medical diagnosis using machine learning research paper

• Open access
• Published: 11 September 2024

Cardiovascular disease diagnosis: a holistic approach using the integration of machine learning and deep learning models

• Hossein Sadr 1 ,
• Arsalan Salari 2 ,
• Mohammad Taghi Ashoobi 3 &
• Mojdeh Nazari 2 , 4  

European Journal of Medical Research volume  29 , Article number:  455 ( 2024 ) Cite this article

Metrics details

The incidence and mortality rates of cardiovascular disease worldwide are a major concern in the healthcare industry. Precise prediction of cardiovascular disease is essential, and the use of machine learning and deep learning can aid in decision-making and enhance predictive abilities.

The goal of this paper is to introduce a model for precise cardiovascular disease prediction by combining machine learning and deep learning.

Two public heart disease classification datasets with 70,000 and 1190 records besides a locally collected dataset with 600 records were used in our experiments. Then, a model which makes use of both machine learning and deep learning was proposed in this paper. The proposed model employed CNN and LSTM, as the representatives of deep learning models, besides KNN and XGB, as the representatives of machine learning models. As each classifier defined the output classes, majority voting was then used as an ensemble learner to predict the final output class.

The proposed model obtained the highest classification performance based on all evaluation metrics on all datasets, demonstrating its suitability and reliability in forecasting the probability of cardiovascular disease.

Introduction

Nowadays massive amounts of data are generated in the healthcare industry and individuals facing these types of data have realized that there is a significant gap between their collection and interpretation [ 1 , 2 ]. In today's data-driven era, the intersection of healthcare and artificial intelligence has paved the way for transformative advancements in healthcare. In this context, machine learning algorithms have emerged as powerful tools capable of analyzing vast amounts of patient data with unprecedented speed and precision. By harnessing the potential of machine learning, healthcare providers can leverage complex patterns within diverse datasets to develop predictive models for disease diagnosis. These models can identify subtle indicators and risk factors that may elude traditional diagnostic methods, empowering healthcare providers to take proactive measures and customize individualized treatment plans for patients. [ 3 , 4 ].

According to the World Health Organization report, cardiovascular diseases are the leading non-communicable disease that causes numerous deaths worldwide, roughly 17.9 million individuals pass away from cardiovascular diseases, making up approximately 31% of total worldwide fatalities [ 5 , 6 ]. Recent information from the American Heart Association indicates that cardiovascular diseases continue to be the main cause of death in the United States in 2023. Given the increasing rate of coronary artery diseases and their transformation into a global concern, the healthcare industry needs to shape and enhance methods of dealing with these diseases to minimize their impact on society [ 7 , 8 ]. In this regard, the integration of machine learning in cardiovascular disease diagnosis not only holds promise for enhancing diagnostic accuracy, but also for optimizing resource allocation within healthcare systems. By streamlining diagnostic processes and prioritizing high-risk individuals for further evaluation, machine learning-driven approaches offer the potential to improve patient care, increase operational efficiency, and ultimately save lives [ 9 ].

The application of machine learning has provided a new approach to predicting cardiovascular diseases [ 10 , 11 , 12 ]. Consequently, various machine learning methods have been used to recognize and extract useful information from clinical datasets with minimal user input and effort. However, the emergence of deep learning has revolutionized cardiovascular disease prediction by offering distinct advantages over traditional machine learning approaches [ 10 , 13 , 14 ]. Deep learning algorithms, such as neural networks, excel in processing vast amounts of complex data, capturing intricate patterns, and extracting high-level features from raw inputs. In the context of cardiovascular disease prediction, the inherent ability of deep learning models to automatically learn hierarchical representations of data enables them to uncover subtle relationships and dependencies that may not be apparent to conventional machine learning algorithms [ 15 , 16 , 17 , 18 , 19 ].

The fusion of deep learning and machine learning methodologies also holds great promise for advancing cardiovascular disease diagnosis and management in recent years [ 20 , 21 , 22 , 23 , 24 ]. By combining the strengths of deep learning in extracting intricate patterns from complex data and machine learning's interpretability and explainability, healthcare professionals can leverage a hybrid approach to enhance the accuracy, efficiency, and transparency of cardiovascular disease diagnosis. Deep learning models can effectively process and analyze large amounts of data to extract comprehensive features and patterns that may be indicative of cardiovascular conditions. These deep learning-derived features can then be integrated with traditional machine learning algorithms to build predictive models that not only provide accurate diagnostic assessments, but also offer insights into the underlying decision-making process, enabling clinicians to understand and trust the predictions made by the hybrid system. Multiple empirical and theoretical studies have shown that the accuracy of combinational models is often better than individual ones [ 11 , 19 , 21 , 25 , 26 ].

While multiple studies have been carried out in this domain, a specific accurate predictive model that can thoroughly identify all essential characteristics of cardiovascular diseases is still lacking. Accordingly, a combinational model is presented in this paper which makes use of both Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) network, as the representative of deep learning models, besides K-Nearest Neighbor (KNN) and Extreme Gradient Boosting (XGB) as the representative of machine learning models to predict cardiovascular disease. This amalgamation of deep learning and machine learning techniques empowers healthcare providers with a powerful toolkit for personalized and precise cardiovascular disease diagnosis, paving the way for more informed treatment strategies and improved patient care outcomes. To implement our proposed model, three distinct datasets were utilized. Two of them are public heart disease datasets sourced from Kaggle and the third one is a local dataset collected from the medical records of patients at Dr.Heshmat Hospital, Guilan, Iran. The key contribution of this paper can be outlined as follows:

Two public datasets besides a locally collected dataset were used in our experiments to identify the primary features and optimal model for predicting cardiovascular disease.

A novel combinational model was proposed for cardiovascular disease prediction that could capture complex relationships in data through both feature learning and traditional algorithmic approaches, leading to improved predictive performance.

Drawing from the results of experiments, a robust and precise framework for predicting cardiovascular disease was proposed, offering substantial value in optimizing resource allocation and aiding cardiologists in effectively categorizing recently diagnosed patients.

The remainder of this paper is classified as follows: Section " Related work " includes a summary of related studies. The employed methodology including dataset description, proposed model, and its details are mentioned in Section " Proposed model ". Section " Experiments and results " includes the results of the experiments and discussion. Conclusion and future research direction are provided in Section " Experiments and results ".

Related work

Using machine learning and deep learning for cardiovascular disease prediction is crucial as it can enhance accuracy in identifying risk factors, enable early detection of potential issues, personalize treatment plans, and ultimately improve patient outcomes through proactive healthcare interventions. Accordingly, numerous have been conducted to investigate and find the appropriate technique for predicting heart disease. While the focus of this paper is to propose a model based on the combination of machine learning and deep learning for cardiovascular prediction, studies conducted in this era are briefly introduced in the following.

Using machine learning for cardiovascular disease prediction involves training algorithms on large amounts of medical data to identify patterns and predict the likelihood of an individual developing heart-related issues. By leveraging machine learning techniques like logistic regression, random forests, or neural networks, healthcare professionals can assess the probability of cardiovascular events, allowing for early intervention and personalized treatment approaches. From the perspective of machine learning, Ahmad et al. [ 27 ] performed experiments using six machine learning algorithms: K-nearest neighbor, logistic regression, SVM, random forest classifier, decision tree, and extreme gradient boosting, which were trained on two datasets related to heart disease. The findings indicated that SVM recorded the highest accuracy, reaching 87.91% on the Cleveland dataset. Meanwhile, Akkaya et al. [ 28 ] examined eight distinct machine learning classification methods using the Cleveland dataset and determined that KNN had the best performance with an accuracy of 85.6%. Similarly, Tougui et al. [ 29 ] applied different data mining techniques, with the random forest method attaining the highest classification accuracy of 87.64%.

Additionally, Shafenoor et al. [ 30 ] explored the effectiveness of data mining methods in recognizing important features and classifying whether heart diseases are present or absent. Their results indicated that using a voting approach with Naïve Bayes and logistic regression achieved the highest classification accuracy of 87.41%.

In a similar research direction, Subanya & Rajalaxmi [ 31 ] utilized the SVM classification technique along with the Swarm intelligence-based Artificial Bee Colony (ABC) algorithm to find optimal features, resulting in an accuracy of 86.76%. Additionally, Mokeddem et al. [ 32 ] utilized the genetic algorithm alongside Naïve Bayes and SVM for classification, yielding accuracies of 85.50% and 83.82%, respectively. Khanna et al. [ 33 ] performed a comparative study of classification methods (logistic regression, SVM, and neural networks) to forecast the prevalence of heart disease, determining that logistic regression achieved the best performance with a classification accuracy of 84.80%. Furthermore, Kumar et al. [ 34 ] employed eight different data mining methods to predict heart disease, finding that Decision Tree C4.5 performed the best with an accuracy of 83.40%. Finally, Acharya [ 35 ] explored the effectiveness of various data mining techniques in predicting the presence of heart disease, concluding that KNN is the most effective algorithm with a classification accuracy of 82%. Arroyo et al. [ 36 ] employed the combination of an optimized neural network and genetic algorithm for cardiovascular disease prediction and proved that their proposed model outperformed the other machine learning algorithms. Lin et al. [ 37 ] employed the two-step Taguchi technique along with an artificial neural network for accurate prediction of cardiovascular disease risk and revealed that not only their proposed model could accurately predict cardiovascular risk but also conserve computational resources. They also proposed a model named TPTM-HANN-GA derived from the combination of the two-phase Taguchi method (TPTM), a hyperparameter artificial neural network (HANN), and a genetic algorithm (GA) that could effectively fine-tune hyperparameters for an artificial neural network in the training phase, leading to a substantial improvement in the accuracy of cardiovascular disease risk prediction [ 38 ]. Chaithra et al. [ 39 ] also used three learning models, namely DT-J48, Naive Bayes, and neural network, to design a predictive model for cardiovascular diseases. They applied these models to a dataset collected from the trans-thoracic echocardiography database. Their empirical results showed that the neural network model performs much better in predicting heart disease with an accuracy of 97.91%. Nazari et al. [ 40 ] also proposed a combinational model based on majority voting and GA for cardiovascular prediction and applied it to the Cleveland and a local dataset which obtained an accuracy of 88.43% on Cleveland dataset.

In the subsequent research area, the emergence of deep learning has significantly impacted the prediction of cardiovascular disease by enabling more complex and accurate models to be developed. Deep learning algorithms, particularly deep neural networks, are capable of automatically learning intricate patterns and relationships within large datasets without the need for explicit feature engineering. Accordingly, Singhal et al. [ 41 ] evaluated a convolutional neural network on the Cleveland dataset. They first used a convolutional neural network with 2 convolutional layers and then the number of layers was increased. Their model was expanded up to 5 layers, and the results were examined and compared. Based on their obtained result, the best accuracy (95%) was obtained with 3 convolutional layers. Dutta et al. [ 42 ] created a dataset similar to the Cleveland dataset using information gathered from the "National Health and Nutrition Examination Survey" database. They trained a neural network model with multiple convolutional layers and obtained an accuracy of 81.78%.

In recent years, the combination of machine learning and deep learning has revolutionized cardiovascular disease prediction by leveraging vast amounts of patient data to identify intricate patterns and subtle relationships that may not be apparent to the human eye. In this regard, Mehmood et al. [ 43 ] proposed a model called CardioHelp that can be used for predicting cardiovascular disease. They used a combination of deep learning algorithms, including multiple convolutional layers, and obtained an accuracy of 97% on a local dataset. Tarawneh and Embarak [ 44 ] examined various machine learning models on the Cleveland dataset to find the best one. They used Naive Bayes, Support Vector Machine, Decision Tree, Neural Network, and K-Nearest Neighbor and then used a combined approach. They concluded that the combinational model that makes use of the benefits of all baselines obtained the highest accuracy can be used as an expert system in hospitals to assist doctors in quickly diagnosing cardiovascular disease. Bhavekar and Goswami [ 45 ] developed a hybrid deep-learning methodology based on RNN and LSTM for cardiac disease categorization. In their classification process, RNN utilized three distinct activation functions while various preprocessing techniques were applied for sorting and classifying. Relational, bigram, and density-based methods were employed for feature extraction. Subhadra and Vikas [ 46 ] tested decision tree models, logistic regression, Naive Bayes, random forest, support vector machine, deep learning, gradient boosting tree algorithm, and generalized linear models on the Cleveland dataset to predict cardiovascular disease. Finally, their proposed MLP-NN model achieved an accuracy of 93.39% with 5 neurons in the hidden layer for predicting cardiovascular disease compared to other models.

Proposed model

Machine learning and deep learning both offer valuable tools for cardiovascular disease prediction, each with its own set of benefits. Notably, machine learning excels in interpretability, allowing for insights into factors influencing predictions. On the other hand, deep learning can automatically learn intricate patterns in raw data, potentially capturing complex relationships that may be missed by traditional machine learning methods. While each of them has its own benefits and pitfalls, using combinational models for cardiovascular disease prediction offers a significant advantage by harnessing the collective power of diverse predictive models to improve the accuracy and robustness via capturing complex relationships within the data. In this regard, the proposed model leverages the combination of CNN and LSTM from deep learning and KNN and XGB from machine learning as the base classifiers for cardiovascular disease classification. As the output classes are defined by each classifier, majority voting is then used as an ensemble learner to predict the final output class.

Noteworthy, the reason behind choosing each of these classifiers refers to their unique structure while CNN can automatically extract relevant features from the input, effectively capturing key aspects of the data that are important for classification. Moreover, LSTM has the inherent ability to extract relevant features and representations from sequential data without the need for explicit feature engineering. The combination of CNN and LSTM networks allows the model to capture complex relationships and patterns within the clinical data, potentially enabling it to discern intricate dependencies and interactions between different clinical variables, ultimately aiding in accurate disease classification. On the other hand, KNN is effective in capturing local patterns within the feature space. In the context of clinical data, where characteristics of patients and their health profiles can exhibit local patterns, KNN can be suitable for identifying similarities between patients based on their clinical attributes, potentially aiding in patient classification. Moreover, clinical data often contain non-linear relationships between patient characteristics and disease outcomes. XGB's ability to capture complex interactions between features and its capability to handle non-linear relationships can be beneficial for accurate cardiovascular disease classification. The use of majority voting also ensures mitigating the impact of individual model biases by aggregating predictions from multiple models, thereby reducing the risk of overfitting and improving generalization to unseen data. The structure of the proposed model is depicted in Fig.  1 and more details are provided in the following.

Structure of the proposed model

The first step in the proposed model is preprocessing, which is essential for preparing the data for the critical phase of the learning model. The preprocessing step includes data cleaning, data transformation, data augmentation, data balancing, and data normalization.

Data cleaning: Data cleaning encompasses the tasks of recognizing and rectifying errors or discrepancies within the dataset, such as addressing missing values, eliminating duplicates, rectifying inaccuracies, and managing outliers.

Data transformation: Data transformation is the process of converting data into a format that is better suited for analysis, helping improve the performance of machine learning models by making the data more understandable and easier to work with.

Data balancing: Data balancing is a preprocessing step used when the dataset is imbalanced to ensure that the model does not favor the majority class and makes more accurate predictions for all classes.

Data normalization: Data normalization is the process of scaling the features of the dataset to a standard range to make sure that no particular feature dominates the learning process due to its larger scale compared to other features.

Once the data have been preprocessed, it is divided into two groups: a training set and a testing set. The training set is utilized to train the model. Our proposed model includes three base classifiers as CNN-LTSM, XGB and KNN. The first classifier is made by combining CNN and LSTM. Notably, combining these two models can help to leverage their strengths and advantages of each model and mitigate the weaknesses to some extent. One of the main advantages of a CNN is its low number of parameters and ease of training. CNN is also capable of extracting local features and, with an increase in the number of layers, extracting more valuable features from the input sequence [ 47 ]. LSTM is good at capturing temporal dependencies in time series data, which can be useful in predicting cardiovascular disease progression over time. The structure of the long short-term memory network is designed in a way that can effectively address the main limitations of CNN. In summary, CNNs can effectively process structured clinical data, extracting relevant patterns and features, while LSTMs are adept at capturing temporal dependencies within sequential clinical records. The architecture of the combined CNN-LSTM model which is used as the first baseline classifier is shown in Fig.  2 .

Proposed CNN-LSTM architecture

XGB and KNN are the next base classifiers. While clinical data often contain non-linear relationships between patient characteristics and disease outcomes, XGB's ability to capture complex interactions between features and its capability to handle non-linear relationships can be beneficial for accurate cardiovascular disease classification. KNN is also a simple yet effective algorithm for classification tasks while it is a non-parametric model and does not make strong assumptions about the underlying distribution of the data. This can be advantageous when dealing with clinical data, as it allows the model to adapt to complex and diverse patient characteristics without imposing strict constraints on the data's distribution. For in-depth information about the algorithms discussed, please refer to Taye et al. [ 48 ], as their comprehensive explanations exceed the scope of this paper.

In the following, while the outputs of these classifiers are generated, majority voting is utilized for combining the predictions of multiple classifiers, where the final prediction is based on the most common prediction among the individual classifiers. Majority voting can help to reduce the impact of individual classifiers that may be biased or have poor performance on certain types of data. In other words, majority voting is effective for combining classifiers because it leverages the wisdom of the crowd, aggregating diverse opinions from multiple classifiers to reduce individual biases and errors, leading to more robust and accurate predictions. By voting on the most frequently predicted class, the combinational model tends to make better decisions, enhancing overall performance and generalization across various datasets and learning tasks. It is worth mentioning that the weights assigned to each classifier's prediction are determined based on their individual performance on a validation set to maximize the overall predictive performance.

Experiments and results

As mentioned earlier, our experiments involved utilizing three distinct datasets. Two of them are freely available heart disease datasets sourced from Kaggle and the next one is a local dataset collected from medical records of patients who visited Dr. Heshmat Hospital, Guilan, Iran. The comparison of these three datasets is shown in Table  1 and their details are provided in the following.

Dataset I ( https://www.kaggle.com/datasets/sulianova/cardiovascular-disease-dataset ) contains 70,000 instances gathered from medical examinations and consists of 12 variables. The first 11 variables serve as input features, while the 12th variable is the output feature indicating the presence or absence of cardiovascular disease. It must be mentioned that this dataset contains many duplicate values and extreme outliers. Accordingly, in the preprocessing step, duplicated values and instances with extreme outliers were removed the number of records was reduced to 62,267. Moreover, the age attributed was converted from days to years. Based on the American Heart Association's normal range, systolic and diastolic blood pressure were changed from numerical to nominal for better analysis.  Feature descriptions and their distribution after preprocessing are provided in Table  2 .

Dataset II ( https://www.kaggle.com/datasets/mexwell/heart-disease-dataset/data ) includes 1190 instances obtained from the combination of five original datasets across 11 shared attributes and one attribute as the predictor presenting the presence or absence of cardiovascular disease in a patient . The compilation of this dataset includes the incorporation of five original datasets including Cleveland (303 samples), Hungarian (294 samples), Switzerland (123 samples), Long Beach VA (200 samples), and Statlog datasets (270 samples). This dataset also includes missing and duplicate values that were removed along the preprocessing step and the number of records was reduced to 918. Features descriptions and their distribution after preprocessing are provided in Table  3 .

To enhance the foundation of our analysis, we opted to gather a local dataset that mirrors the attributes of Dataset II . Accordingly, a local dataset was collected in this paper which includes the information on the medical records of patients who visited Dr. Heshmat Hospital from January to June 2023. The collected dataset is called “ Dataset III ” containing 600 and 12 features while 11 of them are used as predictors and one is the nominator of output indicating the absence or presence of cardiovascular disease. This dataset also includes missing values that were removed along the preprocessing step and the number of records was reduced to 577. Features descriptions and their distribution after preprocessing are provided in Table  4 .

Evaluation metrics

To assess the performance of implemented models, it is necessary to consider an appropriate metric to examine the efficiency of the models. In this study, metrics including accuracy, precision, recall, F1 score, and specificity have been used for evaluating the proposed model that are indicated in equations \(1 to 5\) . In these equations, TP is the count of positive samples accurately identified by the system, while FP is the count of positive samples incorrectly identified. FN refers to negative samples that were wrongly classified as positive, and TN is the number of negative samples correctly recognized.

Implementation details and hyperparameters

To take advantage of the processing power of the graphics processor, all instructions were implemented using Google Colab based on Python 3.10 as the programming language. The hardware infrastructure for running the proposed model was a system with an Intel Core i5 processor, 8 GB RAM, and an Ubuntu distribution as the operating system.

Notably, oversampling techniques along with cross-validation were employed in our implementations to maintain data integrity and prevent data leakage. To this end, the original datasets were split into a training set and a separate holdout test set (80% for training and 20% for testing), ensuring that the test set is not used during cross-validation. Then, the tenfold cross-validation technique was utilized to split the training data into multiple folds. For each fold in the cross-validation process, the oversampling technique was only applied to the training data within that fold to ensure that oversampling is performed independently for each fold, preventing data leakage across folds. Thereafter, the model was trained on the training data within each fold and its performance was evaluated on the validation data within that fold. Finally, performance metrics for each fold were measured and then aggregated to obtain an overall assessment of the model's performance. After completing the cross-validation process, the final model was evaluated on the holdout test set that was initially set aside to provide an additional independent evaluation of our model's performance.

Hyperparameters are external settings that influence an algorithm's behavior and can greatly affect the model's performance and generalization ability. Since they directly impact how well a model performs and predicts, it is important to tune these hyperparameters carefully. In our study, we made rigorous efforts to appropriately configure hyperparameters for each algorithm to optimize performance and ensure effective pattern capture. Details of the hyperparameters employed are outlined in Table  5 .

Performance evaluation

The goal of this paper is to introduce a model based on the combination of machine learning and deep learning models to predict the risk of cardiovascular diseases. Accordingly, basic machine learning and deep learning models besides combinational models were implemented on three mentioned datasets. The average performance metrics across the 10 folds of cross-validation with the standard deviation on each dataset are provided in the following tables (Tables 6 , 7 , and 8 ).

To better assess the classification performance of our proposed model compared to other models, namely CNN-LSTM, KNN, and XGB, their ROC curves on the three mentioned datasets are, respectively, depicted in Fig.  3 .

Roc Curve illustration of our proposed model compared to other models on three datasets

Based on the result of experiments on all three datasets, it can be concluded that:

Among traditional machine learning-based models, KNN has the highest classification accuracy on all datasets. XGB also presents the highest accuracy among ensemble-based models on all datasets.

Comparing baselines with combinational models, it is clear that combinational models outperform any single model alone because they leverage the strengths of multiple individual models by combining their predictions.

Considering the results of deep learning-based models, it can be concluded that they have higher classification accuracy than both traditional and combinational models due to their ability to automatically learn intricate patterns and features from raw data.

The main conclusion can be drawn from the last lines of Tables 6 , 7 , and 8 , which showcase the advantages of our proposed model. It underscores the effectiveness of our combinational learning model in comparison to others. Overall, our model outperforms all evaluation metrics across all datasets, making it a suitable benchmark for future research.

While our proposed model has the highest performance based on all evaluation metrics on all three datasets, it can be stated that it not only can generalize well to new, unseen data, which is an important characteristic for any predictive model, but also it has consistent behavior in its predictive capability which is a good indication that the model is not overfitting to a particular dataset and is generalizing reasonably well.

Considering Fig.  3 , it can be seen that our proposed model has the closest ROC curve to the top-left corner on all three datasets which signals that the model has strong overall performance across different thresholds, making it a dependable predictor. Therefore, it can be claimed that our proposed model has strong discriminative power, meaning it is effective at distinguishing between classes.

Precisely forecasting the risk of cardiovascular disease is essential for early intervention and better patient results. This paper proposed a holistic approach using the integration of machine learning and deep learning models to improve the accuracy of cardiovascular disease prediction. According to the empirical results, our combinational model presented the highest classification performance based on all evaluation metrics indicating that this combination offers a more comprehensive approach to analyzing complex cardiovascular disease data compared to using just one type of model. In order to assess the efficiency of our proposed model, it is crucial to conduct a comparison with current state-of-the-art approaches. While the majority of prior research utilized the Cleveland dataset, a component of Dataset II, for their evaluations, we opted to apply our proposed model using all specified configurations on the Cleveland dataset to ensure a fair and comprehensive comparison. An accurate comparison of the existing studies and our proposed model on the Cleveland dataset is provided in Table  9 . To provide more comparison (Table  10 ), we also compared our proposed model with studies that concluded their experiments on Dataset I. As can be seen, our proposed model obtained the highest classification accuracy compared to the state of the arts on both datasets.

Even if our proposed mode performs well in research settings, transitioning them into clinical practice requires rigorous validation and regulatory approval while it may not generalize well to different settings or populations. Differences in patient demographics, medical procedures, and treatment guidelines can influence how well predictive models work when used in varying environments. Given that disease trends and risk elements can change due to factors like lifestyle shifts, medical progress, and population demographics, the model created using historical data may face challenges in adjusting to these evolving trends, potentially leading to reduced prediction accuracy.

Cardiovascular disease is a prominent global cause of mortality, underscoring the critical need for early detection in healthcare settings. Artificial intelligence plays a crucial role in this area by pinpointing risk factors, facilitating predictive analytics, aiding decision-making, and fostering knowledge exploration. This contributes to proactive and personalized strategies for managing cardiovascular disease. Accordingly, a model which makes use of both machine learning and deep learning is proposed in this paper. The proposed model employed CNN and LSTM, as the representatives of deep learning models, besides KNN and XGB, as the representatives of machine learning models. As the output classes are defined by each classifier, majority voting is then used as an ensemble learner to predict the final output class.

To demonstrate the effectiveness of the proposed model, we utilized two public datasets along with a locally collected dataset in our experiments. To ensure a valid comparison, all datasets were first processed using the same methods. The experimental results across all datasets showed that the proposed model outperformed both individual classifiers and combinations of classifiers. These findings highlight a precise model that can be utilized for predicting the risk of cardiovascular disease. Additionally, it offers an important utility for cardiologists and physicians in categorizing new patients and assessing the necessary human resources, including doctors, technicians, nurses, and vital medical equipment.

There are abundant opportunities to enhance this research and address the constraints of the current study. One strategy involves broadening the study by replicating the experiment using larger real-world datasets. Future research could investigate alternative combinations of machine learning and deep learning models for predicting cardiovascular disease. Moreover, implementing novel feature selection methods could offer a more comprehensive insight into crucial features, leading to improved prediction accuracy. Exploring the application of the proposed model in other domains holds promise and could be considered as a potential avenue for future research.

Availability of data and materials

Three datasets were used in our experiments. Two of them are freely available datasets that can be found on Kaggle. The next one is the locally collected dataset that is freely available for academic purposes upon request.

Bhatt CM, et al. Effective heart disease prediction using machine learning techniques. Algorithms. 2023;16(2):88.

Article   Google Scholar  

Ramesh T, et al. Predictive analysis of heart diseases with machine learning approaches. Malays J Comp Sci. 2022. https://doi.org/10.22452/mjcs.sp2022no1.10 .

Nagavelli U, Samanta D, Chakraborty P. Machine learning technology-based heart disease detection models. J Healthc Eng. 2022;2022:7351061.

Article   PubMed   PubMed Central   Google Scholar  

Al-Jammali K. Prediction of heart diseases using data mining algorithms. Informatica. 2023;47(5):e4467.

Tsao CW, et al. Heart disease and stroke statistics—2022 update: a report from the American heart association. Circulation. 2022;145(8):e153–639.

Article   PubMed   Google Scholar  

Saberi ZA, Sadr H, Yamaghani MR. An intelligent diagnosis system for predicting coronary heart disease. In: Saberi ZA, editor. 2024 10th International conference on artificial intelligence and robotics (QICAR). Qazvin: IEEE; 2024.

Google Scholar  

Solanki A, Barot MP. Study of heart disease diagnosis by comparing various classification algorithms. Int J Eng Adv Technol. 2019;8(2S2):40–2.

Bhavekar GS, et al. Heart disease prediction using machine learning, deep Learning and optimization techniques—a semantic review. Multimed Tools Appl. 2024. https://doi.org/10.1007/s11042-024-19680-0 .

Vijayarani S, Dhayanand S, Phil M. Kidney disease prediction using SVM and ANN algorithms. Int J Comp Bus Res (IJCBR). 2015;6(2):1–12.

Bakar WA, et al. A review: heart disease prediction in machine learning and deep learning. In: Bakar WA, editor., et al., 2023 19th IEEE International colloquium on signal processing and its applications (CSPA). Kedah: IEEE; 2023.

Mohades Deilami F, Sadr H, Tarkhan M. Contextualized multidimensional personality recognition using combination of deep neural network and ensemble learning. Neural Proc Lett. 2022;54(5):3811–28.

Ogunpola A, et al. Machine learning-based predictive models for detection of cardiovascular diseases. Diagnostics. 2024;14(2):144.

Latha CBC, Jeeva SC. Improving the accuracy of prediction of heart disease risk based on ensemble classification techniques. Informatics in Medicine Unlocked. 2019;16: 100203.

Garavand A, et al. The attributes of hospital-based coronary artery diseases registries with a focus on key registry processes: a systematic review. Health Information Management Journal. 2022;51(2):63–78.

Alotaibi N, Alzahrani M. Comparative analysis of machine learning algorithms and data mining techniques for predicting the existence of heart disease. Int J Adv Comp Sci Applx. 2022;13(7):810.

Ayatollahi H, Gholamhosseini L, Salehi M. Predicting coronary artery disease: a comparison between two data mining algorithms. BMC Public Health. 2019;19(1):1–9.

Alizadehsani R, et al. Machine learning-based coronary artery disease diagnosis: a comprehensive review. Comput Biol Med. 2019;111: 103346.

Garavand A, et al. Designing the coronary artery disease registry with data management processes approach: a comparative systematic review in selected registries. Int Cardiovasc Res J. 2020;14(1): 100833.

Khodaverdian Z, et al. An energy aware resource allocation based on combination of CNN and GRU for virtual machine selection. Multimed Tools Appl. 2023. https://doi.org/10.1007/s11042-023-16488-2 .

Alkayyali Z, Idris S, Abu-Naser SS. A systematic literature review of deep and machine learning algorithms in cardiovascular diseases diagnosis. J Theor Appl Inf Technol. 2023;101(4):1353–65.

Naser MA, et al. A review of machine learning’s role in cardiovascular disease prediction: recent advances and future challenges. Algorithms. 2024;17(2):78.

Samavat T, Hojatzadeh E. Programs for prevention and control of cardiovascular diseases. Javan: Ministry of Health; 2012.

Goswami AD, Bhavekar GS, Chafle PV. Electrocardiogram signal classification using VGGNet: a neural network based classification model. Int J Inf Technol. 2023;15(1):119–28.

Bhavekar GS, Goswami AD. Herding exploring algorithm With light gradient boosting machine classifier for effective prediction of heart diseases. Int J Swarm Intell Res (IJSIR). 2022;13(1):1–22.

Subramani S, et al. Cardiovascular diseases prediction by machine learning incorporation with deep learning. Front Med. 2023;10:1150933.

Sadr H et al. Unified topic-based semantic models: a study in computing the semantic relatedness of geographic terms. In: 2019 5th International conference on web research (ICWR). IEEE; 2019.

Ahamad GN, et al. Influence of optimal hyperparameters on the performance of machine learning algorithms for predicting heart disease. Processes. 2023;11(3):734.

Akkaya B, Sener E, Gursu C. A comparative study of heart disease prediction using machine learning techniques. In: 2022 International congress on human–computer interaction, optimization and robotic applications (HORA). IEEE; 2022.

Tougui I, Jilbab A, El Mhamdi J. Heart disease classification using data mining tools and machine learning techniques. Heal Technol. 2020;10:1137–44.

Amin MS, Chiam YK, Varathan KD. Identification of significant features and data mining techniques in predicting heart disease. Telematics Inform. 2019;36:82–93.

Subanya B, Rajalaxmi R.Feature selection using artificial bee colony for cardiovascular disease classification. In: 2014 International conference on electronics and communication systems (ICECS). IEEE; 2014.

Mokeddem S, Atmani B, Mokaddem M. Supervised feature selection for diagnosis of coronary artery disease based on genetic algorithm. arXiv preprint. 2013. https://arxiv.org/abs/1305.6046 .

Khanna D, et al. Comparative study of classification techniques (SVM, logistic regression and neural networks) to predict the prevalence of heart disease. Int J Mach Learn Comput. 2015;5(5):414.

Kumar MN, Koushik K, Deepak K. Prediction of heart diseases using data mining and machine learning algorithms and tools. Int J Sci Res Comput Sci Eng Inform Technol. 2018;3(3):887–98.

Acharya A. Comparative study of machine learning algorithms for heart disease prediction. 2017.

Arroyo JCT, Delima AJP. An optimized neural network using genetic algorithm for cardiovascular disease prediction. J Adv Inform Technol. 2022. https://doi.org/10.12720/jait.13.1.95-99 .

Lin C-M, Lin Y-S. Utilizing a two-stage taguchi method and artificial neural network for the precise forecasting of cardiovascular disease risk. Bioengineering. 2023;10(11):1286.

Article   CAS   PubMed   PubMed Central   Google Scholar  

Lin C-M, Lin Y-S. TPTM-HANN-GA: a novel hyperparameter optimization framework integrating the Taguchi method, an artificial neural network, and a genetic algorithm for the precise prediction of cardiovascular disease risk. Mathematics. 2024;12(9):1303.

Chaithra N, Madhu B. Classification models on cardiovascular disease prediction using data mining techniques. Cardiovasc Dis Diagn. 2018;6(6):1–4.

CAS   Google Scholar  

Nazari M, et al. Detection of cardiovascular diseases using data mining approaches: application of an ensemble-based model. Cogn Comput. 2024. https://doi.org/10.1007/s12559-024-10306-z .

Singhal S, Kumar H, Passricha V. Prediction of heart disease using CNN. Am Int J Res Sci Technol Eng Math. 2018;23(1):257–61.

Dutta A, et al. An efficient convolutional neural network for coronary heart disease prediction. Expert Syst Appl. 2020;159: 113408.

Mehmood A, et al. Prediction of heart disease using deep convolutional neural networks. Arab J Sci Eng. 2021;46(4):3409–22.

Tarawneh M, Embarak O. Hybrid approach for heart disease prediction using data mining techniques. In: Barolli L, Xhafa F, Khan ZA, Odhabi H, editors. International conference on emerging internetworking, data and web technologies. Cham: Springer; 2019.

Bhavekar GS, Goswami AD. A hybrid model for heart disease prediction using recurrent neural network and long short term memory. Int J Inf Technol. 2022;14(4):1781–9.

Subhadra K, Vikas B. Neural network based intelligent system for predicting heart disease. Int J Innov Technol Explor Eng. 2019;8(5):484–7.

Van VD, Thai T, Nghiem MQ. Combining convolution and recursive neural networks for sentiment analysis. In: Proceedings of the 8th international symposium on information and communication technology. 2017.

Taye MM. Understanding of machine learning with deep learning: architectures, workflow, applications and future directions. Computers. 2023;12(5):91.

Download references

This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.

Author information

Authors and affiliations.

Department of Health Informatics and Intelligent Systems, Guilan Road Trauma Research Center, Trauma Institute, Guilan University of Medical Sciences, Rasht, Iran

Hossein Sadr

Cardiovascular Disease Research Center, Department of Cardiology, School of Medicine, Heshmat Hospital, Guilan University of Medical Sciences, Rasht, Iran

Arsalan Salari & Mojdeh Nazari

Department of Surgery, School of Medicine, Razi Hospital, Guilan University of Medical Sciences, Rasht, Iran

Mohammad Taghi Ashoobi

Department of Health Information Technology and Management, School of Allied Medical Sciences, Shahid Beheshti University of Medical Sciences, Tehran, Iran

Mojdeh Nazari

You can also search for this author in PubMed   Google Scholar

Contributions

H.S., A.S., and M.N. conceived of the presented idea. M.N. developed the theory and performed the computations. H.S. and M.A conceived the study and were in charge of overall direction and planning. A.S. and M.A. verified the analytical methods and obtained results. All authors discussed the results and contributed to the final manuscript.

Corresponding authors

Correspondence to Hossein Sadr or Mojdeh Nazari .

Ethics declarations

Ethics approval and consent to participate.

The project was found to be in accordance with the ethical principles and the national norms and standards for conducting Medical Research in Iran (Approval ID: IR.GUMS.REC.1402.516) .

Competing interests

The authors declare no competing interests.

Additional information

Publisher's note.

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/ .

Reprints and permissions

About this article

Cite this article.

Sadr, H., Salari, A., Ashoobi, M.T. et al. Cardiovascular disease diagnosis: a holistic approach using the integration of machine learning and deep learning models. Eur J Med Res 29 , 455 (2024). https://doi.org/10.1186/s40001-024-02044-7

Download citation

Received : 22 March 2024

Accepted : 27 August 2024

Published : 11 September 2024

DOI : https://doi.org/10.1186/s40001-024-02044-7

Share this article

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

• Cardiovascular disease
• Machine learning
• Deep learning
• Combinational model

European Journal of Medical Research

ISSN: 2047-783X

• General enquiries: [email protected]

• Research article
• Open access
• Published: 19 March 2019

Machine learning in medicine: a practical introduction

• Jenni A. M. Sidey-Gibbons 1 &
• Chris J. Sidey-Gibbons   ORCID: orcid.org/0000-0002-4732-7305 2 , 3 , 4  

BMC Medical Research Methodology volume  19 , Article number:  64 ( 2019 ) Cite this article

99k Accesses

659 Citations

48 Altmetric

Metrics details

Following visible successes on a wide range of predictive tasks, machine learning techniques are attracting substantial interest from medical researchers and clinicians. We address the need for capacity development in this area by providing a conceptual introduction to machine learning alongside a practical guide to developing and evaluating predictive algorithms using freely-available open source software and public domain data.

We demonstrate the use of machine learning techniques by developing three predictive models for cancer diagnosis using descriptions of nuclei sampled from breast masses. These algorithms include regularized General Linear Model regression (GLMs), Support Vector Machines (SVMs) with a radial basis function kernel, and single-layer Artificial Neural Networks. The publicly-available dataset describing the breast mass samples ( N =683) was randomly split into evaluation ( n =456) and validation ( n =227) samples.

We trained algorithms on data from the evaluation sample before they were used to predict the diagnostic outcome in the validation dataset. We compared the predictions made on the validation datasets with the real-world diagnostic decisions to calculate the accuracy, sensitivity, and specificity of the three models. We explored the use of averaging and voting ensembles to improve predictive performance. We provide a step-by-step guide to developing algorithms using the open-source R statistical programming environment.

The trained algorithms were able to classify cell nuclei with high accuracy (.94 -.96), sensitivity (.97 -.99), and specificity (.85 -.94). Maximum accuracy (.96) and area under the curve (.97) was achieved using the SVM algorithm. Prediction performance increased marginally (accuracy =.97, sensitivity =.99, specificity =.95) when algorithms were arranged into a voting ensemble.

Conclusions

We use a straightforward example to demonstrate the theory and practice of machine learning for clinicians and medical researchers. The principals which we demonstrate here can be readily applied to other complex tasks including natural language processing and image recognition.

Peer Review reports

Driven by an increase in computational power, storage, memory, and the generation of staggering volumes of data, computers are being used to perform a wide-range of complex tasks with impressive accuracy. Machine learning (ML) is the name given to both the academic discipline and collection of techniques which allow computers to undertake complex tasks. As an academic discipline, ML comprises elements of mathematics, statistics, and computer science. Machine learning is the engine which is helping to drive advances in the development of artificial intelligence. It is impressively employed in both academia and industry to drive the development of ‘intelligent products’ with the ability to make accurate predictions using diverse sources of data [ 1 ]. To date, the key beneficiaries of the 21 st century explosion in the availability of big data, ML, and data science have been industries which were able to collect these data and hire the necessary staff to transform their products. The learning methods developed in and for these industries offer tremendous potential to enhance medical research and clinical care, especially as providers increasingly employ electronic health records.

Two areas which may benefit from the application of ML techniques in the medical field are diagnosis and outcome prediction. This includes a possibility for the identification of high risk for medical emergencies such as relapse or transition into another disease state. ML algorithms have recently been successfully employed to classify skin cancer using images with comparable accuracy to a trained dermatologist [ 2 ] and to predict the progression from pre-diabetes to type 2 diabetes using routinely-collected electronic health record data [ 3 ].

Machine learning will is increasingly employed in combination with Natural Language Processing (NLP) to make sense of unstructured text data. By combining ML with NLP techniques, researchers have been able to derive new insights from comments from clinical incident reports [ 4 ], social media activity [ 5 , 6 ], doctor performance feedback [ 7 ], and patient reports after successful cancer treatments [ 8 ]. Automatically generated information from unstructured data could be exceptionally useful not only in order to gain insight into quality, safety, and performance, but also for early diagnosis. Recently, an automated analysis of free-speech collected during in-person interviews resulted in the ability to predict transition to psychosis with perfect accuracy in a group of high-risk youths [ 9 ].

Machine learning will also play a fundamental role in the development of learning healthcare systems. Learning healthcare systems describe environments which align science, informatics, incentives, and culture for continuous improvement and innovation. In a practical sense, these systems; which could occur on any scale from small group practices to large national providers, will combine diverse data sources with complex ML algorithms. The result will be a continuous source of data-driven insights to optimise biomedical research, public health, and health care quality improvement [ 10 ].

Machine learning

Machine learning techniques are based on algorithms – sets of mathematical procedures which describe the relationships between variables. This paper will explain the process of developing (known as training ) and validating an algorithm to predict the malignancy of a sample of breast tissue based on its characteristics. Though algorithms work in different ways depending on their type there are notable commonalities in the way in which they are developed. Though the complexities of ML algorithms may appear esoteric, they often bear more than a subtle resemblance to conventional statistical analyses.

Given the commonalities shared between statistical and ML techniques, the boundary between the two may seem fuzzy or ill-defined. One way to delineate these bodies of approaches is to consider their primary goals. The goal of statistical methods is inference ; to reach conclusions about populations or derive scientific insights from data which are collected from a representative sample of that population. Though many statistical techniques, such as linear and logistic regression, are capable of creating predictions about new data, the motivator of their use as a statistical methodology is to make inferences about relationships between variables. For example, if we were to create a model which described the relationship between clinical variables and mortality following organ transplant surgery for example, we would need to have insight into the factors which distinguish low mortality risk from high if we were to develop interventions to improve outcomes and reduce mortality in the future. In statistical inference, therefore, the goal is to understand the relationships between variables.

Conversely, in the field of ML, the primary concern is an accurate prediction ; the ‘what’ rather than the ‘how’. For example, in image recognition, the relationship between the individual features (pixels) and the outcome is of little relevance if the prediction is accurate. This is a critical facet of ML techniques as the relationship between many inputs, such as pixels in image or video and geo-location, are complex and usually non-linear. It is exceptionally difficult to describe in a coherent way the relationships between predictors and outcomes both when the relationships are non-linear and when there are a large number of predictors, each of which make a small individual contribution to the model.

Fortunately for the medical field, many relationships of interest are reasonably straightforward, such as those between body mass index and diabetes risk or tobacco use a lung cancer. Because of this, their interaction can often be reasonably well explained using relatively simple models. In many popular applications of ML, such a optimizing navigation, translating documents, and identifying objects in videos, understanding the relationship between features and outcomes is of less importance. This allows the use of complex non-linear algorithms. Given this key difference, it might be useful for researchers to consider that algorithms exist on a continuum between those algorithms which are easily interpretable (i.e., Auditable Algorithms) and those which are not (i.e., Black Boxes), presented visually in Fig.  1 .

The complexity/interpretability trade-off in machine learning tools

Interesting questions remain as to when a conventionally statistical technique becomes a ML technique. In this work, we will introduce some that computational enhancements to traditional statistical techniques, such as elastic net regression, make these algorithms performed well with big data. However, a fuller discussion of the similarities and differences between ML and conventional statistics is beyond the purview of the current paper. Interested readers are directed to materials which develop the ideas discussed here [ 11 ]. It should also be acknowledged that whilst the ’Black Box’ concept does generally apply to models which utilize non-linear transformations, such as the neural networks, work is being carried out to facilitate feature identification in complex algorithms [ 12 ].

The majority of ML methods can be categorised into two types learning techniques: those which are supervised and those which are unsupervised. Both are introduced in the following sections.

Supervised learning

Supervised ML refers to techniques in which a model is trained on a range of inputs (or features) which are associated with a known outcome. In medicine, this might represent training a model to relate a person’s characteristics (e.g., height, weight, smoking status) to a certain outcome (onset of diabetes within five years, for example). Once the algorithm is successfully trained, it will be capable of making outcome predictions when applied to new data. Predictions which are made by models trained using supervised learning can be either discrete (e.g., positive or negative, benign or malignant) or continuous (e.g., a score from 0 to 100).

A model which produces discrete categories (sometimes referred to as classes) is referred to as a classification algorithm. Examples of classification algorithms include those which, predict if a tumour is benign or malignant, or to establish whether comments written by a patient convey a positive or negative sentiment [ 2 , 6 , 13 ]. In practice, classification algorithms return the probability of a class (between 0 for impossible and 1 for definite). Typically, we would transform any probability greater than.50 into a class of 1, but this threshold may be altered to improve algorithm performance as required. This paper provides an example of a classification algorithm in which a diagnosis is predicted.

A model which returns a prediction of a continuous value is known as a regression algorithm. The use of the term regression in ML varies from its use in statistics, where regression is often used to refer to both binary outcomes (i.e., logistic regression) and continuous outcomes (i.e., linear regression). In ML, an algorithm which is referred to as a regression algorithm might be used to predict an individual’s life expectancy or tolerable dose of chemotherapy.

Supervised ML algorithms are typically developed using a dataset which contains a number of variables and a relevant outcome. For some tasks, such as image recognition or language processing, the variables (which would be pixels or words) must be processed by a feature selector. A feature selector picks identifiable characteristics from the dataset which then can be represented in a numerical matrix and understood by the algorithm. In the examples above, a feature may be the colour of a pixel in an image or the number of times that a word appears in a given text. Using the same examples, outcomes may be whether an image shows a malignant or benign tumour or whether transcribed interview responses indicate predisposition to a mental health condition.

Once a dataset has been organised into features and outcomes, a ML algorithm may be applied to it. The algorithm is iteratively improved to reduce the error of prediction using an optimization technique.

Note that, when training ML algorithms, it is possible to over-fit the algorithm to the nuances of a specific dataset, resulting in a prediction model that does not generalise well to new data. The risk of over-fitting can be mitigated using various techniques. Perhaps the most straight-forward approach, which will be employed in this work, is to split our dataset into two segments; a training segment and a testing segment to ensure that the trained model can generalize to predictions beyond the training sample. Each segment contains a randomly-selected proportion of the features and their related outcomes. This allows the algorithm to associate certain features, or characteristics, with a specific outcome, and is known as training the algorithm. Once training is completed, the algorithm is applied to the features in the testing dataset without their associated outcomes. The predictions made by the algorithm are then compared to the known outcomes of the testing dataset to establish model performance. This is a necessary step to increase the likelihood that the algorithm will generalise well to new data. This process is illustrated graphically in Fig.  2 .

Overview of supervised learning. a Training b Validation c Application of algorithm to new data

Unsupervised Machine Learning

In contrast with supervised learning, unsupervised learning does not involve a predefined outcome. In unsupervised learning, patterns are sought by algorithms without any input from the user. Unsupervised techniques are thus exploratory and used to find undefined patterns or clusters which occur within datasets. These techniques are often referred to as dimension reduction techniques and include processes such as principal component analysis, latent Dirichlet analysis and t-Distributed Stochastic Neighbour Embedding (t-SNE) [ 14 – 16 ]. Unsupervised learning techniques are not discussed at length in this work, which focusses primarily on supervised ML. However, unsupervised methods are sometimes employed in conjunction with the methods used in this paper to reduce the number of features in an analysis, and are thereby worth mention. By compressing the information in a dataset into fewer features, or dimensions, issues including multiple-collinearity or high computational cost may be avoided. A visual illustration of an unsupervised dimension reduction technique is given in Fig.  3 . In this figure, the raw data (represented by various shapes in the left panel) are presented to the algorithm which then groups the data into clusters of similar data points (represented in the right panel). Note that data which do not have sufficient commonality to the clustered data are typically excluded, thereby reducing the number of features within of the dataset.

A visual illustration of an unsupervised dimension reduction technique

In a similar way to the supervised learning algorithms described earlier, also share many similarities to statistical techniques which will be familiar to medical researchers. Unsupervised learning techniques make use of similar algorithms used for clustering and dimension reduction in traditional statistics. Those familiar with Principal Component Analysis and factor analysis will already be familiar with many of the techniques used in unsupervised learning.

What this paper will achieve

This paper provides a pragmatic example using supervised ML techniques to derive classifications from a dataset containing multiple inputs. The first algorithm we introduce, the regularized logistic regression, is very closely related to multivariate logistic regression. It is distinguished primarily by the use of a regularization function which both reduces the number of features in the model and attenuates the magnitude of their coefficients. Regularization is, therefore, suitable for datasets which contain many variables and missing data (known as high sparsity datasets ), such as the term-document matrices which are used to represent text in text mining studies.

The second algorithm, a Support Vector Machine (SVM), gained popularity among the ML community for its high performance deriving accurate predictions in situations where the relationship between features and the outcome is non-linear. It uses a mathematical transformation known as the kernel trick , which we describe in more detail below.

Finally, we introduce an Artificial Neural Network (ANN), in which complex architecture and heavily modifiable parameters have led to it’s widespread use in many challenging applications, including image and video recognition. The addition of speciality neural networks, such as recurrent or convolutional networks, to ANNs has resulted in impressive performance on a range of tasks. Being highly parametrized models, ANNs are prone to over-fitting. Their performance may be improved using a regularization technique, such as DropConnect.

The ultimate goal of this manuscript is to imbue clinicians and medical researchers with both a foundational understanding of what ML is, how it may be used, as well as the practical skills to develop, evaluate, and compare their own algorithms to solve prediction problems in medicine.

How to follow this paper

We provide a conceptual introduction alongside practical instructions using code written for the R Statistical Programming Environment, which may be easily modified and applied to other classification or regression tasks. This code will act as a framework upon which researchers can develop their own ML studies. The models presented here may be fitted to diverse types of data and are, with minor modifications, suitable for analysing text and images.

This paper is divided into sections which describe the typical stages of a ML analysis: preparing data, training algorithms, validating algorithms, assessing algorithm performance, and applying new data to the trained models.

Throughout the paper, examples of R code used to the run the analyses are presented. The code is given in full in Additional file  1 . The data which was used for these analyses are available in Addition file 2 .

The dataset used in this work is the Breast Cancer Wisconsin Diagnostic Data Set. This dataset is publicly available from the University of California Irvine (UCI) Machine Learning Repository [ 17 ]. It consists of characteristics, or features, of cell nuclei taken from breast masses which were sampled using fine-needle aspiration (FNA), a common diagnostic procedure in oncology. The clinical samples used to form this dataset were collected from January 1989 to November 1991. Relevant features from digitised images of the FNA samples were extracted through the methods described in Refs. [ 13 , 18 , 19 ]. An example of one of the digitised images from an FNA sample is given in Fig.  4 .

An example of an image of a breast mass from which dataset features were extracted

A total of 699 samples were used to create this dataset. This number will be referred to as the number of instances . Each instance has an I.D. number, diagnosis, and set of features attributed to it. While the Sample I.D. is unique to that instance, the diagnosis, listed as class in the dataset, can either be malignant or benign, depending if the FNA was found to be cancerous or not. In this dataset, 241 instances were diagnosed as malignant, and 458 instances were found to be benign. Malignant cases have a class of four, and benign cases have a class of two. This class, or diagnosis, is the outcome of the instance.

The features of the dataset are characteristics identified or calculated from each FNA image. There are nine features in this dataset, and each is valued on a scale of 1 to 10 for a particular instance, 1 being the closest to benign and 10 being the most malignant [ 18 ]. Features range from descriptors of cell characteristics, such as Uniformity of Cell Size and Uniformity of Cell Shape , to more complex cytological characteristics such as Clump Thickness and Marginal Adhesion . All nine features, along with the Instance No., Sample I.D., and Class are listed in Table  1 . The full dataset is a matrix of 699 × 12 (one identification number, nine features, and one outcome per instance).

This dataset is simple and therefore computationally efficient. The relatively low number of features and instances means that the analysis provided in this paper can be conducted using most modern PCs without long computing times. Although the principals are the same as those described throughout the rest of this paper, using large datasets to train Machine learning algorithms can be computationally intensive and, in some cases, require many days to complete. The principals illustrated here apply to datasets of any size.

The R Statistical Programming Language is an open-source tool for statistics and programming which was developed as an extension of the S language. R is supported by a large community of active users and hosts several excellent packages for ML which are both flexible and easy to use. R is a computationally efficient language which is readily comprehensible without special training in computer science. The R language is similar to many other statistical programming languages, including MATLAB, SAS, and STATA. Packages for R are arranged into different task views on the Comprehensive R Archive Network. The Machine Learning and Statistical Learning task view currently lists almost 100 packages dedicated to ML.

Many, if not most, R users access the R environment using RStudio, an open-source integrated developer environment (IDE) which is designed to make working in R more straightforward. We recommend that readers of the current paper download the latest version of both R and RStudio and access the environment through the RStudio application. Both R and RStudio are free to use and available for use under an open-source license.

Conducting a machine learning analysis

The following section will take you through the necessary steps of a ML analysis using the Wisconsin Cancer dataset.

Importing and preparing the dataset.

Training the ML algorithms.

Testing the ML algorithms.

Assessing sensitivity, specificity and accuracy of the algorithms.

Plotting receiver operating characteristic curves.

Applying new data to the trained models.

1. Importing and preparing the dataset.

The dataset can be downloaded directly from the UCI repository using the code in Fig.  5 .

Import the data and label the columns

We first modify the data by re-scoring missing data from ‘?’ to NA, removing any rows with missing data and re-scoring the class variables from 2 and 4 to 0 and 1, where 0 indicates the tumour was benign and 1 indicates that it was malignant. Recall that a dataset with many missing data points is referred to as a sparse dataset. In this dataset there are small number of cases (n =16) with at least one missing value. To simplify the analytical steps, we will remove these cases, using the code in Fig.  6 .

Remove missing items and restore the outcome data

Datasets used for supervised ML are most easily represented in a matrix similar to the way Table  1 is presented. The n columns are populated with the n −1 features, with the single remaining column containing the outcome. Each row contains an individual instance. The features which make up the training dataset may also be described as inputs or variables and are denoted in code as x . The outcomes may be referred to as the label or the class and are denoted using y .

Recall that it is necessary to train a supervised algorithm on a training dataset in order to ensure it generalises well to new data. The code in Fig.  7 will divide the dataset into two required segments, one which contains 67% of the dataset, to be used for training; and the other, to be used for evaluation, which contains the remaining 33%.

Split the data into training and testing datasets

2. Training the ML algorithms

Now that we have arranged our dataset into a suitable format, we may begin training our algorithms. These ML algorithms which we will use are listed below and detailed in the following section.

Logistic regression using Generalised Linear Models (GLMs) with \(\mathscr {L}_{1}\) Least Absolute Selection and Shrinkage Operator (LASSO) regularisation.

Support Vector Machines (SVMs) with a radial basis function (RBF) kernel.

Artificial Neural Networks (ANNs) with a single hidden layer.

Regularised regression using Generalised Linear Models (GLMs)

Regularised General Linear Models (GLMs) have demonstrated excellent performance in some complex learning problems, including predicting individual traits from on-line digital footprints [ 20 ], classifying open-text reports of doctors’ performance [ 7 ], and identifying prostate cancer by desorption electro-spray ionization mass spectrometric imaging of small metabolites and lipids [ 21 ].

When fitting GLMs using datasets which have a large number of features and substantial sparsity, model performance may be increased when the contribution of each of the included features to the model is reduced (or penalised) using regularisation, a process which also reduces the risk of over-fitting. Regularisation effectively reduces both the number of coefficients in the model and their magnitudes, making especially it suitable for big datasets that may have more features than instances. In this example, feature selection is guided by the Least Absolute Shrinkage and Selection Operator (LASSO). Other forms of regularisation are available, including Ridge Regression and the Elastic Net (which is a linear blend of both Ridge and LASSO regularisation) [ 22 ]. An accessible, up-to-date summary of LASSO and other regularisation techniques is given in Ref [ 23 ].

Regularised GLMs are operationalised in R using the glmnet package [ 24 ]. The code below demonstrates how the GLM algorithm is fitted to the training dataset. In the glmnet package, the regularistion parameter is chosen using the numerical value referred to as alpha. In this package, a alpha value of 1 selects LASSO regularisation where as alpha 0 selects Ridge regularization, a value between between 0 and 1 selects a linear blend of the two techniques known as the Elastic Net [ 22 ].

nFold cross-validation is used to ascertain the optimal value of lambda ( λ ), the regularisation parameter. The value of ( λ ) which minimizes prediction error is stored in the glm_model$lambda.min object. The smaller the λ value, the greater the effect of regularisation upon the number of features in the model and their respective coefficients. Figure  8 shows the effect of different levels of log( λ ). The optimal value of log( λ ) is indicated using the vertical broken line (shown here at x = -5.75). The rightmost dotted line indicates the most parsimonious value of log( λ ) which is within 1 standard deviation of the absolute minimum value. Note that the random nature of cross-validation means that values of log( λ ) may differ slightly between analyses. The integers are given above Fig.  8 (0-9) relate to the number of features included in the model. The code shown in Fig.  9 fits the GLM algorithm to the data and extracts the minimum value of λ and the weights of the coefficients.

Regression coefficients for the GLM model. The figure shows the coefficients for the 9 model features for different values of log( λ ). log( λ ) values are given on the lower x-axis and number of features in the model are displayed above the figure. As the size of log( λ ) decreases the number of variables in the model (i.e. those with a nonzero coefficient) increases as does the magnitude of each feature. The vertical dotted line indicates the value of log( λ ) at which the accuracy of the predictions is maximized

Fit the GLM model to the data and extract the coefficients and minimum value of lambda

Figure  10 shows the cross-validation curves for different levels of log( λ ). This figure can be plotted using the code in Fig.  11 .

Cross-validation curves for the GLM model. The figure shows the cross-validation curves as the red dots with upper and lower standard deviation shown as error bars

Plot the cross-validation curves for the GLM algorithm

Figure  8 shows magnitude of the coefficients for each of the variables within the model for different values of log( λ ). The vertical dotted line indicates the value of log( λ ) which minimises the mean squared error established during cross-validation. This figure can be augmented with a dotted vertical line indicating the value of log( λ ) using the abline() function, shown in Fig.  12 .

Plot the coefficients and their magnitudes

Support Vector Machines (SVMs)

Support Vector Machine (SVM) classifiers operate by separating the two classes using a linear decision boundary called the hyperplane. The hyperplane is placed at a location that maximises the distance between the hyperplane and instances [ 25 ].

Fig.  13 depicts an example of a linear hyperplane that perfectly separates between two classes. In real-world examples, it may not be possible to adequately separate the two classes using a linear hyperplane. By maximising the width of the decision boundary then the generalisability of the model to new data is optimised. Rather than employ a non-linear separator such as a high-order polynomial, SVM techniques use a method to transform the feature space such that the classes do become linearly separable. This technique, known as the kernel trick, is demonstrated in Fig.  14 .

A SVM Hyperplane The hyperplane maximises the width of the decision boundary between the two classes

The kernel trick The kernel trick modifies the feature space allowing separation of the classes with a linear hyperplane

Fig.  14 shows an example of a two classes that are not separable using a linear separator. By projecting the data to X 2 , they become linearly separable using the y =5 hyperplane. A popular method for kernel transformation in high-dimensional space is the radial basis function (RBF).

The SVM algorithm is fitted to the data using a function, given in Fig.  15 , which is arranged in a similar way to the regularised regression shown above.

Fit the SVM algorithm to the data

Further exploration of SVM which attempt to fit separating hyperplanes following different feature space transformations is possible by altering the kernel argument to “linear”, “radial”, “polynomial”, or “sigmoid”.

Artificial Neural Networks (ANNs)

Artificial Neural Networks (ANNs) are algorithms which are loosely modelled on the neuronal structure observed in the mammalian cortex. Neural networks are arranged with a number of input neurons, which represent the information taken from each of the features in the dataset. which feed into any number of hidden layers before passing to an output layer in which the final decision is presented. As information passes through the ’neurons’, or nodes, where is is multiplied by the weight of the neuron (plus a constant bias term) and transformed by an activation function. The activation function applies a non-linear transformation using a simple equation shown in Eq. 1 .

In recurrent ANNs, a process is undertaken in which the prediction errors are fed back through the network before modifying the weights of each neural connection is altered until error level is minimised, a process known as backpropagation [ 26 ].

Deep Neural Networks (DNNs) refers to neural networks which have many hidden layers. Deep learning, which may utilise DNNs, has produced impressive results when employed in complex tasks using very high dimensional data, such as image recognition [ 27 ] and computer-assisted diagnosis of melanoma [ 2 ].

DNNs are heavily parametrised and, resultantly, can be prone to over-fitting models to data. Regularisation can, like the GLM algorithm described above, be used prevent this. Other strategies to improve performance can include dropout regularisation, where some number of randomly-selected units are omitted from the hidden layers during training [ 28 ].

The code in Fig.  16 demonstrates the code for fitting a neural network. This is straightforward, requiring the x and y datasets to be defined, as well as the number of units in the hidden layer using the size argument.

Fit the ANN algorithm to the data

3. Testing the ML algorithms

In order to test the performance of the trained algorithms, it is necessary to compare the predictions which the algorithm has made on data other than the data upon which it was trained with the true outcomes for that data which we have known but we did not expose the algorithm to. To accomplish this in he R programming environment, we would create a vector of model predictions using the x_test matrix, which can be compared to the y_test vector to establish performance metrics. This is easily achievable using the predict() function, which is included in the stats package in the R distribution. The nnet package contains a minor modification to the predict() function, and as such the type argument is set to ‘raw’, rather than ‘response’ for the neural network. This code is given in Fig.  17 .

Extract predictions from the trained models on the new data

4. Assessing the sensitivity, specificity and accuracy of the algorithms

Machine learning algorithms for classification are typically evaluated using simple methodologies that will be familiar to many medical researchers and clinicians. In the current study, we will use sensitivity, specificity, and accuracy to evaluate the performance of the three algorithms. Sensitivity is the proportion of true positives that are correctly identified by the test, specificity is the proportion of true negatives that are correctly identified by the test and the accuracy is the proportion of the times which the classifier is correct [ 29 ]. Equations used to calculate sensitivity, specificity, and accuracy are given below.

Confusion matrices

Data from classifiers are often represented in a confusion matrix in which the classifications made by the algorithm (e.g., pred_y_svm ) are compared to the true classifications (which the algorithms were blinded to) in the dataset (i.e., y_test ). Once populated, the confusion matrix provides all of the information needed to calculate sensitivity, specificity, and accuracy manually. An example of an unpopulated confusion matrix is demonstrated in Table  2 .

Confusion matrices can be easily created in R using the caret package. The confusionMatrix() function creates a confusion matrix and calculates sensitivity, specificity, and accuracy. The confusionMatrix() function requires a binary input for the predictors whereas the pred() functions used earlier produce a vector of continuous values between 0 and 1, in which a larger value reflects greater certainty that the sample was positive. Before evaluating a binary classifier, a cut-off threshold must be decided upon. The round() function used in the code shown in Fig.  18 effectively sets a threshold of >.50 for a positive prediction by rounding values ≤.50 down to 0 and values >.50 up to 1. While this is sufficient for this teaching example, users may wish to evaluate the optimal threshold for a positive prediction as this may differ from.50. The populated confusion matrix for this example is shown in Table  3 and is displayed alongside sensitivity, specificity, and accuracy.

Create confusion matrices for the three algorithms

5. Plotting receiver operating characteristic curves

Receiver operating characteristics curves are useful and are shown in the code in Fig.  19 using the pROC package. An example output is given in Fig.  20 . These curves illustrate the relationship between the model’s sensitivity (plotted on the y -axis) and specificity (plotted on the x -axis). The grey diagonal line is reflective of as-good-as-chance performance and any curves which are plotted to the left of that line are performing better than chance. Interpretation of ROC curves is facilitated by calculating the area under each curve (AUC) [ 30 ]. The AUC gives a single value which explains the probability that a random sample would be correctly classified by each algorithm. In this example all models perform very well but the SVM algorithm shows the best performance, with AUC =.97 compared to the ANN (AUC =.95) and the LASSO-regularized regression (AUC =.94).

Draw received operating curves and calculate the area under them

Receiver Operating Characteristics curves

6. Applying new data to the trained models

Despite many similarities, ML is differentiated from statistical inference by its focus on predicting real-life outcomes from new data. As such, we develop models not to infer the relationships between variables but rather to produce reliable predictions from new data (though, as we have demonstrated, prediction and inference are not mutually exclusive).

In order to use the trained models to make predictions from data we need to construct either a vector (if there is a single new case) or a matrix (if there are multiple new cases). We need to ensure that the new data are entered into the model in the same order as the x_train and x_test matrices. In this case, we need to enter new data in the order of thickness , cell size , cell shape , adhesion , epithelial size , bare nuclei , bland cromatin , normal nucleoli , and mitoses . The code in Fig.  21 demonstrates how these data are represented in a manner that allows them to be processed by the trained model. Note that all three algorithms return predictions that suggest there is a near-certainty that this particular sample is malignant.

Apply new data to the trained and validated algorithm

Additional ML techniques

Reducing prediction error; the case for ensembles..

When working to maximise the performance of a predictive model, it can be beneficial to group different algorithms together to create a more robust prediction in a process known as ensemble learning [ 24 ]. There are too many ensemble techniques to adequately summarize here, but more information can be found in Ref. [ 23 ].

The principal of ensemble learning can be demonstrated using a un-weighted voting algorithm with R code. The code in Fig.  22 can be used to demonstrate the process of developing both an averaging and and voting algorithm.

Create predictions from the ensemble

Natural language processing

Another common use for classification algorithms is in Natural Language Processing (NLP), the branch of ML in which computers are taught to interpret linguistic data. One popular example of NLP is in sentiment analysis, which involves ML algorithms trained to classify texts into different categories relating to the sentiment they convey; usually positive, negative, or neutral. We will give an overview of how features can be extracted from text and then used in the framework we have introduced above.

A linguistic dataset (also known as a corpus ) comprises a number of distinct documents . The documents can be broken down into smaller tokens of text, such as the individual words contained within. These tokens can be used as the features in a ML analysis as demonstrated above. In such an analysis, we arrange the x_train matrix such that the rows represent the individual documents and the tokenized features are represented in the columns. This arrangement for linguistic analysis is known as a term-document matrix (TDM).

In its most basic form, each row of the TDM represents a simple count of the words which were used in a document. In this case, the width of a TDM is equal to the number of unique words in the entire corpus and, for each document, the value any given cell will either be 0 if the word does not appear in that comment or 1 if it does. Arranging a document this way leads to two issues: firstly, that the majority of the matrix likely contains null values (an issue known as sparsity ); and secondly, that many of the documents contain the most common words in a language (e.g., “the”, “a”, or “and”) which are not very informative in analysis. Refining the TDM using a technique known as a term-frequency-inverse document frequency (TF-IDF) weighting can reduce the value of certain common words in the matrix which may be less informative and increase the value of less common words, which may be more informative. It is also possible to remove uninformative words using a pre-defined dictionary known as a stop words dictionary.

In a TDM, words can be tokenized individually, known as unigrams , or as groups of sequential words, known a nGrams where n is the number of words extracted in the token ( i.e, bi-gram or tri-gram extraction ). Such extraction can mitigate issues caused by grammatical nuances such as negation (e.g., “I never said she stole my money.”). Some nuances are more difficult to analyse robustly, especially those used commonly in spoken language, such as emphasis or sarcasm. For example, the sentence above about the stolen money could have at least 7 different meanings depending on where the emphasis was placed.

A TDM can be easily developed in R using the tools provided in the tm package. In Table  4 , we demonstrate a simple uniGram (single word) TDM without TF-IDF weighting.

The code in Fig.  23 demonstrates the process for creating a term document management for a vector of open-text comments called ’comments’. modifications are made to the open text comments including the removal of punctuation and weighting using the TF-DF technique. The final matrix which is saved to an objects names ’x’ could The linked to a vector of outcomes ‘y’ and used to train and validate machine learning algorithms using the process described above listings 3 to 11.

Create a term document matrix

Once created, documents in the TDM can be combined with a vector of outcomes using the cbind() function, as shown in Table  4 , and processed in the same way as demonstrated in Fig.  7 . Interested readers can explore the informative tm package documentation to learn more about term-document matrices [ 31 ].

When trained on a proportion of the dataset, the three algorithms were able to classify cell nuclei in the remainder of the dataset with high accuracy (.94 -.96), sensitivity (.97 -.99), and specificity (.85 -.94). Though each algorithm performed well individually, maximum accuracy (.96) and area under the curve (.97) was achieved using the SVM algorithm (see Table  3 ).

Model performance was marginally increased when the three algorithms were arranged into a voting ensemble, with an overall accuracy of.97, sensitivity of.99 and specificity of.95 (see the attached R Code for further details.).

Machine learning has the potential to transform the way that medicine works [ 32 ], however, increased enthusiasm has hitherto not been met by increased access to training materials aimed at the knowledge and skill sets of medical practitioners.

In this paper, we introduce basic ML concepts within a context which medical researchers and clinicians will find familiar and accessible. We demonstrate three commonly-used algorithms; a regularized general linear model, support vector machines (SVM), and an artificial neural network to classify tumour biopsies with high accuracy as either benign or malignant. Our results show that all algorithms can perform with high accuracy, sensitivity, and specificity despite substantial differences in the way that the algorithms work. The best-performing algorithm, the SVM, is very similar to the method demonstrated by Wolberg and Mangasarian who used different versions of the same dataset with fewer observations to achieve similar results [ 18 , 33 ]. It is noteworthy that the LASSO-regularized linear regression also performed exceptionally well whilst preserving the ability to understand which features were guiding the predictions (see Table  5 ). In contrast, the archetypal ’black box’ of the heavily-parametrized neural network could not improve classification accuracy.

In parallel to our analysis, we demonstrate techniques which can be applied with a commonly-used and open-source programming software (the R environment) which does not require prior experience with command-line computing. The presented code is designed to be re-usable and easily adaptable, so that readers may apply these techniques to their own datasets. With some modification, the same code may be used to develop linguistic classifiers or object recognition algorithms using open-text or image-based data respectively. Though the R environment now provides many options for advanced ML analyses, including deep learning, the framework of the code can be easily translated to other programming languages, such as Python, if desired. After working through examples in this paper we suggest that user apply their knowledge to problems within their own datasets. Doing so will elucidate specific issue which need to be overcome and will form a foundation for continued learning in this area. Further information can be from any number of excellent textbooks, websites, and online courses. Additional practice data sets can be obtained from the University of California Irvine Machine learning data sets repository which at the time of writing, includes an additional 334 datasets suitable for classification tasks, including 35 which contain open-text data [ 17 ].

Further, this paper acts to demystify ML and endow clinicians and researchers without a previous ML experience with the ability to critically evaluate these techniques. This is particularly important because without a clear understanding of the way in which algorithms are trained, medical practitioners are at risk of relying too heavily on these tools which might not always perform as expected. In their paper demonstrating a multi-surface pattern separation technique using a similar dataset, Wolberg and Mangasarian stress the importance of training algorithms on data which does not itself contain errors; their model was unable to achieve perfect performance as the sample in the dataset appeared to have been incorrectly extracted from an area beyond the tumour. The oft-told parable of the failure of the Google Flu Trends model offers an accessible example of the risks and consequences posed by a lack of understanding of ML models deployed ostensibly to improve health [ 34 ]. In short, the Google Flu Trends model was not generalizable over time as the Google Search data it was trained on was temporally sensitive. Looking to applications of ML beyond the medical field offers further insight into some risks that these algorithms might engender. For example, concerns have been raised about predictive policing algorithms and, in particular, the risk of entrenching certain prejudices in an algorithm which may be apparent in police practice. Though the evidence of whether predictive policing algorithms leads to biases in practice is unclear [ 35 ], it stands to reason that if biases exist in routine police work then models taught to recognize patterns in routinely collected data would have no means to exclude these biases when making predictions about future crime risk. Similar bias-based risks have been identified in some areas of medical practice and, if left unchecked, threaten the ethical use of data-driven automation in those areas [ 36 ]. An understanding of the way ML algorithms are trained is essential to minimize and mitigate the risks of entrenching biases in predictive algorithms in medicine.

The approach which we have taken in this paper entails some notable strengths and weaknesses. We have chosen to use a publicly-available dataset which contains a relatively small number of inputs and cases. The data is arranged in such a way that will allow those trained in medical disciplines to easily draw parallels between familiar statistical and novel ML techniques. Additionally, the compact dataset enables short computational times on almost all modern computers. A caveat of this approach is that many of the nuances and complexities of ML analyses, such as sparsity or high dimensionality, are not well represented in the data. Despite the omission of these common features of a ML dataset, we are confident that users who have worked through the examples given here with the code provided in the appendix will be well-placed to further develop their skills working on more complex datasets using the scalable code framework which we provide. In addition, this data also usefully demonstrates an important principle of ML: more complex algorithms do not necessarily beget more useful predictions.

We look toward a future of medical research and practice greatly enhanced by the power of ML. In the provision of this paper, we hope that the enthusiasm for new and transformative ML techniques is tempered by a critical appreciation for the way in which they work and the risks that they could pose.

Abbreviations

Artificial neural network

Area under the curve

Fine needle aspiration

Generalized linear model

Integrated developer environment

Least absolute shrinkage and selection operator

Radial basis function

Received operating characteristics

Support vector machine

Term document - inverse document frequency

Term document matrix

t-embedded stochastic neighbor embedding

University of California, Irvine

Jordan MI, Mitchell TM. Machine learning: Trends, perspectives, and prospects. Sci (NY). 2015; 349(6245):255–60. https://doi.org/10.1126/science.aaa8415 .

Article   CAS   Google Scholar  

Esteva A, Kuprel B, Novoa RA, Ko J, Swetter SM, Blau HM, Thrun S. Dermatologist-level classification of skin cancer with deep neural networks. Nature. 2017; 542(7639):115–8. https://doi.org/10.1038/nature21056 .

Anderson J, Parikh J, Shenfeld D. Reverse Engineering and Evaluation of Prediction Models for Progression to Type 2 Diabetes: Application of Machine Learning Using Electronic Health Records. J Diabetes. 2016.

Ong M-S, Magrabi F, Coiera E. Automated identification of extreme-risk events in clinical incident reports. J Am Med Inform Assoc. 2012; 19(e1):e110–e18.

Article   Google Scholar  

Greaves F, Ramirez-Cano D, Millett C, Darzi A, Donaldson L. Use of sentiment analysis for capturing patient experience from free-text comments posted online,. J Med Internet Res. 2013; 15(11):239. https://doi.org/10.2196/jmir.2721 .

Hawkins JB, Brownstein JS, Tuli G, Runels T, Broecker K, Nsoesie EO, McIver DJ, Rozenblum R, Wright A, Bourgeois FT, Greaves F. Measuring patient-perceived quality of care in US hospitals using Twitter,. BMJ Qual Saf. 2016; 25(6):404–13. https://doi.org/10.1136/bmjqs-2015-004309 .

Gibbons C, Richards S, Valderas JM, Campbell J. Supervised Machine Learning Algorithms Can Classify Open-Text Feedback of Doctor Performance With Human-Level Accuracy,. J Med Internet Res. 2017; 19(3):65. https://doi.org/10.2196/jmir.6533 .

Wagland R, Recio-Saucedo A, Simon M, Bracher M, Hunt K, Foster C, Downing A, Glaser A, Corner J. Development and testing of a text-mining approach to analyse patients’ comments on their experiences of colorectal cancer care. Qual Saf BMJ. 2015:2015–004063. https://doi.org/10.1136/bmjqs-2015-004063 .

Bedi G, Carrillo F, Cecchi GA, Slezak DF, Sigman M, Mota NB, Ribeiro S, Javitt DC, Copelli M, Corcoran CM. Automated analysis of free speech predicts psychosis onset in high-risk youths. npj Schizophr. 2015; 1(1):15030. https://doi.org/10.1038/npjschz.2015.30 .

Friedman CP, Wong AK, Blumenthal D. Achieving a Nationwide Learning Health System. Sci Transl Med. 2010; 2(57):57–29.

Beam A, Kohane I. Big Data and Machine Learning in Health Care. J Am Med Assoc. 2018; 319(13):1317–8.

Lei T, Barzilay R, Jaakkola T. Rationalizing Neural Predictions. In: Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining - KDD ’16: 2016. p. 1135–1144. https://doi.org/10.1145/2939672.2939778 .

Mangasarian OL, Street WN, Wolberg WH. Breast Cancer Diagnosis and Prognosis via Linear Programming: AAAI; 1994, pp. 83 - 86.

Jolliffe I, Jolliffe I. Principal Component Analysis. In: Wiley StatsRef: Statistics Reference Online. Chichester: John Wiley & Sons, Ltd: 2014.

Google Scholar  

Blei DM, Ng AY, Jordan MI. Latent Dirichlet Allocation. J Mach Learn Res. 2003; 3(Jan):993–1022.

Maaten Lvd, Hinton G. Visualizing Data using t-SNE. J Mach Learn Res. 2008; 9(Nov):2579–605.

Lichman M. UCI Machine Learning Repository: Breast Cancer Wisconsin (Diagnostic) Data Set. 2014. http://archive.ics.uci.edu/ml . Accessed 8 Aug 2017.

Wolberg WH, Mangasariant OL. Multisurface method of pattern separation for medical diagnosis applied to breast cytology. Proc Natl Acad Sci USA. 1990; 87:9193–6.

Bennett KP. Decision tree construction via linear programming: University of Wisconsin-Madison Department of Computer Sciences; 1992, pp. 97–101.

Kosinski M, Stillwell D, Graepel T. Private traits and attributes are predictable from digital records of human behavior. Proc Natl Acad Sci. 2013; 110(15):5802–5. https://doi.org/10.1073/pnas.1218772110 .

Banerjee S, Zare RN, Tibshirani RJ, Kunder CA, Nolley R, Fan R, Brooks JD, Sonn GA. Diagnosis of prostate cancer by desorption electrospray ionization mass spectrometric imaging of small metabolites and lipids. Proc Natl Acad Sci U S A. 2017; 114(13):3334–9. https://doi.org/10.1073/pnas.1700677114 .

Zou H, Zou H, Hastie T. Regularization and variable selection via the Elastic Net. J R Stat Soc Ser B. 2005; 67:301–20.

Efron B, Hastie T. Computer Age Statistical Inference, 1st edn. Cambridge: Cambridge University Press; 2016.

Book   Google Scholar  

Hastie T, Tibshirani R, Friedman J. Elements of statistical learning. 2001; 1(10). New York: Springer series in statistics.

Cortes C, Vapnik V. Support-vector networks. Mach Learn. 1995; 20(3):273–97. https://doi.org/10.1007/BF00994018 .

Hecht-Nielsen. Theory of the backpropagation neural network. 1989:593–605. https://doi.org/10.1109/IJCNN.1989.118638 .

Krizhevsky A, Sutskever I, Hinton GE. ImageNet Classification with Deep Convolutional Neural Networks. In: Advances in neural information processing systems: 2012. p. 1097–1105.

Dahl GE, Sainath TN, Hinton GE. Improving deep neural networks for LVCSR using rectified linear units and dropout. 2013:8609–8613. https://doi.org/10.1109/ICASSP.2013.6639346 .

Martin Bland J, Altman D. Statistical methods for assessing agreement between two methods of clinical measurement. Lancet. 1986; 327(8476):307–10. https://doi.org/10.1016/S0140-6736(86)90837-8 .

Hanley JA, McNeil BJ. The meaning and use of the area under a receiver operating characteristic (ROC) curve,. Radiology. 1982; 143(1):29–36. https://doi.org/10.1148/radiology.143.1.7063747 .

Meyer D, Hornik K, Fienerer I. Text mining infrastructure in R. J Stat Softw. 2008; 25(5):1–54.

Darcy AM, Louie AK, Roberts LW. Machine Learning and the Profession of Medicine. J Am Med Assoc. 2016; 315(6):551. https://doi.org/10.1001/jama.2015.18421 .

Wolberg WH, Street WN, Mangasarian OL. Machine learning techniques to diagnose breast cancer from image-processed nuclear features of fine needle aspirates. Cancer Lett. 1994; 77(2-3):163–71. https://doi.org/10.1016/0304-3835(94)90099-X .

Lazer D, Kennedy R, King G, Vespignani A. The Parable of Google Flu: Traps in Big Data Analysis. Science. 2014; 343(6176):1203–5. https://doi.org/10.1126/science.1248506 .

Brantingham PJ, Valasik M, Mohler GO. Does Predictive Policing Lead to Biased Arrests? Results From a Randomized Controlled Trial. Stat Public Policy. 2018; 5(1):1–6. https://doi.org/10.1080/2330443X.2018.1438940 .

Haider AH, Chang DC, Efron DT, Haut ER, Crandall M, Cornwell EE. Race and Insurance Status as Risk Factors for Trauma Mortality. Arch Surg. 2008; 143(10):945. https://doi.org/10.1001/archsurg.143.10.945 .

Download references

Acknowledgments

We acknowledge and thank the investigators, scientists, and developers who have contributed to the scientific community by making their data, code, and software freely available. We thank our colleagues in Cambridge, Boston, and beyond who provided critical insight into this work.

CSG was funded by National Institute for Health Research Trainees Coordinating Centre Fellowships (NIHR-PDF-2014-07-028 and NIHR-CDF-2017-10-19). The funders had no role in the design or execution of this study.

Availability of data and materials

In this manuscript we use de-identified data from a public repository [ 17 ]. The data are included on the BMC Med Res Method website. As such, ethical approval was not required.

Author information

Authors and affiliations.

Department of Engineering, University of Cambridge, Trumpington Street, Cambridge, CB2 1PZ, UK

Jenni A. M. Sidey-Gibbons

Department of Surgery, Harvard Medical School, 25 Shattuck Street, Boston, 01225, Massachusetts, USA

Chris J. Sidey-Gibbons

Department of Surgery, Brigham and Women’s Hospital, 75 Francis Street, Boston, 01225, Massachusetts, USA

University of Cambridge Psychometrics Centre, Trumpington Street, Cambridge, CB2 1AG, UK

You can also search for this author in PubMed   Google Scholar

Contributions

JSG contributed to the conception and design of the work, interpretation of data and presentation of results, and drafted the manuscript. CSG contributed to the conception and design of the work, conducted the analyses, and drafted the manuscript. Both JSG and CSG approve of the final versions and agree to be accountable for their own contributions. Both authors read and approved the final manuscript.

Corresponding author

Correspondence to Chris J. Sidey-Gibbons .

Ethics declarations

Ethics approval and consent to participate, consent for publication.

All contributing parties consent for the publication of this work.

Competing interests

The authors report no competing interests relating to this work.

Publisher’s Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Additional files

Additional file 1.

Breast Cancer Wisconsin Dataset. Anonomised dataset used in this work. (CSV 24.9 kb)

Additional file 2

R Markdown Supplementary Material. R Code accompanying the work described in this paper and its output. (PDF 207 kb)

Rights and permissions

Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License ( http://creativecommons.org/licenses/by/4.0/ ), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver ( http://creativecommons.org/publicdomain/zero/1.0/ ) applies to the data made available in this article, unless otherwise stated.

Reprints and permissions

About this article

Cite this article.

Sidey-Gibbons, J., Sidey-Gibbons, C. Machine learning in medicine: a practical introduction. BMC Med Res Methodol 19 , 64 (2019). https://doi.org/10.1186/s12874-019-0681-4

Download citation

Received : 11 June 2018

Accepted : 14 February 2019

Published : 19 March 2019

DOI : https://doi.org/10.1186/s12874-019-0681-4

Share this article

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

• Medical informatics
• Classification
• Supervised machine learning
• Programming languages
• Computer-assisted
• Decision making

BMC Medical Research Methodology

ISSN: 1471-2288

• General enquiries: [email protected]

IEEE Account

• Change Username/Password
• Update Address

Purchase Details

• Payment Options
• Order History
• View Purchased Documents

Profile Information

• Communications Preferences
• Profession and Education
• Technical Interests
• US & Canada: +1 800 678 4333
• Worldwide: +1 732 981 0060
• Contact & Support
• About IEEE Xplore
• Accessibility
• Terms of Use
• Nondiscrimination Policy
• Privacy & Opting Out of Cookies

A not-for-profit organization, IEEE is the world's largest technical professional organization dedicated to advancing technology for the benefit of humanity. © Copyright 2024 IEEE - All rights reserved. Use of this web site signifies your agreement to the terms and conditions.

• Academic Journals
• Books & Monographs
• Conferences
• Standard Editing Service
• Expert Scientific Editing Service
• Academic Translation Service
• News & Announcements

Table of Content

• Introduction
• Research Methodology

Medical Diagnosis Using Machine Learning: A Statistical Review

Kaustubh Arun Bhavsar 1 , Jimmy Singla 1 , Yasser D. Al-Otaibi 2 , Oh-Young Song 3,* , Yousaf Bin Zikria 4 , Ali Kashif Bashir 5

1 Lovely Professional University, Jalandhar, 144411, India 2 Department of Information Systems, Faculty of Computing and Information Technology in Rabigh, King Abdulaziz University, Jeddah, 21589, Saudi Arabia 3 Software Department, Sejong University, Seoul, 05006, Korea 4 Department of Information and Communication Engineering, Yeungnam University, Gyeongsan, 38541, Korea 5 Department of Computing and Mathematics, Manchester Metropolitan University, Manchester, M15 6BH, UK

(This article belongs to the Special Issue: Intelligent Decision Support Systems for Complex Healthcare Applications )

Computers, Materials & Continua 2021 , 67 (1), 107-125. https://doi.org/10.32604/cmc.2021.014604

Received 02 October 2020; Accepted 20 October 2020; Issue published 12 January 2021

Cite This Article

• Full-Text PDF
• Full-Text HTML
• Full-Text XML
• Full-Text Epub

Related articles

Crowdsourcing-Based Framework for Teaching Quality Evaluation and Feedback Using Linguistic 2-Tuple

An Improved Memory Cache Management Study Based on Spark

An Empirical Comparison on Multi-Target Regression Learning

Machine Learning Based Resource Allocation of Cloud Computing in Auction

Feature Selection Method Based on Class Discriminative Degree for Intelligent Medical Diagnosis

Further Information

About Tech Science Press

Open Access Policy

Article Processing Charges

Special Issue Policy

Research Topic Policy

Terms and Conditions

Privacy Policy

Advertising Policy

Contact TSP

For Editors

For Reviewers

For Authors

For Conference Organizers

For Subscribers

Join TSP editorial community

Disease Prediction From Various Symptoms Using Machine Learning

7 Pages Posted: 8 Oct 2020

Rinkal Keniya

K. J. Somaiya College of Engineering

Aman Khakharia

K. J. Somaiya college of engineering

Vruddhi Shah

University of Mumbai - K. J. Somaiya College of Engineering (K.J.S.C.E.)

Vrushabh Gada

Ruchi manjalkar, tirth thaker, mahesh warang, ninad mehendale.

University of Mumbai - K. J. Somaiya College of Engineering (K.J.S.C.E.); Ninad's research Lab

Date Written: July 27, 2020

Accurate and on-time analysis of any health-related problem is important for the prevention and treatment of the illness. The traditional way of diagnosis may not be sufficient in the case of a serious ailment. Developing a medical diagnosis system based on machine learning (ML) algorithms for prediction of any disease can help in a more accurate diagnosis than the conventional method. We have designed a disease prediction system using multiple ML algorithms. The data set used had more than 230 diseases for processing. Based on the symptoms, age, and gender of an individual, the diagnosis system gives the output as the disease that the individual might be suffering from. The weighted KNN algorithm gave the best results as compared to the other algorithms. The accuracy of the weighted KNN algorithm for the prediction was 93.5 %. Our diagnosis model can act as a doctor for the early diagnosis of a disease to ensure the treatment can take place on time and lives can be saved.

Keywords: Disease Prediction, Machine Learning, Symptoms

JEL Classification: I

Suggested Citation: Suggested Citation

K. J. Somaiya College of Engineering ( email )

K. j. somaiya college of engineering ( email ), university of mumbai - k. j. somaiya college of engineering (k.j.s.c.e.) ( email ).

Mumbai, MA Maharashtra 400007 India

Ninad Mehendale (Contact Author)

Ninad's research lab ( email ).

M.G. Road, Naupada Thane Thane, 400602 India

Do you have a job opening that you would like to promote on SSRN?

Paper statistics, related ejournals, computation theory ejournal.

Subscribe to this fee journal for more curated articles on this topic

Public Health eJournal

• Health Care Services
• Public Health
• Delivery of Health Care

Machine learning in healthcare diagnosis

• November 2020
• In book: Blockchain and Machine Learning for e-Healthcare Systems (pp.343-366)

• Pondicherry Engineering College

• IFET College of Engineering

• Galgotias University

• government polytechnic college Sevvapet

Abstract and Figures

Discover the world's research

• 25+ million members
• 160+ million publication pages
• 2.3+ billion citations
• Yingxue Mei

• Waleed Salih

• Seyma Yasar

• S. Rajeshwaran

• N. Ragapriya

• D. Raghu Raman
• A Prabagaran
• Vinay Dwivedi
• Padmanabhan P
• Anisha Sharma
• Samarth Gupta

• S Kalavathi
• Angelin Blessy J
• R Y. Rajesh

• H.D. Maheshappa
• SENSORS-BASEL

• Guanghong Gong

• Danni Cheng

• Yaping Wang
• Miguel Aires Barros Monteiro

• INT J ARTIF INTELL T
• Athanasios Tagaris

• H. M. Tarek Ullah
• ZishanAhmed Onik
• Riashat Islam

• Azhar Hussein Alkeshuosh

• Inas Al Mansoori

• Recruit researchers
• Join for free
• Login Email Tip: Most researchers use their institutional email address as their ResearchGate login Password Forgot password? Keep me logged in Log in or Continue with Google Welcome back! Please log in. Email · Hint Tip: Most researchers use their institutional email address as their ResearchGate login Password Forgot password? Keep me logged in Log in or Continue with Google No account? Sign up

• Open access
• Published: 29 August 2023

Healthcare predictive analytics using machine learning and deep learning techniques: a survey

• Mohammed Badawy   ORCID: orcid.org/0000-0001-9494-1386 1 ,
• Nagy Ramadan 1 &
• Hesham Ahmed Hefny 2  

Journal of Electrical Systems and Information Technology volume  10 , Article number:  40 ( 2023 ) Cite this article

14k Accesses

18 Citations

Metrics details

Healthcare prediction has been a significant factor in saving lives in recent years. In the domain of health care, there is a rapid development of intelligent systems for analyzing complicated data relationships and transforming them into real information for use in the prediction process. Consequently, artificial intelligence is rapidly transforming the healthcare industry, and thus comes the role of systems depending on machine learning and deep learning in the creation of steps that diagnose and predict diseases, whether from clinical data or based on images, that provide tremendous clinical support by simulating human perception and can even diagnose diseases that are difficult to detect by human intelligence. Predictive analytics for healthcare a critical imperative in the healthcare industry. It can significantly affect the accuracy of disease prediction, which may lead to saving patients' lives in the case of accurate and timely prediction; on the contrary, in the case of an incorrect prediction, it may endanger patients' lives. Therefore, diseases must be accurately predicted and estimated. Hence, reliable and efficient methods for healthcare predictive analysis are essential. Therefore, this paper aims to present a comprehensive survey of existing machine learning and deep learning approaches utilized in healthcare prediction and identify the inherent obstacles to applying these approaches in the healthcare domain.

Introduction

Each day, human existence evolves, yet the health of each generation either improves or deteriorates. There are always uncertainties in life. Occasionally encounter many individuals with fatal health problems due to the late detection of diseases. Concerning the adult population, chronic liver disease would affect more than 50 million individuals worldwide. However, if the sickness is diagnosed early, it can be stopped. Disease prediction based on machine learning can be utilized to identify common diseases at an earlier stage. Currently, health is a secondary concern, which has led to numerous problems. Many patients cannot afford to see a doctor, and others are extremely busy and on a tight schedule, yet ignoring recurring symptoms for an extended length of time can have significant health repercussions [ 1 ].

Diseases are a global issue; thus, medical specialists and researchers are exerting their utmost efforts to reduce disease-related mortality. In recent years, predictive analytic models has played a pivotal role in the medical profession because of the increasing volume of healthcare data from a wide range of disparate and incompatible data sources. Nonetheless, processing, storing, and analyzing the massive amount of historical data and the constant inflow of streaming data created by healthcare services has become an unprecedented challenge utilizing traditional database storage [ 2 , 3 , 4 ]. A medical diagnosis is a form of problem-solving and a crucial and significant issue in the real world. Illness diagnosis is the process of translating observational evidence into disease names. The evidence comprises data received from evaluating a patient and substances generated from the patient; illnesses are conceptual medical entities that detect anomalies in the observed evidence [ 5 ].

Healthcare is the collective effort of society to ensure, provide, finance, and promote health. In the twentieth century, there was a significant shift toward the ideal of wellness and the prevention of sickness and incapacity. The delivery of healthcare services entails organized public or private efforts to aid persons in regaining health and preventing disease and impairment [ 6 ]. Health care can be described as standardized rules that help evaluate actions or situations that affect decision-making [ 7 ]. Healthcare is a multi-dimensional system. The basic goal of health care is to diagnose and treat illnesses or disabilities. A healthcare system’s key components are health experts (physicians or nurses), health facilities (clinics and hospitals that provide medications and other diagnostic services), and a funding institution to support the first two [ 8 ].

With the introduction of systems based on computers, the digitalization of all medical records and the evaluation of clinical data in healthcare systems have become widespread routine practices. The phrase "electronic health records" was chosen by the Institute of Medicine, a division of the National Academies of Sciences, Engineering, and Medicine, in 2003 to define the records that continued to enhance the healthcare sector for the benefit of both patients and physicians. Electronic Health Records (EHR) are "computerized medical records for patients that include all information in an individual's past, present, or future that occurs in an electronic system used to capture, store, retrieve, and link data primarily to offer healthcare and health-related services," according to Murphy, Hanken, and Waters [ 8 ].

Daily, healthcare services produce an enormous amount of data, making it increasingly complicated to analyze and handle it in "conventional ways." Using machine learning and deep learning, this data may be properly analyzed to generate actionable insights. In addition, genomics, medical data, social media data, environmental data, and other data sources can be used to supplement healthcare data. Figure  1 provides a visual picture of these data sources. The four key healthcare applications that can benefit from machine learning are prognosis, diagnosis, therapy, and clinical workflow, as outlined in the following section [ 9 ].

Illustration of heterogeneous sources contributing to healthcare data [ 9 ]

The long-term investment in developing novel technologies based on machine learning as well as deep learning techniques to improve the health of individuals via the prediction of future events reflects the increased interest in predictive analytics techniques to enhance healthcare. Clinical predictive models, as they have been formerly referred to, assisted in the diagnosis of people with an increased probability of disease. These prediction algorithms are utilized to make clinical treatment decisions and counsel patients based on some patient characteristics [ 10 ].

The concept of medical care is used to stress the organization and administration of curative care, which is a subset of health care. The ecology of medical care was first introduced by White in 1961. White also proposed a framework for perceiving patterns of health concerning symptoms experienced by populations of interest, along with individuals’ choices in getting medical treatment. In this framework, it is possible to calculate the proportion of the population that used medical services over a specific period of time. The "ecology of medical care" theory has become widely accepted in academic circles over the past few decades [ 6 ].

Medical personnel usually face new problems, changing tasks, and frequent interruptions because of the system's dynamism and scalability. This variability often makes disease recognition a secondary concern for medical experts. Moreover, the clinical interpretation of medical data is a challenging task from an epistemological point of view. This not only applies to professionals with extensive experience but also to representatives, such as young physician assistants, with varied or little experience [ 11 ]. The limited time available to medical personnel, the speedy progression of diseases, and the fluctuating patient dynamics make diagnosis a particularly complex process. However, a precise method of diagnosis is critical to ensuring speedy treatment and, thus, patient safety [ 12 ].

Predictive analytics for health care are critical industry requirements. It can have a significant impact on the accuracy of disease prediction, which can save patients' lives in the case of an accurate and timely prediction but can also endanger patients' lives in the case of an incorrect prediction. Diseases must therefore be accurately predicted and estimated. As a result, dependable and efficient methods for healthcare predictive analysis are required.

The purpose of this paper is to present a comprehensive review of common machine learning and deep learning techniques that are utilized in healthcare prediction, in addition to identifying the inherent obstacles that are associated with applying these approaches in the healthcare domain.

The rest of the paper is organized as follows: Section  " Background " gives a theoretical background on artificial intelligence, machine learning, and deep learning techniques. Section  " Disease prediction with analytics " outlines the survey methodology and presents a literature review of machine learning as well as deep learning approaches employed in healthcare prediction. Section  " Results and Discussion " gives a discussion of the results of previous works related to healthcare prediction. Section  " Challenges " covers the existing challenges related to the topic of this survey. Finally, Section  " Conclusion " concludes the paper.

The extensive research and development of cutting-edge tools based on machine learning and deep learning for predicting individual health outcomes demonstrate the increased interest in predictive analytics techniques to improve health care. Clinical predictive models assisted physicians in better identifying and treating patients who were at a higher risk of developing a serious illness. Based on a variety of factors unique to each individual patient, these prediction algorithms are used to advise patients and guide clinical practice.

Artificial intelligence (AI) is the ability of a system to interpret data, and it makes use of computers and machines to improve humans' capacity for decision-making, problem-solving, and technological innovation [ 13 ]. Figure  2 depicts machine learning and deep learning as subsets of AI.

AI, ML, and DL

Machine learning

Machine learning (ML) is a subfield of AI that aims to develop predictive algorithms based on the idea that machines should have the capability to access data and learn on their own [ 14 ]. ML utilizes algorithms, methods, and processes to detect basic correlations within data and create descriptive and predictive tools that process those correlations. ML is usually associated with data mining, pattern recognition, and deep learning. Although there are no clear boundaries between these areas and they often overlap, it is generally accepted that deep learning is a relatively new subfield of ML that uses extensive computational algorithms and large amounts of data to define complex relationships within data. As shown in Fig.  3 , ML algorithms can be divided into three categories: supervised learning, unsupervised learning, and reinforcement learning [ 15 ].

Different types of machine learning algorithms

Supervised learning

Supervised learning is an ML model for investigating the input–output correlation information of a system depending on a given set of training examples that are paired between the inputs and the outputs [ 16 ]. The model is trained with a labeled dataset. It matches how a student learns fundamental math from a teacher. This kind of learning requires labeled data with predicted correct answers based on algorithm output [ 17 ]. The most widely used supervised learning-based techniques include linear regression, logistic regression, decision trees, random forests, support vector machines, K-nearest neighbor, and naive Bayes.

A. Linear regression

Linear regression is a statistical method commonly used in predictive investigations. It succeeds in forecasting the dependent, output, variable (Y) based on the independent, input, variable (X). The connection between X and Y is represented as shown in Eq.  1 assuming continuous, real, and numeric parameters.

where m indicates the slope and c indicates the intercept. According to Eq.  1 , the association between the independent parameters (X) and the dependent parameters (Y) can be inferred [ 18 ].

The advantage of linear regression is that it is straightforward to learn and easy to-eliminate overfitting through regularization. One drawback of linear regression is that it is not convenient when applied to nonlinear relationships. However, it is not recommended for most practical applications as it greatly simplifies real-world problems [ 19 ]. The implementation tools utilized in linear regression are Python, R, MATLAB, and Excel.

As shown in Fig.  4 , observations are highlighted in red, and random deviations' result (shown in green) from the basic relationship (shown in yellow) between the independent variable (x) and the dependent variable (y) [ 20 ].

Linear regression model

B. Logistic regression

Logistic regression, also known as the logistic model, investigates the correlation between many independent variables and a categorical dependent variable and calculates the probability of an event by fitting the data to a logistic curve [ 21 ]. Discrete mean values must be binary, i.e., have only two outcomes: true or false, 0 or 1, yes or no, or either superscript or subscript. In logistic regression, categorical variables need to be predicted and classification problems should be solved. Logistic regression can be implemented using various tools such as R, Python, Java, and MATLAB [ 18 ]. Logistic regression has many benefits; for example, it shows the linear relationship between dependent and independent variables with the best results. It is also simple to understand. On the other hand, it can only predict numerical output, is not relevant to nonlinear data, and is sensitive to outliers [ 22 ].

C. Decision tree

The decision tree (DT) is the supervised learning technique used for classification. It combines the values of attributes based on their order, either ascending or descending [ 23 ]. As a tree-based strategy, DT defines each path starting from the root using a data-separating sequence until a Boolean conclusion is attained at the leaf node [ 24 , 25 ]. DT is a hierarchical representation of knowledge interactions that contains nodes and links. When relations are employed to classify, nodes reflect purposes [ 26 , 27 ]. An example of DT is presented in Fig.  5 .

Example of a DT

DTs have various drawbacks, such as increased complexity with increasing nomenclature, small modifications that may lead to a different architecture, and more processing time to train data [ 18 ]. The implementation tools used in DT are Python (Scikit-Learn), RStudio, Orange, KNIME, and Weka [ 22 ].

D. Random forest

Random forest (RF) is a basic technique that produces correct results most of the time. It may be utilized for classification and regression. The program produces an ensemble of DTs and blends them [ 28 ].

In the RF classifier, the higher the number of trees in the forest, the more accurate the results. So, the RF has generated a collection of DTs called the forest and combined them to achieve more accurate prediction results. In RF, each DT is built only on a part of the given dataset and trained on approximations. The RF brings together several DTs to reach the optimal decision [ 18 ].

As indicated in Fig.  6 , RF randomly selects a subset of features from the data, and from each subset it generates n random trees [ 20 ]. RF will combine the results from all DTs and provide them in the final output.

Random forest architecture

Two parameters are being used for tuning RF models: mtry —the count of randomly selected features to be considered in each division; and ntree —the model trees count. The mtry parameter has a trade-off: Large values raise the correlation between trees, but enhance the per-tree accuracy [ 29 ].

The RF works with a labeled dataset to do predictions and build a model. The final model is utilized to classify unlabeled data. The model integrates the concept of bagging with a random selection of traits to build variance-controlled DTs [ 30 ].

RF offers significant benefits. First, it can be utilized for determining the relevance of the variables in a regression and classification task [ 31 , 32 ]. This relevance is measured on a scale, based on the impurity drop at each node used for data segmentation [ 33 ]. Second, it automates missing values contained in the data and resolves the overfitting problem of DT. Finally, RF can efficiently handle huge datasets. On the other side, RF suffers from drawbacks; for example, it needs more computing and resources to generate the output results and it requires training effort due to the multiple DTs involved in it. The implementation tools used in RF are Python Scikit-Learn and R [ 18 ].

E. Support vector machine

The supervised ML technique for classification issues and regression models is called the support vector machine (SVM). SVM is a linear model that offers solutions to issues that are both linear and nonlinear. as shown in Fig.  7 . Its foundation is the idea of margin calculation. The dataset is divided into several groups to build relations between them [ 18 ].

Support vector machine

SVM is a statistics-based learning method that follows the principle of structural risk minimization and aims to locate decision bounds, also known as hyperplanes, that can optimally separate classes by finding a hyperplane in a usable N-dimensional space that explicitly classifies data points [ 34 , 35 , 36 ]. SVM indicates the decision boundary between two classes by defining the value of each data point, in particular the support vector points placed on the boundary between the respective classes [ 37 ].

SVM has several advantages; for example, it works perfectly with both semi-structured and unstructured data. The kernel trick is a strong point of SVM. Moreover, it can handle any complex problem with the right functionality and can also handle high-dimensional data. Furthermore, SVM generalization has less allocation risk. On the other hand, SVM has many downsides. The model training time is increased on a large dataset. Choosing the right kernel function is also a difficult process. In addition, it is not working well with noisy data. Implementation tools used in SVM include SVMlight with C, LibSVM with Python, MATLAB or Ruby, SAS, Kernlab, Scikit-Learn, and Weka [ 22 ].

F. K-nearest neighbor

K-nearest neighbor (KNN) is an "instance-based learning" or non-generalized learning algorithm, which is often known as a “lazy learning” algorithm [ 38 ]. KNN is used for solving classification problems. To anticipate the target label of the novel test data, KNN determines the distance of the nearest training data class labels with a new test data point in the existence of a K value, as shown in Fig.  8 . It then calculates the number of nearest data points using the K value and terminates the label of the new test data class. To determine the number of nearest-distance training data points, KNN usually sets the value of K according to (1): k  =  n ^(1/2), where n is the size of the dataset [ 22 ].

K-nearest neighbor

KNN has many benefits; for example, it is sufficiently powerful if the size of the training data is large. It is also simple and flexible, with attributes and distance functions. Moreover, it can handle multi-class datasets. KNN has many drawbacks, such as the difficulty of choosing the appropriate K value, it being very tedious to choose the distance function type for a particular dataset, and the computation cost being a little high due to the distance between all the training data points, the implementation tools used in KNN are Python (Scikit-Learn), WEKA, R, KNIME, and Orange [ 22 ].

G. Naive Bayes

Naive Bayes (NB) focuses on the probabilistic model of Bayes' theorem and is simple to set up as the complex recursive parameter estimation is basically none, making it suitable for huge datasets [ 39 ]. NB determines the class membership degree based on a given class designation [ 40 ]. It scans the data once, and thus, classification is easy [ 41 ]. Simply, the NB classifier assumes that there is no relation between the presence of a particular feature in a class and the presence of any other characteristic. It is mainly targeted at the text classification industry [ 42 ].

NB has great benefits such as ease of implementation, can provide a good result even using fewer training data, can manage both continuous and discrete data, and is ideal to solve the prediction of multi-class problems, and the irrelevant feature does not affect the prediction. NB, on the other hand, has the following drawbacks: It assumes that all features are independent which is not always viable in real-world problems, suffers from zero frequency problems, and the prediction of NB is not usually accurate. Implementation tools are WEKA, Python, RStudio, and Mahout [ 22 ].

To summarize the previously discussed models, Table 1 demonstrates the advantages and disadvantages of each model.

Unsupervised learning

Unlike supervised learning, there are no correct answers and no teachers in unsupervised learning [ 42 ]. It follows the concept that a machine can learn to understand complex processes and patterns on its own without external guidance. This approach is particularly useful in cases where experts have no knowledge of what to look for in the data and the data itself do not include the objectives. The machine predicts the outcome based on past experiences and learns to predict the real-valued outcome from the information previously provided, as shown in Fig.  9 .

Workflow of unsupervised learning [ 23 ]

Unsupervised learning is widely used in the processing of multimedia content, as clustering and partitioning of data in the lack of class labels is often a requirement [ 43 ]. Some of the most popular unsupervised learning-based approaches are k-means, principal component analysis (PCA), and apriori algorithm.

The k-means algorithm is the common portioning method [ 44 ] and one of the most popular unsupervised learning algorithms that deal with the well-known clustering problem. The procedure classifies a particular dataset by a certain number of preselected (assuming k -sets) clusters [ 45 ]. The pseudocode of the K-means algorithm is shown in Pseudocode 1.

K means has several benefits such as being more computationally efficient than hierarchical grouping in case of large variables. It provides more compact clusters than hierarchical ones when a small k is used. Also, the ease of implementation and comprehension of assembly results is another benefit. However, K -means have disadvantages such as the difficulty of predicting the value of K . Also, as different starting sections lead to various final combinations, the performance is affected. It is accurate for raw points and local optimization, and there is no single solution for a given K value—so the average of the K value must be run multiple times (20–100 times) and then pick the results with the minimum J [ 19 ].

B. Principal component analysis

In modern data analysis, principal component analysis (PCA) is an essential tool as it provides a guide for extracting the most important information from a dataset, compressing the data size by keeping only those important features without losing much information, and simplifying the description of a dataset [ 46 , 47 ].

PCA is frequently used to reduce data dimensions before applying classification models. Moreover, unsupervised methods, such as dimensionality reduction or clustering algorithms, are commonly used for data visualizations, detection of common trends or behaviors, and decreasing the data quantity to name a few only [ 48 ].

PCA converts the 2D data into 1D data. This is done by changing the set of variables into new variables known as principal components (PC) which are orthogonal [ 23 ]. In PCA, data dimensions are reduced to make calculations faster and easier. To illustrate how PCA works, let us consider an example of 2D data. When these data are plotted on a graph, it will take two axes. Applying PCA, the data turn into 1D. This process is illustrated in Fig.  10 [ 49 ].

Visualization of data before and after applying PCA [ 49 ]

Apriori algorithm is considered an important algorithm, which was first introduced by R. Agrawal and R. Srikant, and published in [ 50 , 51 ].

The principle of the apriori algorithm is to represent the filter generation strategy. It creates a filter element set ( k  + 1) based on the repeated k element groups. Apriori uses an iterative strategy called planar search, where k item sets are employed to explore ( k  + 1) item sets. First, the set of repeating 1 item is produced by scanning the dataset to collect the number for each item and then collecting items that meet the minimum support. The resulting group is called L1. Then L1 is used to find L2, the recursive set of two elements is used to find L3, and so on until no repeated k element groups are found. Finding every Lk needs a full dataset scan. To improve production efficiency at the level-wise of repeated element groups, a key property called the apriori property is used to reduce the search space. Apriori property states that all non-empty subsets of a recursive element group must be iterative. A two-step technique is used to identify groups of common elements: join and prune activities [ 52 ].

Although it is simple, the apriori algorithm suffers from several drawbacks. The main limitation is the costly wasted time to contain many candidates sets with a lot of redundant item sets. It also suffers from low minimum support or large item sets, and multiple rounds of data are needed for data mining which usually results in irrelevant items, in addition to difficulties in discovering individual elements of events [ 53 , 54 ].

To summarize the previously discussed models, Table 2 demonstrates the advantages and disadvantages of each model.

Reinforcement learning

Reinforcement learning (RL) is different from supervised learning and unsupervised learning. It is a goal-oriented learning approach. RL is closely related to an agent (controller) that takes responsibility for the learning process to achieve a goal. The agent chooses actions, and as a result, the environment changes its state and returns rewards. Positive or negative numerical values are used as rewards. An agent's goal is to maximize the rewards accumulated over time. A job is a complete environment specification that identifies how to generate rewards [ 55 ]. Some of the most popular reinforcement learning-based algorithms are the Q-learning algorithm and the Monte Carlo tree search (MCTS).

A. Q-learning

Q-learning is a type of model-free RL. It can be considered an asynchronous dynamic programming approach. It enables agents to learn how to operate optimally in Markovian domains by exploring the effects of actions, without the need to generate domain maps [ 56 ]. It represented an incremental method of dynamic programming that imposed low computing requirements. It works through the successive improvement of the assessment of individual activity quality in particular states [ 57 ].

In information theory, Q-learning is strongly employed, and other related investigations are underway. Recently, Q-learning combined with information theory has been employed in different disciplines such as natural language processing (NLP), pattern recognition, anomaly detection, and image classification [ 58 , 59 , 60 , 60 ]. Moreover, a framework has been created to provide a satisfying response based on the user’s utterance using RL in a voice interaction system [ 61 ]. Furthermore, a high-resolution deep learning-based prediction system for local rainfall has been constructed [ 62 ].

The advantage of developmental Q-learning is that it is possible to identify the reward value effectively on a given multi-agent environment method as agents in ant Q-learning are interacting with each other. The problem with Q-learning is that its output can be stuck in the local minimum as agents just take the shortest path [ 63 ].

B. Monte Carlo tree search

Monte Carlo tree search (MCTS) is an effective technique for solving sequential selection problems. Its strategy is based on a smart tree search that balances exploration and exploitation. MCTS presents random samples in the form of simulations and keeps activity statistics for better educated choices in each future iteration. MCTS is a decision-making algorithm that is employed in searching tree-like huge complex regions. In such trees, each node refers to a state, which is also referred to as problem configuration, while edges represent transitions from one state to another [ 64 ].

The MCTS is related directly to cases that can be represented by a Markov decision process (MDP), which is a type of discrete-time random control process. Some modifications of the MCTS make it possible to apply it to partially observable Markov decision processes (POMDP) [ 65 ]. Recently, MCTS coupled with deep RL became the base of AlphaGo developed by Google DeepMind and documented in [ 66 ]. The basic MCTS method is conceptually simple, as shown in Fig.  11 .

Basic MCTS process

Tree 1 is constructed progressively and unevenly. The tree policy is utilized to get the critical node of the current tree for each iteration of the method. The tree strategy seeks to strike a balance between exploration and exploitation concerns. Then, from the specified node, simulation 2 is run, and the search tree is then updated according to the obtained results. This comprises adding a child node that matches the specified node's activity and updating its ancestor's statistics. During this simulation, movements are performed based on some default policy, which in its simplest case is to make uniform random movements. The benefit of MCTS is that there is no need to evaluate the values of the intermediate state, which significantly minimizes the amount of required knowledge in the field [ 67 ].

To summarize the previously discussed models, Table 3 demonstrates the advantages and disadvantages of each model.

Deep learning

Over the past decades, ML has had a significant impact on our daily lives with examples including efficient computer vision, web search, and recognition of optical characters. In addition, by applying ML approaches, AI at the human level has also been improved [ 68 , 69 , 70 ]. However, when it comes to the mechanisms of human information processing (such as sound and vision), the performance of traditional ML algorithms is far from satisfactory. The idea of deep learning (DL) was formed in the late 20th inspired by the deep hierarchical structures of human voice recognition and production systems. DL breaks have been introduced in 2006 when Hinton built a deep-structured learning architecture called deep belief network (DBN) [ 71 ].

The performance of classifiers using DL has been extensively improved with the increased complexity of data compared to classical learning methods. Figure  12 shows the performance of classic ML algorithms and DL methods [ 72 ]. The performance of typical ML algorithms becomes stable when they reach the training data threshold, but DL improves its performance as the complexity of data increases [ 73 ].

Performance of deep learning concerning the complexity of data

DL (deep ML, or deep-structured learning) is a subset of ML that involves a collection of algorithms attempting to represent high-level abstractions for data through a model that has complicated structures or is otherwise, composed of numerous nonlinear transformations. The most important characteristic of DL is the depth of the network. Another essential aspect of DL is the ability to replace handcrafted features generated by efficient algorithms for unsupervised or semi-supervised feature learning and hierarchical feature extraction [ 74 ].

DL has significantly advanced the latest technologies in a variety of applications, including machine translation, speech, and visual object recognition, NLP, and text automation, using multilayer artificial neural networks (ANNs) [ 15 ].

Different DL designs in the past two decades give enormous potential for employment in various sectors such as automatic voice recognition, computer vision, NLP, and bioinformatics. This section discusses the most common architectures of DL such as convolutional neural networks (CNNs), long short-term memory (LSTM), and recurrent convolution neural networks (RCNNs) [ 75 ].

A. Convolutional neural network

CNNs are special types of neural networks inspired by the human visual cortex and used in computer vision. It is an automatic feed-forward neural network in which information transfers exclusively in the forward direction [ 76 ]. CNN is frequently applied in face recognition, human organ localization, text analysis, and biological image recognition [ 77 ].

Since CNN was first created in 1989, it has done well in disease diagnosis over the past three decades [ 78 ]. Figure  13 depicts the general architecture of a CNN composed of feature extractors and a classifier. Each layer of the network accepts the output of the previous layer as input and passes it on to the next layer in feature extraction layers. A typical CNN architecture consists of three types of layers: convolution, pooling, and classification. There are two types of layers at the network's low and middle levels: convolutional layers and pooling layers. Even-numbered layers are used for convolutions, while odd-numbered layers are used for pooling operations. The convolution and pooling layers' output nodes are categorized in a two-dimensional plane called feature mapping. Each layer level is typically generated by combining one or more previous layers [ 79 ].

Architecture of CNN [ 79 ]

CNN has a lot of benefits, including a human optical processing system, greatly improved 2D and 3D image processing structure, and is effective in learning and extracting abstract information from 2D information. The max-pooling layer in CNN is efficient in absorbing shape anisotropy. Furthermore, they are constructed from sparse connections with paired weights and contain far fewer parameters than a fully connected network of equal size. CNNs are trained using a gradient-based learning algorithm and are less susceptible to the diminishing gradient problem because the gradient-based approach trains the entire network to directly reduce the error criterion, allowing CNNs to provide highly optimized weights [ 79 ].

B. Long short-term memory

LSTM is a special type of recurrent neural network (RNN) with internal memory and multiplicative gates. Since the original LSTM introduction in 1997 by Sepp Hochrieiter and Jürgen Schmidhuber, a variety of LSTM cell configurations have been described [ 80 ].

LSTM has contributed to the development of well-known software such as Alexa, Siri, Cortana, Google Translate, and Google voice assistant [ 81 ]. LSTM is an implementation of RNN with a special connection between nodes. The special components within the LSTM unit include the input, output, and forget gates. Figure  14 depicts a single LSTM cell.

LSTM unit [ 82 ]

x t  = Input vector at the time t.

h t-1  = Previous hidden state.

c t-1  = Previous memory state.

h t  = Current hidden state.

c t  = Current memory state.

[ x ] = Multiplication operation.

[+] = Addition operation.

LSTM is an RNN module that handles gradient loss problems. In general, RNN uses LSTM to eliminate propagation errors. This allows the RNN to learn over multiple time steps. LSTM is characterized by cells that hold information outside the recurring network. This cell enables the RNN to learn over many time steps. The basic principle of LSTMs is the state of the cell, which contains information outside the recurrent network. A cell is like a memory in a computer, which decides when data should be stored, written, read, or erased via the LSTM gateway [ 82 ]. Many network architectures use LSTM such as bidirectional LSTM, hierarchical and attention-based LSTM, convolutional LSTM, autoencoder LSTM, grid LSTM, cross-modal, and associative LSTM [ 83 ].

Bidirectional LSTM networks move the state vector forward and backward in both directions. This implies that dependencies must be considered in both temporal directions. As a result of inverse state propagation, the expected future correlations can be included in the network's current output [ 84 ]. Bidirectional LSTM investigates and analyzes this because it encapsulates spatially and temporally scattered information and can tolerate incomplete inputs via a flexible cell state vector propagation communication mechanism. Based on the detected gaps in data, this filtering mechanism reidentifies the connections between cells for each data sequence. Figure  15 depicts the architecture. A bidirectional network is used in this study to process properties from multiple dimensions into a parallel and integrated architecture [ 83 ].

(left) Bidirectional LSTM and (right) filter mechanism for processing incomplete data [ 84 ]

Hierarchical LSTM networks solve multi-dimensional problems by breaking them down into subproblems and organizing them in a hierarchical structure. This has the advantage of focusing on a single or multiple subproblems. This is accomplished by adjusting the weights within the network to generate a certain level of interest [ 83 ]. A weighting-based attention mechanism that analyzes and filters input sequences is also used in hierarchical LSTM networks for long-term dependency prediction [ 85 ].

Convolutional LSTM reduces and filters input data collected over a longer period using convolutional operations applied in LSTM networks or the LSTM cell architecture directly. Furthermore, due to their distinct characteristics, convolutional LSTM networks are useful for modeling many quantities such as spatially and temporally distributed relationships. However, many quantities can be expected collectively in terms of reduced feature representation. Decoding or decoherence layers are required to predict different output quantities not as features but based on their parent units [ 83 ].

The LSTM autoencoder solves the problem of predicting high-dimensional parameters by shrinking and expanding the network [ 86 ]. The autoencoder architecture is separately trained with the aim of accurate reconstruction of the input data as reported in [ 87 ]. Only the encoder is used during testing and commissioning to extract the low-dimensional properties that are transmitted to the LSTM. The LSTM was extended to multimodal prediction using this strategy. To compress the input data and cell states, the encoder and decoder are directly integrated into the LSTM cell architecture. This combined reduction improves the flow of information in the cell and results in an improved cell state update mechanism for both short-term and long-term dependency [ 83 ].

Grid long short-term memory is a network of LSTM cells organized into a multi-dimensional grid that can be applied to sequences, vectors, or higher-dimensional data like images [ 88 ]. Grid LSTM has connections to the spatial or temporal dimensions of input sequences. Thus, connections of different dimensions within cells extend the normal flow of information. As a result, grid LSTM is appropriate for the parallel prediction of several output quantities that may be independent, linear, or nonlinear. The network's dimensions and structure are influenced by the nature of the input data and the goal of the prediction [ 89 ].

A novel method for the collaborative prediction of numerous quantities is the cross-modal and associative LSTM. It uses several standard LSTMs to separately model different quantities. To calculate the dependencies of the quantities, these LSTM streams communicate with one another via recursive connections. The chosen layers' outputs are added as new inputs to the layers before and after them in other streams. Consequently, a multimodal forecast can be made. The benefit of this approach is that the correlation vectors that are produced have the same dimensions as the input vectors. As a result, neither the parameter space nor the computation time increases [ 90 ].

C. Recurrent convolution neural network

CNN is a key method for handling various computer vision challenges. In recent years, a new generation of CNNs has been developed, the recurrent convolution neural network (RCNN), which is inspired by large-scale recurrent connections in the visual systems of animals. The recurrent convolutional layer (RCL) is the main feature of RCNN, which integrates repetitive connections among neurons in the normal convolutional layer. With the increase in the number of repetitive computations, the receptive domains (RFs) of neurons in the RCL expand infinitely, which is contrary to biological facts [ 91 ].

The RCNN prototype was proposed by Ming Liang and Xiaolin Hu [ 92 , 93 ], and the structure is illustrated in Fig.  16 , in which both forward and redundant connections have local connectivity and weights shared between distinct sites. This design is quite like the recurrent multilayer perceptron (RMLP) concept which is often used for dynamic control [ 94 , 95 ] (Fig.  17 , middle). Like the distinction between MLP and CNN, the primary distinction is that in RMLP, common local connections are used in place of full connections. For this reason, the proposed model is known as RCNN [ 96 ].

Illustration of the architectures of CNN, RMLP, and RCNN [ 85 ]

Illustration of the total number of reviewed papers

The main unit of RCNN is the RCL. RCLs develop through discrete time steps. RCNN offers three basic advantages. First, it allows each unit to accommodate background information in an arbitrarily wide area in the current layer. Second, recursive connections improve the depth of the network while keeping the number of mutable parameters constant through weight sharing. This is consistent with the trend of modern CNN architecture to grow deeper with a relatively limited number of parameters. The third aspect of RCNN is the time exposed in RCNN which is a CNN with many paths between the input layer and the output layer, which makes learning simple. On one hand, having longer paths makes it possible for the model to learn very complex features. On the other hand, having shorter paths may improve the inverse gradient during training [ 91 ].

To summarize the previously discussed models, Table 4 demonstrates the advantages and disadvantages of each model.

Disease prediction with analytics

The studies discussed in this paper have been presented and published in high-quality journals and international conferences published by IEEE, Springer, and Elsevier, and other major scientific publishers such as Hindawi, Frontiers, Taylor, and MDPI. The search engines used are Google Scholar, Scopus, and Science Direct. All papers selected covered the period from 2019 to 2022. Machine learning, deep learning, health care, surgery, cardiology, radiology, hepatology, and nephrology are some of the terms used to search for these studies. The studies chosen for this survey are concerned with the use of machine learning as well as deep learning algorithms in healthcare prediction. For this survey, empirical and review articles on the topics were considered. This section discusses existing research efforts that healthcare prediction using various techniques in ML and DL. This survey gives a detailed discussion about the methods and algorithms which are used for predictions, performance metrics, and tools of their model.

ML-based healthcare prediction

To predict diabetes patients, the authors of [ 97 ] utilized a framework to develop and evaluate ML classification models like logistic regression, KNN, SVM, and RF. ML method was implemented on the Pima Indian Diabetes Database (PIDD) which has 768 rows and 9 columns. The forecast accuracy delivers 83%. Results of the implementation approach indicate how the logistic regression outperformed other algorithms of ML, in addition only a structured dataset was selected but unstructured data are not considered, also model should be implemented in other healthcare domains like heart disease, and COVID-19, finally other factors should be considered for diabetes prediction, like family history of diabetes, smoking habits, and physical inactivity.

The authors created a diagnosis system in [ 98 ] that uses two different datasets (Frankfurt Hospital in Germany and PIDD provided by the UCI ML repository) and four prediction models (RF, SVM, NB, and DT) to predict diabetes. the SVM algorithm performed with an accuracy of 83.1 percent. There are some aspects of this study that need to be improved; such as, using a DL approach to predict diabetes may lead to achieving better results; furthermore, the model should be tested in other healthcare domains such as heart disease and COVID-19 prediction datasets.

In [ 99 ], the authors proposed three ML methods (logistic regression, DT, and boosted RF) to assess COVID-19 using OpenData Resources from Mexico and Brazil. To predict rescue and death, the proposed model incorporates just the COVID-19 patient's geographical, social, and economic conditions, as well as clinical risk factors, medical reports, and demographic data. On the dataset utilized, the model for Mexico has a 93 percent accuracy, and an F1 score is 0.79. On the other hand, on the used dataset, the Brazil model has a 69 percent accuracy and an F1 score is 0.75. The three ML algorithms have been examined and the acquired results showed that logistic regression is the best way of processing data. The authors should be concerned about the usage of authentication and privacy management of the created data.

A new model for predicting type 2 diabetes using a network approach and ML techniques was presented by the authors in [ 100 ] (logistic regression, SVM, NB, KNN, decision tree, RF, XGBoost, and ANN). To predict the risk of type 2 diabetes, the healthcare data of 1,028 type 2 diabetes patients and 1,028 non-type 2 diabetes patients were extracted from de-identified data. The experimental findings reveal the models’ effectiveness with an area under curve (AUC) varied from 0.79 to 0.91. The RF model achieved higher accuracy than others. This study relies only on the dataset providing hospital admission and discharge summaries from one insurance company. External hospital visits and information from other insurance companies are missing for people with many insurance providers.

The authors of [ 101 ] proposed a healthcare management system that can be used by patients to schedule appointments with doctors and verify prescriptions. It gives support for ML to detect ailments and determine medicines. ML models including DT, RF, logistic regression, and NB classifiers are applied to the datasets of diabetes, heart disease, chronic kidney disease, and liver. The results showed that among all the other models, logistic regression had the highest accuracy of 98.5 percent in the heart dataset. while the least accuracy is of the DT classifier which came out to be 92 percent. In the liver dataset the logistic regression with maximum accuracy of 75.17% among all others. In the chronic renal disease dataset, the logistic regression, RF, and Gaussian NB, all performed well with an accuracy of 1, the accuracy of 100% should be verified by using k-fold cross-validation to test the reliability of the models. In the diabetes dataset random forest with maximum accuracy of 83.67 percent. The authors should include a hospital directory as then various hospitals and clinics can be accessed through a single portal. Additionally, image datasets could be included to allow image processing of reports and the deployment of DL to detect diseases.

In [ 102 ], the authors developed an ML model to predict the occurrence of Type 2 Diabetes in the following year (Y + 1) using factors in the present year (Y). Between 2013 and 2018, the dataset was obtained as an electronic health record from a private medical institute. The authors applied logistic regression, RF, SVM, XGBoost, and ensemble ML algorithms to predict the outcome of non-diabetic, prediabetes, and diabetes. Feature selection was applied to choose the three classes efficiently. FPG, HbA1c, triglycerides, BMI, gamma-GTP, gender, age, uric acid, smoking, drinking, physical activity, and family history were among the features selected. According to the experimental results, the maximum accuracy was 73% from RF, while the lowest was 71% from the logistic regression model. The authors presented a model that used only one dataset. As a result, additional data sources should be applied to verify the models developed in this study.

The authors of [ 103 ] classified the diabetes dataset using SVM and NB algorithms with feature selection to improve the model's accuracy. PIDD is taken from the UCI Repository for analysis. For training and testing purposes the authors employed the k-fold cross-validation model, the SVM classifier was performing better than the NB method it offers around 91% correct predictions; however, the authors acknowledge that they need to extend to the latest dataset that will contain additional attributes and rows.

K-means clustering is an unsupervised ML algorithm that was introduced by the authors of [ 104 ] for the purpose of detecting heart disease in its earliest stages using the UCI heart disease dataset. PCA is used for dimensionality reduction. The outcome of the method demonstrates early cardiac disease prediction with 94.06% accuracy. The authors should apply the proposed technique using more than one algorithm and use more than one dataset.

In [ 105 ], the authors constructed a predictive model for the classification of diabetes data using the logistic regression classification technique. The dataset includes 459 patients for training data and 128 cases for testing data. The prediction accuracy using logistic regression was obtained at 92%. The main limitation of this research is that the authors have not compared the model with other diabetes prediction algorithms, so it cannot be confirmed.

The authors of [ 106 ] developed a prediction model that analyzes the user's symptoms and predicts the disease using ML algorithms (DT classifier, RF classifier, and NB classifier). The purpose of this study was to solve health-related problems by allowing medical professionals to predict diseases at an early stage. The dataset is a sample of 4920 patient records with 41 illnesses diagnosed. A total of 41 disorders were included as a dependent variable. All algorithms achieved the same accuracy score of 95.12%. The authors noticed that overfitting occurred when all 132 symptoms from the original dataset were assessed instead of 95 symptoms. That is, the tree appears to remember the dataset provided and thus fails to classify new data. As a result, just 95 symptoms were assessed during the data-cleansing process, with the best ones being chosen.

In [ 107 ], the authors built a decision-making system that assists practitioners to anticipate cardiac problems in exact classification through a simpler method and will deliver automated predictions about the condition of the patient’s heart. implemented 4 algorithms (KNN, RF, DT, and NB), all these algorithms were used in the Cleveland Heart Disease dataset. The accuracy varies for different classification methods. The maximum accuracy is given when they utilized the KNN algorithm with the Correlation factor which is almost 94 percent. The authors should extend the presented technique to leverage more than one dataset and forecast different diseases.

The authors of [ 108 ] used the Cleveland dataset, which included 303 cases and 76 attributes, to test out three different classification strategies: NB, SVM, and DT in addition to KNN. Only 14 of these 76 characteristics are going to be put through the testing process. The authors performed data preprocessing to remove noisy data. The KNN obtained the greatest accuracy with 90.79 percent. The authors need to use more sophisticated models to improve the accuracy of early heart disease prediction.

The authors of [ 109 ] proposed a model to predict heart disease by making use of a cardiovascular dataset, which was then classified through the application of supervised machine learning algorithms (DT, NB, logistic regression, RF, SVM, and KNN). The results reveal that the DT classification model predicted cardiovascular disorders better than other algorithms with an accuracy of 73 percent. The authors highlighted that the ensemble ML techniques employing the CVD dataset can generate a better illness prediction model.

In [ 110 ], the authors attempted to increase the accuracy of heart disease prediction by applying a logistic regression using a healthcare dataset to determine whether patients have heart illness problems or not. The dataset was acquired from an ongoing cardiovascular study on people of the town of Framingham, Massachusetts. The model reached an accuracy prediction of 87 percent. The authors acknowledge the model could be improved with more data and the use of more ML models.

Because breast cancer affects one in every 28 women in India, the author of [ 111 ] presented an accurate classification technique to examine the breast cancer dataset containing 569 rows and 32 columns. Similarly employing a heart disease dataset and Lung cancer dataset, this research offered A novel way to function selection. This method of selection is based on genetic algorithms mixed with the SVM classification. The classifier results are Lung cancer 81.8182, Diabetes 78.9272. noticed that the size, kind, and source of data used are not indicated.

In [ 112 ], the authors predicted the risk factors that cause heart disease using the K-means clustering algorithm and analyzed with a visualization tool using a Cleveland heart disease dataset with 76 features of 303 patients, holds 209 records with 8 attributes such as age, chest pain type, blood pressure, blood glucose level, ECG in rest, heart rate as well as four types of chest pain. The authors forecast cardiac diseases by taking into consideration the primary characteristics of four types of chest discomfort solely and K-means clustering is a common unsupervised ML technique.

The aim of the article [ 113 ] was to report the advantages of using a variety of data mining (DM) methods and validated heart disease survival prediction models. From the observations, the authors proposed that logistic regression and NB achieved the highest accuracy when performed on a high-dimensional dataset on the Cleveland hospital dataset and DT and RF produce better results on low-dimensional datasets. RF delivers more accuracy than the DT classifier as the algorithm is an optimized learning algorithm. The author mentioned that this work can be extended to other ML algorithms, the model could be developed in a distributed environment such as Map–Reduce, Apache Mahout, and HBase.

In [ 114 ], the authors proposed a single algorithm named hybridization to predict heart disease that combines used techniques into one single algorithm. The presented method has three phases. Preprocessing phase, classification phase, and diagnosis phase. They employed the Cleveland database and algorithms NB, SVM, KNN, NN, J4.8, RF, and GA. NB and SVM always perform better than others, whereas others depend on the specified features. results attained an accuracy of 89.2 percent. The authors need to is the key goal. Notice that the dataset is little; hence, the system was not able to train adequately, so the accuracy of the method was bad.

Using six algorithms (logistic regression, KNN, DT, SVM, NB, and RF), the authors of [ 115 ] explored different data representations to better understand how to use clinical data for predicting liver disease. The original dataset was taken from the northeast of Andhra Pradesh, India. includes 583 liver patient data, whereas 75.64 percent are male, and 24.36 percent are female. The analysis result indicated that the logistic regression classifier delivers the most increased order exactness of 75 percent depending on the f1 measure to forecast the liver illness and NB gives the least precision of 53 percent. The authors merely studied a few prominent supervised ML algorithms; more algorithms can be picked to create an increasingly exact model of liver disease prediction and performance can be steadily improved.

In [ 116 ], the authors aimed to predict coronary heart disease (CHD) based on historical medical data using ML technology. The goal of this study is to use three supervised learning approaches, NB, SVM, and DT, to find correlations in CHD data that could aid improve prediction rates. The dataset contains a retrospective sample of males from KEEL, a high-risk heart disease location in the Western Cape of South Africa. The model utilized NB, SVM, and DT. NB achieved the most accurate among the three models. SVM and DT J48 outperformed NB with a specificity rate of 82 percent but showed an inadequate sensitivity rate of less than 50 percent.

With the help of DM and network analysis methods, the authors of [ 117 ] created a chronic disease risk prediction framework that was created and evaluated in the Australian healthcare system to predict type 2 diabetes risk. Using a private healthcare funds dataset from Australia that spans six years and three different predictive algorithms (regression, parameter optimization, and DT). The accuracy of the prediction ranges from 82 to 87 percent. The hospital admission and discharge summary are the dataset's source. As a result, it does not provide information about general physician visits or future diagnoses.

DL-based healthcare prediction

With the help of DL algorithms such as CNN for autofeature extraction and illness prediction, the authors of [ 118 ] proposed a system for predicting patients with the more common inveterate diseases, and they used KNN for distance calculation to locate the exact matching in the dataset and the outcome of the final sickness prediction. A combination of disease symptoms was made for the structure of the dataset, the living habits of a person, and the specific attaches to doctor consultations which are acceptable in this general disease prediction. In this study, the Indian chronic kidney disease dataset was utilized that comprises 400 occurrences, 24 characteristics, and 2 classes were restored from the UCI ML store. Finally, a comparative study of the proposed system with other algorithms such as NB, DT, and logistic regression has been demonstrated in this study. The findings showed that the proposed system gives an accuracy of 95% which is higher than the other two methods. So, the proposed technique should be applied using more than one dataset.

In [ 119 ], the authors developed a DL approach that uses chest radiography images to differentiate between patients with mild, pneumonia, and COVID-19 infections, providing a valid mechanism for COVID-19 diagnosis. To increase the intensity of the chest X-ray image and eliminate noise, image-enhancing techniques were used in the proposed system. Two distinct DL approaches based on a pertained neural network model (ResNet-50) for COVID-19 identification utilizing chest X-ray (CXR) pictures are proposed in this work to minimize overfitting and increase the overall capabilities of the suggested DL systems. The authors emphasized that tests using a vast and hard dataset encompassing several COVID-19 cases are necessary to establish the efficacy of the suggested system.

Diabetes disease prediction was the topic of the article [ 120 ], in which the authors presented a cuckoo search-based deep LSTM classifier for prediction. The deep convLSTM classifier is used in cuckoo search optimization, which is a nature-inspired method for accurately predicting disease by transferring information and therefore reducing time consumption. The PIMA dataset is used to predict the onset of diabetes. The National Institute of Diabetes and Digestive and Kidney Diseases provided the data. The dataset is made up of independent variables including insulin level, age, and BMI index, as well as one dependent variable. The new technique was compared to traditional methods, and the results showed that the proposed method achieved 97.591 percent accuracy, 95.874 percent sensitivity, and 97.094 percent specificity, respectively. The authors noticed more datasets are needed, as well as new approaches to improve the classifier's effectiveness.

In [ 121 ], the authors presented a wavelet-based convolutional neural network to handle data limitations in this time of COVID-19 fast emergence. By investigating the influence of discrete wavelet transform decomposition up to 4 levels, the model demonstrated the capability of multi-resolution analysis for detecting COVID-19 chest X-rays. The wavelet sub-bands are CNN’s inputs at each decomposition level. COVID-19 chest X-ray-12 is a collection of 1,944 chest X-ray pictures divided into 12 groups that were compiled from two open-source datasets (National Institute Health containing several X-rays of pneumonia-related diseases, whereas the COVID-19 dataset is collected from Radiology Society North America). COVID-Neuro wavelet, a suggested model, was trained alongside other well-known ImageNet pre-trained models on COVID-CXR-12. The authors acknowledge they hope to investigate the effects of other wavelet functions besides the Haar wavelet.

A CNN framework for COVID-19 identification was suggested in [ 122 ] it made use of computed tomography images that was developed by the authors. The proposed framework employs a public CT dataset of 2482 CT images from patients of both classifications. the system attained an accuracy of 96.16 percent and recall of 95.41 percent after training using only 20 percent of the dataset. The authors stated that the use of the framework should be extended to multimodal medical pictures in the future.

Using an LSTM network enhanced by two processes to perform multi-label classification based on patients' clinical visit records, the authors of [ 123 ] performed multi-disease prediction for intelligent clinical decision support. A massive dataset of electronic health records was collected from a prominent hospital in southeast China. The suggested LSTM approach outperforms several standard and DL models in predicting future disease diagnoses, according to model evaluation results. The F1 score rises from 78.9 to 86.4 percent, respectively, with the state-of-the-art conventional and DL models, to 88.0 percent with the suggested technique. The authors stated that the model prediction performance may be enhanced further by including new input variables and that to reduce computational complexity, the method only uses one data source.

In [ 124 ], the authors introduced an approach to creating a supervised ANN structure based on the subnets (the group of neurons) instead of layers, in the cases of low datasets, this effectively predicted the disease. The model was evaluated using textual data and compared to multilayer perceptrons (MLPs) as well as LSTM recurrent neural network models using three small-scale publicly accessible benchmark datasets. On the Iris dataset, the experimental findings for classification reached 97% accuracy, compared to 92% for RNN (LSTM) with three layers, and the model had a lower error rate, 81, than RNN (LSTM) and MLP on the diabetic dataset, while RNN (LSTM) has a high error rate of 84. For larger datasets, however, this method is useless. This model is useless because it has not been implemented on large textual and image datasets.

The authors of [ 125 ] presented a novel AI and Internet of Things (IoT) convergence-based disease detection model for a smart healthcare system. Data collection, reprocessing, categorization, and parameter optimization are all stages of the proposed model. IoT devices, such as wearables and sensors, collect data, which AI algorithms then use to diagnose diseases. The forest technique is then used to remove any outliers found in the patient data. Healthcare data were used to assess the performance of the CSO-LSTM model. During the study, the CSO-LSTM model had a maximum accuracy of 96.16% on heart disease diagnoses and 97.26% on diabetes diagnoses. This method offered a greater prediction accuracy for heart disease and diabetes diagnosis, but there was no feature selection mechanism; hence, it requires extensive computations.

The global health crisis posed by coronaviruses was a subject of [ 126 ]. The authors aimed at detecting disease in people whose X-ray had been selected as potential COVID-19 candidates. Chest X-rays of people with COVID-19, viral pneumonia, and healthy people are included in the dataset. The study compared the performance of two DL algorithms, namely CNN and RNN. DL techniques were used to evaluate a total of 657 chest X-ray images for the diagnosis of COVID-19. VGG19 is the most successful model, with a 95% accuracy rate. The VGG19 model successfully categorizes COVID-19 patients, healthy individuals, and viral pneumonia cases. The dataset's most failing approach is InceptionV3. The success percentage can be improved, according to the authors, by improving data collection. In addition to chest radiography, lung tomography can be used. The success ratio and performance can be enhanced by creating numerous DL models.

In [ 127 ], the authors developed a method based on the RNN algorithm for predicting blood glucose levels for diabetics a maximum of one hour in the future, which required the patient's glucose level history. The Ohio T1DM dataset for blood glucose level prediction, which included blood glucose level values for six people with type 1 diabetes, was used to train and assess the approach. The distribution features were further honed with the use of studies that revealed the procedure's certainty estimate nature. The authors point out that they can only evaluate prediction goals with enough glucose level history; thus, they cannot anticipate the beginning levels after a gap, which does not improve the prediction's quality.

To build a new deep anomaly detection model for fast, reliable screening, the authors of [ 128 ] used an 18-layer residual CNN pre-trained on ImageNet with a different anomaly detection mechanism for the classification of COVID-19. On the X-ray dataset, which contains 100 images from 70 COVID-19 persons and 1431 images from 1008 non-COVID-19 pneumonia subjects, the model obtains a sensitivity of 90.00 percent specificity of 87.84 percent or sensitivity of 96.00 percent specificity of 70.65 percent. The authors noted that the model still has certain flaws, such as missing 4% of COVID-19 cases and having a 30% false positive rate. In addition, more clinical data are required to confirm and improve the model's usefulness.

In [ 129 ], the authors developed COVIDX-Net, a novel DL framework that allows radiologists to diagnose COVID-19 in X-ray images automatically. Seven algorithms (MobileNetV2, ResNetV2, VGG19, DenseNet201, InceptionV3, Inception, and Xception) were evaluated using a small dataset of 50 photographs (MobileNetV2, ResNetV2, VGG19, DenseNet201, InceptionV3, Inception, and Xception). Each deep neural network model can classify the patient's status as a negative or positive COVID-19 case based on the normalized intensities of the X-ray image. The f1-scores for the VGG19 and dense convolutional network (DenseNet) models were 0.89 and 0.91, respectively. With f1-scores of 0.67, the InceptionV3 model has the weakest classification performance.

The authors of [ 130 ] designed a DL approach for delivering 30-min predictions about future glucose levels based on a Dilated RNN (DRNN). The performance of the DRNN models was evaluated using data from two electronic health records datasets: OhioT1DM from clinical trials and the in silicon dataset from the UVA-Padova simulator. It outperformed established glucose prediction approaches such as neural networks (NNs), support vector regression (SVR), and autoregressive models (ARX). The results demonstrated that it significantly improved glucose prediction performance, although there are still some limits, such as the authors' creation of a data-driven model that heavily relies on past EHR. The quality of the data has a significant impact on the accuracy of the prediction. The number of clinical datasets is limited and , however, often restricted. Because certain data fields are manually entered, they are occasionally incorrect.

In [ 131 ], the authors utilized a deep neural network (DNN) to discover 15,099 stroke patients, researchers were able to predict stroke death based on medical history and human behaviors utilizing large-scale electronic health information. The Korea Centers for Disease Control and Prevention collected data from 2013 to 2016 and found that there are around 150 hospitals in the country, all having more than 100 beds. Gender, age, type of insurance, mode of admission, necessary brain surgery, area, length of hospital stays, hospital location, number of hospital beds, stroke kind, and CCI were among the 11 variables in the DL model. To automatically create features from the data and identify risk factors for stroke, researchers used a DNN/scaled principal component analysis (PCA). 15,099 people with a history of stroke were enrolled in the study. The data were divided into a training set (66%) and a testing set (34%), with 30 percent of the samples used for validation in the training set. DNN is used to examine the variables of interest, while scaled PCA is utilized to improve the DNN's continuous inputs. The sensitivity, specificity, and AUC values were 64.32%, 85.56%, and 83.48%, respectively.

The authors of [ 132 ] proposed (GluNet), an approach to glucose forecasting. This method made use of a personalized DNN to forecast the probabilistic distribution of short-term measurements for people with Type 1 diabetes based on their historical data. These data included insulin doses, meal information, glucose measurements, and a variety of other factors. It utilized the newest DL techniques consisting of four components: post-processing, dilated CNN, label recovery/ transform, and data preprocessing. The authors run the models on the subjects from the OhioT1DM datasets. The outcomes revealed significant enhancements over the previous procedures via a comprehensive comparison concerning the and root mean square error (RMSE) having a time lag of 60 min prediction horizons (PH) and RMSE having a small-time lag for the case of prediction horizons in the virtual adult participants. If the PH is properly matched to the lag between input and output, the user may learn the control of the system more frequently and it achieves good performance. Additionally, GluNet was validated on two clinical datasets. It attained an RMSE with a time lag of 60 min PH and RMSE with a time lag of 30-min PH. The authors point out that the model does not consider physiological knowledge, and that they need to test GluNet with larger prediction horizons and use it to predict overnight hypoglycemia.

The authors of [ 133 ] proposed the short-term blood glucose prediction model (VMD-IPSO-LSTM), which is a short-term strategy for predicting blood glucose (VMD-IPSO-LSTM). Initially, the intrinsic modal functions (IMF) in various frequency bands were obtained using the variational modal decomposition (VMD) technique, which deconstructed the blood glucose content. The short- and long-term memory networks then constructed a prediction mechanism for each blood glucose component’s intrinsic modal functions (IMF). Because the time window length, learning rate, and neuron count are difficult to set, the upgraded PSO approach optimized these parameters. The improved LSTM network anticipated each IMF, and the projected subsequence was superimposed in the final step to arrive at the ultimate prediction result. The data of 56 participants were chosen as experimental data among 451 diabetic Mellitus patients. The experiments revealed that it improved prediction accuracy at "30 min, 45 min, and 60 min." The RMSE and MAPE were lower than the "VMD-PSO-LSTM, VMD-LSTM, and LSTM," indicating that the suggested model is effective. The longer time it took to anticipate blood glucose levels and the higher accuracy of the predictions gave patients and doctors more time to improve the effectiveness of diabetes therapy and manage blood glucose levels. The authors noted that they still faced challenges, such as an increase in calculation volume and operation time. The time it takes to estimate glucose levels in the short term will be reduced.

To speed up diagnosis and cut down on mistakes, the authors of [ 134 ] proposed a new paradigm for primary COVID-19 detection based on a radiology review of chest radiography or chest X-ray. The authors used a dataset of chest X-rays from verified COVID-19 patients (408 photographs), confirmed pneumonia patients (4273 images), and healthy people (1590 images) to perform a three-class image classification (1590 images). There are 6271 people in total in the dataset. To fulfill this image categorization problem, the authors plan to use CNN and transfer learning. For all the folds of data, the model's accuracy ranged from 93.90 percent to 98.37 percent. Even the lowest level of accuracy, 93.90 percent, is still quite good. The authors will face a restriction, particularly when it comes to adopting such a model on a large scale for practical usage.

In [ 135 ], the authors proposed DL models for predicting the number of COVID-19-positive cases in Indian states. The Ministry of Health and Family Welfare dataset contains time series data for 32 individual confirmed COVID-19 cases in each of the states (28) and union territories (4) since March 14, 2020. This dataset was used to conduct an exploratory analysis of the increase in the number of positive cases in India. As prediction models, RNN-based LSTMs are used. Deep LSTM, convolutional LSTM, and bidirectional LSTM models were tested on 32 states/union territories, and the model with the best accuracy was chosen based on absolute error. Bidirectional LSTM produced the best performance in terms of prediction errors, while convolutional LSTM produced the worst performance. For all states, daily and weekly forecasts were calculated, and bi-LSTM produced accurate results (error less than 3%) for short-term prediction (1–3 days).

With the goal of increasing the reliability and precision of type 1 diabetes predictions, the authors of [ 136 ] proposed a new method based on CNNs and DL. It was about figuring out how to extract the behavioral pattern. Numerous observations of identical behaviors were used to fill in the gaps in the data. The suggested model was trained and verified using data from 759 people with type 1 diabetes who visited Sheffield Teaching Hospitals between 2013 and 2015. A subject's type 1 diabetes test, demographic data (age, gender, years with diabetes), and the final 84 days (12 weeks) of self-monitored blood glucose (SMBG) measurements preceding the test formed each item in the training set. In the presence of insufficient data and certain physiological specificities, prediction accuracy deteriorates, according to the authors.

The authors of [ 137 ] constructed a framework using the PIDD. PID's participants are all female and at least 21 years old. PID comprises 768 incidences, with 268 samples diagnosed as diabetic and 500 samples not diagnosed as diabetic. The eight most important characteristics that led to diabetes prediction. The accuracy of functional classifiers such as ANN, NB, DT, and DL is between 90 and 98 percent. On the PIMA dataset, DL had the best results for diabetes onset among the four, with an accuracy rate of 98.07 percent. The technique uses a variety of classifiers to accurately predict the disease, but it failed to diagnose it at an early stage.

To summarize all previous works discussed in this section, we will categorize them according to the diseases along with the techniques used to predict each disease, the datasets used, and the main findings, as shown in Table 5 .

Results and discussion

This study conducted a systematic review to examine the latest developments in ML and DL for healthcare prediction. It focused on healthcare forecasting and how the use of ML and DL can be relevant and robust. A total of 41 papers were reviewed, 21 in ML and 20 in DL as depicted in Fig.  17 .

In this study, the reviewed paper were classified by diseases predicted; as a result, 5 diseases were discussed including diabetes, COVID-19, heart, liver, and chronic kidney). Table 6 illustrates the number of reviewed papers for each disease in addition to the adopted prediction techniques in each disease.

Table 6 provides a comprehensive summary of the various ML and DL models used for disease prediction. It indicates the number of studies conducted on each disease, the techniques employed, and the highest level of accuracy attained. As shown in Table 6 , the optimal diagnostic accuracy for each disease varies. For diabetes, the DL model achieved a 98.07% accuracy rate. For COVID-19, the accuracy of the logistic regression model was 98.5%. The CSO-LSTM model achieved an accuracy of 96.16 percent for heart disease. For liver disease, the accuracy of the logistic regression model was 75%. The accuracy of the logistic regression model for predicting multiple diseases was 98.5%. It is essential to note that these are merely the best accuracy included in this survey. In addition, it is essential to consider the size and quality of the datasets used to train and validate the models. It is more likely that models trained on larger and more diverse datasets will generalize well to new data. Overall, the results presented in Table 6 indicate that ML and DL models can be used to accurately predict disease. When selecting a model for a specific disease, it is essential to carefully consider the various models and techniques.

Although ML and DL have made incredible strides in recent years, they still have a long way to go before they can effectively be used to solve the fundamental problems plaguing the healthcare systems. Some of the challenges associated with implementing ML and DL approaches in healthcare prediction are discussed here.

The Biomedical Data Stream is the primary challenge that needs to be handled. Significant amounts of new medical data are being generated rapidly, and the healthcare industry as a whole is evolving rapidly. Some examples of such real-time biological signals include measurements of blood pressure, oxygen saturation, and glucose levels. While some variants of DL architecture have attempted to address this problem, many challenges remain before effective analyses of rapidly evolving, massive amounts of streaming data can be conducted. These include problems with memory consumption, feature selection, missing data, and computational complexity. Another challenge for ML and DL is tackling the complexity of the healthcare domain.

Healthcare and biomedical research present more intricate challenges than other fields. There is still a lot we do not know about the origins, transmission, and cures for many of these incredibly diverse diseases. It is hard to collect sufficient data because there are not always enough patients. A solution to this issue may be found, however. The small number of patients necessitates exhaustive patient profiling, innovative data processing, and the incorporation of additional datasets. Researchers can process each dataset independently using the appropriate DL technique and then represent the results in a unified model to extract patient data.

The use of ML and DL techniques for healthcare prediction has the potential to change the way traditional healthcare services are delivered. In the case of ML and DL applications, healthcare data is deemed the most significant component that contributes to medical care systems. This paper aims to present a comprehensive review of the most significant ML and DL techniques employed in healthcare predictive analytics. In addition, it discussed the obstacles and challenges of applying ML and DL Techniques in the healthcare domain. As a result of this survey, a total of 41 papers covering the period from 2019 to 2022 were selected and thoroughly reviewed. In addition, the methodology for each paper was discussed in detail. The reviewed studies have shown that AI techniques (ML and DL) play a significant role in accurately diagnosing diseases and helping to anticipate and analyze healthcare data by linking hundreds of clinical records and rebuilding a patient's history using these data. This work advances research in the field of healthcare predictive analytics using ML and DL approaches and contributes to the literature and future studies by serving as a resource for other academics and researchers.

Availability of data and materials

Not applicable.

Abbreviations

Artificial Intelligence

Machine Learning

Decision Tree

Electronic Health Records

Random Forest

Support Vector Machine

K-Nearest Neighbor

Naive Bayes

Reinforcement Learning

Natural Language Processing

Monte Carlo Tree Search

Partially Observable Markov Decision Processes

Deep Learning

Deep Belief Network

Artificial Neural Networks

Convolutional Neural Networks

Long Short-Term Memory

Recurrent Convolution Neural Networks

Recurrent Neural Networks

Recurrent Convolutional Layer

Receptive Domains

Recurrent Multilayer Perceptron

Pima Indian Diabetes Database

Coronary Heart Disease

Chest X-Ray

Multilayer Perceptrons

Internet of Things

Dilated RNN

Neural Networks

Support Vector Regression

Principal Component Analysis

Deep Neural Network

Prediction Horizons

Root Mean Square Error

Intrinsic Modal Functions

Variational Modal Decomposition

Self-Monitored Blood Glucose

Latha MH, Ramakrishna A, Reddy BSC, Venkateswarlu C, Saraswathi SY (2022) Disease prediction by stacking algorithms over big data from healthcare communities. Intell Manuf Energy Sustain: Proc ICIMES 2021(265):355

Google Scholar  

Van Calster B, Wynants L, Timmerman D, Steyerberg EW, Collins GS (2019) Predictive analytics in health care: how can we know it works? J Am Med Inform Assoc 26(12):1651–1654

Sahoo PK, Mohapatra SK, Wu SL (2018) SLA based healthcare big data analysis and computing in cloud network. J Parallel Distrib Comput 119:121–135

Thanigaivasan V, Narayanan SJ, Iyengar SN, Ch N (2018) Analysis of parallel SVM based classification technique on healthcare using big data management in cloud storage. Recent Patents Comput Sci 11(3):169–178

Elmahdy HN (2014) Medical diagnosis enhancements through artificial intelligence

Xiong X, Cao X, Luo L (2021) The ecology of medical care in Shanghai. BMC Health Serv Res 21:1–9

Donev D, Kovacic L, Laaser U (2013) The role and organization of health care systems. Health: systems, lifestyles, policies, 2nd edn. Jacobs Verlag, Lage, pp 3–144

Murphy G F, Hanken M A, & Waters K A (1999) Electronic health records: changing the vision

Qayyum A, Qadir J, Bilal M, Al-Fuqaha A (2020) Secure and robust machine learning for healthcare: a survey. IEEE Rev Biomed Eng 14:156–180

El Seddawy AB, Moawad R, Hana MA (2018) Applying data mining techniques in CRM

Wang Y, Kung L, Wang WYC, Cegielski CG (2018) An integrated big data analytics-enabled transformation model: application to health care. Inform Manag 55(1):64–79

Mirbabaie M, Stieglitz S, Frick NR (2021) Artificial intelligence in disease diagnostics: a critical review and classification on the current state of research guiding future direction. Heal Technol 11(4):693–731

Tang R, De Donato L, Besinović N, Flammini F, Goverde RM, Lin Z, Wang Z (2022) A literature review of artificial intelligence applications in railway systems. Transp Res Part C: Emerg Technol 140:103679

Singh G, Al’Aref SJ, Van Assen M, Kim TS, van Rosendael A, Kolli KK, Dwivedi A, Maliakal G, Pandey M, Wang J, Do V (2018) Machine learning in cardiac CT: basic concepts and contemporary data. J Cardiovasc Comput Tomograph 12(3):192–201

Kim KJ, Tagkopoulos I (2019) Application of machine learning in rheumatic disease research. Korean J Intern Med 34(4):708

Liu B (2011) Web data mining: exploring hyperlinks, contents, and usage data. Spriger, Berlin

MATH   Google Scholar  

Haykin S, Lippmann R (1994) Neural networks, a comprehensive foundation. Int J Neural Syst 5(4):363–364

Gupta M, Pandya SD (2022) A comparative study on supervised machine learning algorithm. Int J Res Appl Sci Eng Technol (IJRASET) 10(1):1023–1028

Ray S (2019) A quick review of machine learning algorithms. In: 2019 international conference on machine learning, big data, cloud and parallel computing (COMITCon) (pp 35–39). IEEE

Srivastava A, Saini S, & Gupta D (2019) Comparison of various machine learning techniques and its uses in different fields. In: 2019 3rd international conference on electronics, communication and aerospace technology (ICECA) (pp 81–86). IEEE

Park HA (2013) An introduction to logistic regression: from basic concepts to interpretation with particular attention to nursing domain. J Korean Acad Nurs 43(2):154–164

Obulesu O, Mahendra M, & Thrilok Reddy M (2018) Machine learning techniques and tools: a survey. In: 2018 international conference on inventive research in computing applications (ICIRCA) (pp 605–611). IEEE

Dhall D, Kaur R, & Juneja M (2020) Machine learning: a review of the algorithms and its applications. Proceedings of ICRIC 2019: recent innovations in computing 47–63

Yang F J (2019) An extended idea about Decision Trees. In: 2019 international conference on computational science and computational intelligence (CSCI) (pp 349–354). IEEE

Eesa AS, Orman Z, Brifcani AMA (2015) A novel feature-selection approach based on the cuttlefish optimization algorithm for intrusion detection systems. Expert Syst Appl 42(5):2670–2679

Shamim A, Hussain H, & Shaikh M U (2010) A framework for generation of rules from Decision Tree and decision table. In: 2010 international conference on information and emerging technologies (pp 1–6). IEEE

Eesa AS, Abdulazeez AM, Orman Z (2017) A dids based on the combination of cuttlefish algorithm and Decision Tree. Sci J Univ Zakho 5(4):313–318

Bakyarani ES, Srimathi H, Bagavandas M (2019) A survey of machine learning algorithms in health care. Int J Sci Technol Res 8(11):223

Resende PAA, Drummond AC (2018) A survey of random forest based methods for intrusion detection systems. ACM Comput Surv (CSUR) 51(3):1–36

Breiman L (2001) Random forests. Mach learn 45:5–32

Ho TK (1998) The random subspace method for constructing decision forests. IEEE Trans Pattern Anal Mach Intell 20(8):832–844

Hofmann M, & Klinkenberg R (2016) RapidMiner: data mining use cases and business analytics applications. CRC Press

Chow CKCN, Liu C (1968) Approximating discrete probability distributions with dependence trees. IEEE Trans Inf Theory 14(3):462–467

Burges CJ (1998) A tutorial on support vector machines for pattern recognition. Data Min Knowl Disc 2(2):121–167

Han J, Pei J, Kamber M (1999) Data mining: concepts and techniques. 2011

Cortes C, Vapnik V (1995) Support-vector networks. Mach learn 20:273–297

Aldahiri A, Alrashed B, Hussain W (2021) Trends in using IoT with machine learning in health prediction system. Forecasting 3(1):181–206

Sarker IH (2021) Machine learning: Algorithms, real-world applications and research directions. SN Comput Sci 2(3):160

Ting K M, & Zheng Z (1999) Improving the performance of boosting for naive Bayesian classification. In: Methodologies for knowledge discovery and data mining: third Pacific-Asia conference, PAKDD-99 Beijing, China, Apr 26–28, 1999 proceedings 3 (pp 296–305). Springer Berlin Heidelberg

Oladipo ID, AbdulRaheem M, Awotunde JB, Bhoi AK, Adeniyi EA, Abiodun MK (2022) Machine learning and deep learning algorithms for smart cities: a start-of-the-art review. In: IoT and IoE driven smart cities, pp 143–162

Shailaja K, Seetharamulu B, & Jabbar M A Machine learning in healthcare: a review. In: 2018 second international conference on electronics, communication and aerospace technology (ICECA) 2018 Mar 29 (pp 910–914)

Mahesh B (2020) Machine learning algorithms-a review. Int J Sci Res (IJSR) 9:381–386

Greene D, Cunningham P, & Mayer R (2008) Unsupervised learning and clustering. Mach learn Techn Multimed: Case Stud Organ Retriev 51–90

Jain AK, Dubes RC (1988) Algorithms for clustering data. Prentice-Hall, Inc, USA

Kodinariya TM, Makwana PR (2013) Review on determining number of cluster in K-means clustering. Int J 1(6):90–95

Smith LI (2002) A tutorial on principal components analysis

Mishra SP, Sarkar U, Taraphder S, Datta S, Swain D, Saikhom R, Laishram M (2017) Multivariate statistical data analysis-principal component analysis (PCA). Int J Livestock Res 7(5):60–78

Kamani M, Farzin Haddadpour M, Forsati R, and Mahdavi M (2019) "Efficient Fair Principal Component Analysis." arXiv e-prints: arXiv-1911.

Dey A (2016) Machine learning algorithms: a review. Int J Comput Sci Inf Technol 7(3):1174–1179

Agrawal R, Imieliński T, & Swami A (1993) Mining association rules between sets of items in large databases. In: proceedings of the 1993 ACM SIGMOD international conference on Management of data (pp 207–216)

Agrawal R, & Srikant R (1994) Fast algorithms for mining association rules. In: Proceeding of 20th international conference very large data bases, VLDB (Vol 1215, pp 487-499)

Singh J, Ram H, Sodhi DJ (2013) Improving efficiency of apriori algorithm using transaction reduction. Int J Sci Res Publ 3(1):1–4

Al-Maolegi M, & Arkok B (2014) An improved Apriori algorithm for association rules. arXiv preprint arXiv:1403.3948

Abaya SA (2012) Association rule mining based on Apriori algorithm in minimizing candidate generation. Int J Sci Eng Res 3(7):1–4

Coronato A, Naeem M, De Pietro G, Paragliola G (2020) Reinforcement learning for intelligent healthcare applications: a survey. Artif Intell Med 109:101964

Watkins CJ, Dayan P (1992) Q-learning. Mach Learn 8:279–292

Jang B, Kim M, Harerimana G, Kim JW (2019) Q-learning algorithms: a comprehensive classification and applications. IEEE access 7:133653–133667

Achille A, Soatto S (2018) Information dropout: Learning optimal representations through noisy computation. IEEE Trans Pattern Anal Mach Intell 40(12):2897–2905

Williams G, Wagener N, Goldfain B, Drews P, Rehg J M, Boots B, & Theodorou E A (2017) Information theoretic MPC for model-based reinforcement learning. In: 2017 IEEE international conference on robotics and automation (ICRA) (pp 1714–1721). IEEE

Wilkes JT, Gallistel CR (2017) Information theory, memory, prediction, and timing in associative learning. Comput Models Brain Behav 29:481–492

Ning Y, Jia J, Wu Z, Li R, An Y, Wang Y, Meng H (2017) Multi-task deep learning for user intention understanding in speech interaction systems. In: Proceedings of the AAAI conference on artificial intelligence (Vol 31, No. 1)

Shi X, Gao Z, Lausen L, Wang H, Yeung DY, Wong WK, Woo WC (2017) Deep learning for precipitation nowcasting: a benchmark and a new model. In: Guyon I, Von Luxburg U, Bengio S, Wallach H, Fergus R, Vishwanathan S, Garnett R (Eds) Advances in neural information processing systems, vol 30. Curran Associates, Inc.,. https://proceedings.neurips.cc/paper_files/paper/2017/file/a6db4ed04f1621a119799fd3d7545d3d-Paper.pdf

Juang CF, Lu CM (2009) Ant colony optimization incorporated with fuzzy Q-learning for reinforcement fuzzy control. IEEE Trans Syst, Man, Cybernet-Part A: Syst Humans 39(3):597–608

Świechowski M, Godlewski K, Sawicki B, Mańdziuk J (2022) Monte Carlo tree search: a review of recent modifications and applications. Artif Intell Rev 56:1–66

Lizotte DJ, Laber EB (2016) Multi-objective Markov decision processes for data-driven decision support. J Mach Learn Res 17(1):7378–7405

MathSciNet   MATH   Google Scholar  

Silver D, Huang A, Maddison CJ, Guez A, Sifre L, Van Den Driessche G, Hassabis D (2016) Mastering the game of go with deep neural networks and tree search. Nature 529(7587):484–489

Browne CB, Powley E, Whitehouse D, Lucas SM, Cowling PI, Rohlfshagen P, Colton S (2012) A survey of monte carlo tree search methods. IEEE Trans Comput Intell AI Games 4(1):1–43

Ling ZH, Kang SY, Zen H, Senior A, Schuster M, Qian XJ, Deng L (2015) Deep learning for acoustic modeling in parametric speech generation: a systematic review of existing techniques and future trends. IEEE Signal Process Magaz 32(3):35–52

Schmidhuber J (2015) Deep learning in neural networks: an overview. Neural Netw 61:85–117

Yu D, Deng L (2010) Deep learning and its applications to signal and information processing [exploratory dsp]. IEEE Signal Process Mag 28(1):145–154

Hinton GE, Osindero S, Teh YW (2006) A fast learning algorithm for deep belief nets. Neural Comput 18(7):1527–1554

Goyal P, Pandey S, Jain K, Goyal P, Pandey S, Jain K (2018) Introduction to natural language processing and deep learning. Deep Learn Nat Language Process: Creat Neural Netw Python 1–74. https://doi.org/10.1007/978-1-4842-3685-7

Mathew A, Amudha P, Sivakumari S (2021) Deep learning techniques: an overview. Adv Mach Learn Technol Appl: Proc AMLTA 2020:599–608

Goodfellow I, Bengio Y, Courville A (2016) Deep learning. MIT press, USA

Gomes L (2014) Machine-learning maestro Michael Jordan on the delusions of big data and other huge engineering efforts. IEEE Spectrum 20. https://spectrum.ieee.org/machinelearning-maestro-michael-jordan-on-the-delusions-of-big-data-and-other-huge-engineering-efforts

Huang G, Liu Z, Van Der Maaten L, & Weinberger K Q (2017) Densely connected convolutional networks. In: proceedings of the IEEE conference on computer vision and pattern recognition (pp 4700–4708)

Yap MH, Pons G, Marti J, Ganau S, Sentis M, Zwiggelaar R, Marti R (2017) Automated breast ultrasound lesions detection using convolutional neural networks. IEEE J Biomed Health Inform 22(4):1218–1226

Hayashi Y (2019) The right direction needed to develop white-box deep learning in radiology, pathology, and ophthalmology: a short review. Front Robot AI 6:24

Alom MZ, Taha TM, Yakopcic C, Westberg S, Sidike P, Nasrin MS, Asari VK (2019) A state-of-the-art survey on deep learning theory and architectures. Electronics 8(3):292

Schmidhuber J, Hochreiter S (1997) Long short-term memory. Neural Comput 9(8):1735–1780

Smagulova K, James AP (2019) A survey on LSTM memristive neural network architectures and applications. Eur Phys J Spec Topics 228(10):2313–2324

Setyanto A, Laksito A, Alarfaj F, Alreshoodi M, Oyong I, Hayaty M, Kurniasari L (2022) Arabic language opinion mining based on long short-term memory (LSTM). Appl Sci 12(9):4140

Lindemann B, Müller T, Vietz H, Jazdi N, Weyrich M (2021) A survey on long short-term memory networks for time series prediction. Procedia CIRP 99:650–655

Cui Z, Ke R, Pu Z, & Wang Y (2018) Deep bidirectional and unidirectional LSTM recurrent neural network for network-wide traffic speed prediction. arXiv preprint arXiv:1801.02143

Villegas R, Yang J, Zou Y, Sohn S, Lin X, & Lee H (2017) Learning to generate long-term future via hierarchical prediction. In: international conference on machine learning (pp 3560–3569). PMLR

Gensler A, Henze J, Sick B, & Raabe N (2016) Deep learning for solar power forecasting—an approach using autoencoder and LSTM neural networks. In: 2016 IEEE international conference on systems, man, and cybernetics (SMC) (pp 002858–002865). IEEE

Lindemann B, Fesenmayr F, Jazdi N, Weyrich M (2019) Anomaly detection in discrete manufacturing using self-learning approaches. Procedia CIRP 79:313–318

Kalchbrenner N, Danihelka I, & Graves A (2015) Grid long short-term memory. arXiv preprint arXiv:1507.01526

Cheng B, Xu X, Zeng Y, Ren J, Jung S (2018) Pedestrian trajectory prediction via the social-grid LSTM model. J Eng 2018(16):1468–1474

Veličković P, Karazija L, Lane N D, Bhattacharya S, Liberis E, Liò P & Vegreville M (2018) Cross-modal recurrent models for weight objective prediction from multimodal time-series data. In: proceedings of the 12th EAI international conference on pervasive computing technologies for healthcare (pp 178–186)

Wang J, Hu X (2021) Convolutional neural networks with gated recurrent connections. IEEE Trans Pattern Anal Mach Intell 44(7):3421–3435

Liang M, & Hu X (2015) Recurrent convolutional neural network for object recognition. In: proceedings of the IEEE conference on computer vision and pattern recognition (pp 3367–3375)

Liang M, Hu X, Zhang B (2015) Convolutional neural networks with intra-layer recurrent connections for scene labeling. In: Cortes C, Lawrence N, Lee D, Sugiyama M, Garnett R (Eds) Advances in Neural Information Processing Systems, vol 28. Curran Associates, Inc. https://proceedings.neurips.cc/paper_files/paper/2015/file/9cf81d8026a9018052c429cc4e56739b-Paper.pdf

Fernandez B, Parlos A G, & Tsai W K (1990) Nonlinear dynamic system identification using artificial neural networks (ANNs). In: 1990 IJCNN international joint conference on neural networks (pp 133–141). IEEE

Puskorius GV, Feldkamp LA (1994) Neurocontrol of nonlinear dynamical systems with Kalman filter trained recurrent networks. IEEE Trans Neural Netw 5(2):279–297

Rumelhart DE (1986) Learning representations by error propagation. In: DE Rumelhart and JL McClelland & PDP Research Group, eds, Parallel distributed processing: explorations in the microstructure of cognition. Bradford Books MITPress, Cambridge, Mass

Krishnamoorthi R, Joshi S, Almarzouki H Z, Shukla P K, Rizwan A, Kalpana C, & Tiwari B (2022) A novel diabetes healthcare disease prediction framework using machine learning techniques. J Healthcare Eng. https://doi.org/10.1155/2022/1684017

Edeh MO, Khalaf OI, Tavera CA, Tayeb S, Ghouali S, Abdulsahib GM, Louni A (2022) A classification algorithm-based hybrid diabetes prediction model. Front Publ Health 10:829510

Iwendi C, Huescas C G Y, Chakraborty C, & Mohan S (2022) COVID-19 health analysis and prediction using machine learning algorithms for Mexico and Brazil patients. J Experiment Theor Artif Intell 1–21. https://doi.org/10.1080/0952813X.2022.2058097

Lu H, Uddin S, Hajati F, Moni MA, Khushi M (2022) A patient network-based machine learning model for disease prediction: the case of type 2 diabetes mellitus. Appl Intell 52(3):2411–2422

Chugh M, Johari R, & Goel A (2022) MATHS: machine learning techniques in healthcare system. In: international conference on innovative computing and communications: proceedings of ICICC 2021, Volume 3 (pp 693–702). Springer Singapore

Deberneh HM, Kim I (2021) Prediction of type 2 diabetes based on machine learning algorithm. Int J Environ Res Public Health 18(6):3317

Gupta S, Verma H K, & Bhardwaj D (2021) Classification of diabetes using Naive Bayes and support vector machine as a technique. In: operations management and systems engineering: select proceedings of CPIE 2019 (pp 365–376). Springer Singapore

Islam M T, Rafa S R, & Kibria M G (2020) Early prediction of heart disease using PCA and hybrid genetic algorithm with k-means. In: 2020 23rd international conference on computer and information technology (ICCIT) (pp 1–6). IEEE

Qawqzeh Y K, Bajahzar A S, Jemmali M, Otoom M M, Thaljaoui A (2020) Classification of diabetes using photoplethysmogram (PPG) waveform analysis: logistic regression modeling. BioMed Res Int. https://doi.org/10.1155/2020/3764653

Grampurohit S, Sagarnal C (2020) Disease prediction using machine learning algorithms. In: 2020 international conference for emerging technology (INCET) (pp 1–7). IEEE

Moturi S, Srikanth Vemuru DS (2020) Classification model for prediction of heart disease using correlation coefficient technique. Int J 9(2). https://doi.org/10.30534/ijatcse/2020/185922020

Barik S, Mohanty S, Rout D, Mohanty S, Patra A K, & Mishra A K (2020) Heart disease prediction using machine learning techniques. In: advances in electrical control and signal systems: select proceedings of AECSS 2019 (pp 879–888). Springer, Singapore

Princy R J P, Parthasarathy S, Jose P S H, Lakshminarayanan A R, & Jeganathan S (2020) Prediction of cardiac disease using supervised machine learning algorithms. In: 2020 4th international conference on intelligent computing and control systems (ICICCS) (pp 570–575). IEEE

Saw M, Saxena T, Kaithwas S, Yadav R, & Lal N (2020) Estimation of prediction for getting heart disease using logistic regression model of machine learning. In: 2020 international conference on computer communication and informatics (ICCCI) (pp 1–6). IEEE

Soni VD (2020) Chronic disease detection model using machine learning techniques. Int J Sci Technol Res 9(9):262–266

Indrakumari R, Poongodi T, Jena SR (2020) Heart disease prediction using exploratory data analysis. Procedia Comput Sci 173:130–139

Wu C S M, Badshah M, & Bhagwat V (2019) Heart disease prediction using data mining techniques. In: proceedings of the 2019 2nd international conference on data science and information technology (pp 7–11)

Tarawneh M, & Embarak O (2019) Hybrid approach for heart disease prediction using data mining techniques. In: advances in internet, data and web technologies: the 7th international conference on emerging internet, data and web technologies (EIDWT-2019) (pp 447–454). Springer International Publishing

Rahman AS, Shamrat FJM, Tasnim Z, Roy J, Hossain SA (2019) A comparative study on liver disease prediction using supervised machine learning algorithms. Int J Sci Technol Res 8(11):419–422

Gonsalves A H, Thabtah F, Mohammad R M A, & Singh G (2019) Prediction of coronary heart disease using machine learning: an experimental analysis. In: proceedings of the 2019 3rd international conference on deep learning technologies (pp 51–56)

Khan A, Uddin S, Srinivasan U (2019) Chronic disease prediction using administrative data and graph theory: the case of type 2 diabetes. Expert Syst Appl 136:230–241

Alanazi R (2022) Identification and prediction of chronic diseases using machine learning approach. J Healthcare Eng. https://doi.org/10.1155/2022/2826127

Gouda W, Almurafeh M, Humayun M, Jhanjhi NZ (2022) Detection of COVID-19 based on chest X-rays using deep learning. Healthcare 10(2):343

Kumar A, Satyanarayana Reddy S S, Mahommad G B, Khan B, & Sharma R (2022) Smart healthcare: disease prediction using the cuckoo-enabled deep classifier in IoT framework. Sci Progr. https://doi.org/10.1155/2022/2090681

Monday H N, Li J P, Nneji G U, James E C, Chikwendu I A, Ejiyi C J, & Mgbejime G T (2021) The capability of multi resolution analysis: a case study of COVID-19 diagnosis. In: 2021 4th international conference on pattern recognition and artificial intelligence (PRAI) (pp 236–242). IEEE

Al Rahhal MM, Bazi Y, Jomaa RM, Zuair M, Al Ajlan N (2021) Deep learning approach for COVID-19 detection in computed tomography images. Cmc-Comput Mater Continua 67(2):2093–2110

Men L, Ilk N, Tang X, Liu Y (2021) Multi-disease prediction using LSTM recurrent neural networks. Expert Syst Appl 177:114905

Ahmad U, Song H, Bilal A, Mahmood S, Alazab M, Jolfaei A & Saeed U (2021) A novel deep learning model to secure internet of things in healthcare. Mach Intell Big Data Anal Cybersec Appl 341–353

Mansour RF, El Amraoui A, Nouaouri I, Díaz VG, Gupta D, Kumar S (2021) Artificial intelligence and internet of things enabled disease diagnosis model for smart healthcare systems. IEEE Access 9:45137–45146

Sevi M, & Aydin İ (2020) COVID-19 detection using deep learning methods. In: 2020 international conference on data analytics for business and industry: way towards a sustainable economy (ICDABI) (pp 1–6). IEEE

Martinsson J, Schliep A, Eliasson B, Mogren O (2020) Blood glucose prediction with variance estimation using recurrent neural networks. J Healthc Inform Res 4:1–18

Zhang J, Xie Y, Pang G, Liao Z, Verjans J, Li W, Xia Y (2020) Viral pneumonia screening on chest X-rays using confidence-aware anomaly detection. IEEE Trans Med Imaging 40(3):879–890

Hemdan E E D, Shouman M A, & Karar M E (2020) Covidx-net: a framework of deep learning classifiers to diagnose covid-19 in x-ray images. arXiv preprint arXiv:2003.11055

Zhu T, Li K, Chen J, Herrero P, Georgiou P (2020) Dilated recurrent neural networks for glucose forecasting in type 1 diabetes. J Healthc Inform Res 4:308–324

Cheon S, Kim J, Lim J (2019) The use of deep learning to predict stroke patient mortality. Int J Environ Res Public Health 16(11):1876

Li K, Liu C, Zhu T, Herrero P, Georgiou P (2019) GluNet: a deep learning framework for accurate glucose forecasting. IEEE J Biomed Health Inform 24(2):414–423

Wang W, Tong M, Yu M (2020) Blood glucose prediction with VMD and LSTM optimized by improved particle swarm optimization. IEEE Access 8:217908–217916

Rashid N, Hossain M A F, Ali M, Sukanya M I, Mahmud T, & Fattah S A (2020) Transfer learning based method for COVID-19 detection from chest X-ray images. In: 2020 IEEE region 10 conference (TENCON) (pp 585–590). IEEE

Arora P, Kumar H, Panigrahi BK (2020) Prediction and analysis of COVID-19 positive cases using deep learning models: a descriptive case study of India. Chaos, Solitons Fractals 139:110017

MathSciNet   Google Scholar  

Zaitcev A, Eissa MR, Hui Z, Good T, Elliott J, Benaissa M (2020) A deep neural network application for improved prediction of in type 1 diabetes. IEEE J Biomed Health Inform 24(10):2932–2941

Naz H, Ahuja S (2020) Deep learning approach for diabetes prediction using PIMA Indian dataset. J Diabetes Metab Disord 19:391–403

Download references

Acknowledgements

Author information, authors and affiliations.

Department of Information Systems and Technology, Faculty of Graduate Studies for Statistical Research, Cairo University, Giza, Egypt

Mohammed Badawy & Nagy Ramadan

Department of Computer Sciences, Faculty of Graduate Studies for Statistical Research, Cairo University, Giza, Egypt

Hesham Ahmed Hefny

You can also search for this author in PubMed   Google Scholar

Contributions

MB wrote the main text of the manuscript; NR and HAH revised the manuscript. All authors read and approved the final manuscript.

Corresponding author

Correspondence to Mohammed Badawy .

Ethics declarations

Competing interests.

The authors declare that they have no competing interests. All authors approved the final manuscript.

Additional information

Publisher's note.

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/ .

Reprints and permissions

About this article

Cite this article.

Badawy, M., Ramadan, N. & Hefny, H.A. Healthcare predictive analytics using machine learning and deep learning techniques: a survey. Journal of Electrical Systems and Inf Technol 10 , 40 (2023). https://doi.org/10.1186/s43067-023-00108-y

Download citation

Received : 27 December 2022

Accepted : 31 July 2023

Published : 29 August 2023

DOI : https://doi.org/10.1186/s43067-023-00108-y

Share this article

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

• Healthcare prediction
• Artificial intelligence (AI)
• Machine learning (ML)
• Deep learning (DL)
• Medical diagnosis

An official website of the United States government

The .gov means it’s official. Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

The site is secure. The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

• Publications
• Account settings

Preview improvements coming to the PMC website in October 2024. Learn More or Try it out now .

• Advanced Search
• Journal List
• Springer Nature - PMC COVID-19 Collection

Machine learning and deep learning approach for medical image analysis: diagnosis to detection

Meghavi rana.

School of Computing, DIT University, Dehradun, India

Megha Bhushan

Associated data.

Data sharing is applicable to this article as MRI datasets was analyzed during the current study.

Computer-aided detection using Deep Learning (DL) and Machine Learning (ML) shows tremendous growth in the medical field. Medical images are considered as the actual origin of appropriate information required for diagnosis of disease. Detection of disease at the initial stage, using various modalities, is one of the most important factors to decrease mortality rate occurring due to cancer and tumors. Modalities help radiologists and doctors to study the internal structure of the detected disease for retrieving the required features. ML has limitations with the present modalities due to large amounts of data, whereas DL works efficiently with any amount of data. Hence, DL is considered as the enhanced technique of ML where ML uses the learning techniques and DL acquires details on how machines should react around people. DL uses a multilayered neural network to get more information about the used datasets. This study aims to present a systematic literature review related to applications of ML and DL for the detection along with classification of multiple diseases. A detailed analysis of 40 primary studies acquired from the well-known journals and conferences between Jan 2014–2022 was done. It provides an overview of different approaches based on ML and DL for the detection along with the classification of multiple diseases, modalities for medical imaging, tools and techniques used for the evaluation, description of datasets. Further, experiments are performed using MRI dataset to provide a comparative analysis of ML classifiers and DL models. This study will assist the healthcare community by enabling medical practitioners and researchers to choose an appropriate diagnosis technique for a given disease with reduced time and high accuracy.

Introduction

The significance of disease classification and prediction can be observed from the previous years. The important properties and features given in a dataset should be well-known to identify the exact cause along with the symptom of the disease. Artificial Intelligence (AI) has shown promising results by classifying and assisting in decision making. Machine Learning (ML), a subset of AI, has accelerated many research related to the medical field. Whereas, Deep Learning (DL) is a subset of ML that deals with neural network layers, analyzing the exact features required for disease detection [ 34 , 71 , 94 ]. The existing studies from 2014 to present, discusses many applications and algorithms developed for enhancing the medical field by providing accurate results for a patient. Using data, ML has driven advanced technologies in many areas including natural language processing, automatic speech recognition, and computer vision to deliver robust systems such as driverless cars, automated translation, etc. Despite all advances, the application of ML in medical care remained affected with hazards. Many of these issues were raised from medical care stating the goal of making accurate predictions using the collected data and managed by the medical system.

AI examines a given dataset using various techniques to get the required features or highlights from a huge amount of data resulting in difficulty for tracking down an ideal arrangement of significant features and excluding repetitive ones. Considering such features is inconvenient and accuracy metrics becomes erroneous. Hence, choosing a small subset from a wide scope of features will upgrade the efficiency of the model. Subsequently, the exclusion of inconvenient and repetitive features will decline the dimensionality of the information, speed up the learned model similar to boosting [ 37 ]. From the existing features, the significant features are extracted using practical approaches such as Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA). Particularly, choosing a feature has two essential clashing objectives, first, boosting the presentation of arrangement and second, limiting the count of features to conquer the issue of dimensionality. Hence, selection of features is considered as an essential task for aforementioned objectives. Later, research related to the features improvement was enhanced by using choice-based multi-target strategies. Thus, in this review, strategies to choose efficient features will be focused.

Cancer disease was identified using multiple techniques of image segmentation, feature selection, and regression using Root Mean Square Error (RMSE), with the parameters such as recognizing patterns, detecting objects, and classifying the image [ 7 ]. Brain tumor was detected using six classifiers and Transfer Learning (TL) techniques for image segmentation with Magnetic Resonance Imaging (MRI) of the brain [ 28 ]. Also, a TL approach was implemented to identify lung cancer and brain disease in [ 55 ]. It analyzed MRI and Computer-Tomography (CT) scan images by using supervised learning Support Vector Machine (SVM) classifiers. The image analysis process has been well understood in the existing studies. However, the techniques using ML and DL are continuously being updated. Therefore, it is a complex task for researchers to identify an accurate method for analyzing images and feature selection techniques varying with every method. The key contributions of this study include:

• (i) Classification of diseases after reviewing primary studies,
• (ii) Recognition of various image modalities provided by existing articles,
• (iii) Description of tools along with reliable ML and DL techniques for disease prediction,
• (iv) Dataset description to provide awareness of available sources,
• (v) Experimental results using MRI dataset to compare different ML and DL methods,
• (vi) Selection of suitable features and classifiers to get better accuracy, and.
• (vii) Insights on classification as well as review of the techniques to infer future research.

The significance of this review is to enable physicians or clinicians to use ML or DL techniques for precise and reliable detection, classification and diagnosis of the disease. Also, it will assist clinicians and researchers to avoid misinterpretation of datasets and derive efficient algorithms for disease diagnosis along with information on the multiple modern medical imaging modalities of ML and DL.

The study presented consists of 11 sections. The organization of the section is described as follows: Section 2  discusses the background of study, Section  3  discusses the review techniques, search criteria, source material and the quality assessment. Section 4  summarizes the current techniques and important parameters to acquire good accuracy. Section 5  gives an insight of medical image modalities. Section 6  sums up the tools and techniques being used in ML and DL models. Section 7  discusses the datasets used by the authors previously and gives an insight of data. Section 8  represents the experimental section using ML classifiers and DL models over brain MRI dataset. Section 9  recaps the analytic discussion about the techniques, datasets being used, tools in ML and DL, journals studied for the given article. Discussion, conclusion and future scope is discussed in Sections  10 and 11 , respectively.

This section discusses the preliminary terms which are required to comprehend this review. Further, it also presents the statistical analysis of ML and DL techniques used for medical image diagnosis.

Machine learning

ML is a branch of AI where a machine learns from the data by identifying patterns and automates decision-making with minimum human intervention [ 96 , 24 , 12 ]. The most important characteristic of a ML model is to adapt independently, learn from previous calculations and produce reliable results when new datasets are exposed to models repeatedly. The two main aspects include (i) ML techniques help the physicians to interpret medical images using Computer Aided Design (CAD) in a small period of time, and (ii) algorithms used for challenging tasks like segmentation with CT scan [ 81 ], breast cancer and mammography, segmenting brain tumors with MRI. Traditional ML models worked on structured datasets where the techniques were predefined for every step, the applied technique fails if any of the steps were missed. The process of evaluating the data quality used by ML and DL algorithms is essential [ 16 – 22 , 61 ]. Whereas, new algorithms adapt the omission of data based on the requirement for robustness of the algorithm. Figure ​ Figure1 1 illustrates the process used by ML algorithms for the prediction and diagnosis of disease.

ML process [ 10 ]

Deep learning

DL models enable machines to achieve the accuracy by advancements in techniques to analyze medical images. In [ 58 ], the heart disease was diagnosed using the labelled chest X-Rays, cardiologist reviewed and relabelled all the data while discarding the data other than heart failure and normal images. To extract the exact features from the images, data augmentation and TL were used with 82% accuracy, 74% specificity and 95% sensitivity for heart failure. In [ 14 ], an automatic feature selection, using histopathology images with the labelling of positive and negative cancer images, was developed with minimum manual work. Two networks named Deep Neural Network (DNN) 2-F and DNN1-F were used with PCA to reduce features in DNN whereas for unsupervised feature learning a single-layer network of K-means centroids was used. Later, the results of unsupervised (93.56%) and supervised (94.52%) learning were compared. The DL model automates the feature extraction procedure to handle data efficiently [ 14 , 74 ]. Figure ​ Figure2 2 depicts the process used by DL algorithms for the prediction and diagnosis of various diseases.

To process the medical images for better prediction and accuracy, ML and DL techniques were used as shown in Figs.  1 and ​ and2, 2 , respectively. As input, medical images from various modalities are taken into consideration, and then algorithms are applied to these images. Further, the input image is segmented based on various factors, these segments were used to extract the essential and maximum features using feature extraction techniques. After the extraction of the required features, they are further refined to obtain actual features used for the identification of diseases [ 60 ]. Also, ML approaches were used to denoise the medical images for better prediction and accuracy in [ 46 ]. Once the feature selection and noise removal from the data are achieved, the classification of the images according to the disease using classifiers like SVM, Decision Tree (DT), etc. was attained.

ML is the process where computers learn from data and use algorithms to carry out a task without being explicitly programmed. It uses pattern recognition to make predictions with new dataset. Alternatively, DL is modeled according to the human brain including a complex structure of algorithms enabling machines to process images, text and documents. It uses layered-structure algorithms such as Convolutional Neural Network (CNN), Artificial Neural Network (ANN), etc., to analyze the data with logics. Comparatively, DL is more capable of processing huge amount of data than ML models.

Review technique

In this section, an overview of the technique used to conduct this review systematically is discussed. It provides the details of the electronic databases used to search, retrieve information, and discuss the research questions framed to execute the review successfully. The systematic review guidelines implemented by [ 49 , 50 ] were followed for this literature review.

Research questions

In this review, following review questions will be discussed:

• 1.1 What are the considered parameters while selecting the classifiers?
• 1.2 What are the evaluation metrics used to evaluate classification models?
• What are various medical image modalities for classifying the diseases?
• What are the tools and techniques used for medical imaging?
• What are various datasets used by several researchers in the domain of healthcare?
• What are the results of comparative analysis of ML classifiers and DL models based on experiments using MRI dataset?

Source material

The guidelines given in [ 49 , 50 ] are followed for searching the existing literature related to the area of ML and DL in medical imaging. Following electronic database sources are used for searching:

• ScienceDirect ( https://www.sciencedirect.com/ ) .
• IEEE Xplore ( https://ieeexplore.ieee.org/Xplore/home.jsp ) .
• Springer ( https://www.springer.com/in ) .
• PubMed ( https://pubmed.ncbi.nlm.nih.gov/16495534/ ) .
• Wiley Interscience ( https://onlinelibrary.wiley.com/ ) .
• Google Scholar ( https://scholar.google.co.in/ ) .
• IOP ( https://www.iop.org/#gref ) .
• Oxford Publications ( https://india.oup.com/ ) .
• Elsevier ( Elsevier Books and Journals - Elsevier ).
• Hindawi ( https://www.hindawi.com ) .
• Bentham science ( Bentham Science - International Publisher of Journals and Books ).

Search criteria

This review consists of the articles written in English language between the years 2014–2022. The review process can be considered as the filtering process for attaining the quality research articles with the inclusion and exclusion criteria at various stages. The search was based on the keywords as shown in Table ​ Table1 1 to retrieve research articles from various journals, conferences, book chapters, and other sources.

Keywords used

ML  Machine Learning, DL  Deep Learning, ANFC  Adaptive Neuro-Fuzzy Classifier, BI-RADS  Breast Imaging Reporting and Data System, CNN  Convolutional Neural Network, AI  Artificial Intelligence, ANN  Artificial Neural Network, DNN  Deep Neural Network, TL  Transfer Learning

The journals and conferences included were taken from IEEE, Science Direct, Springer, Oxford Publication, etc. The article selection method is depicted in Fig.  3 . As depicted in Fig.  3 , the initial search consisted of 16,900 articles which were refined to 250 based on the specific keywords used as shown in Table ​ Table1. 1 . Then 100 articles were retrieved based on their titles and were reduced to 75 articles based on their abstract and introduction. Finally, 40 articles were selected as primary studies based on the criteria of exclusion and inclusion.

Article selection method

Quality assessment

The quality of this review was assured after inclusion and exclusion criteria discussed in sub-section 3.3 . These primary studies were from various journals, conferences, workshops, and others (manuscripts, online records, and society publications). To retrieve the quality articles, analysis of each article was done to maintain fairness and validation (external and internal) of the results based on the CRD guidelines [ 50 ].

Table ​ Table2 2 presents the top 20 highly influential and cited articles related to the classification of diseases, identification of tools and techniques, explanation for the cause of disease, and solutions to the diagnosed disease (source: https://scholar.google.co.in ).

Top 20 cited articles

J  Journal, BC  Book Chapter, C  Conference, O  Other (Manuscripts, online records, Society publications, Proceedings)

Data extraction

Initially, many challenges were faced to extract the relevant data for this review, therefore, some researchers were approached to acquire the necessary information. The method for extracting the required data in this review is as follow:

• One of the authors extracted the data after a thorough review of 40 articles.
• The acquired results of the review were cross checked by another author to maintain consistency.
• During the process of cross checking (in case of a conflict), issues were resolved by meetings between the authors.

ML and DL techniques for medical imaging

Research question 1 is answered in this section to provide an overview of the current techniques of ML and DL for medical imaging. Further, followed by various parameters considered for selecting the classifiers and the evaluation metrics used to evaluate classification models. The existing literature review is divided according to the diseases such as breast cancer, brain tumor, lung disease, diabetes, multiple disease detection, etc.

Breast disease

In this subsection, articles related to breast disease symptoms, detection, classification, prediction and diagnosis using ML and DL methods are discussed. In [ 33 ], significant features were identified using BI-RADS (Breast Imaging Reporting and Data System) to develop a CAD system for obtaining breast ultrasound. Also, 10-fold cross validation technique was used upon the benign and malignant lesions. As a result, 77% accuracy was achieved using the SVM classifier. However, some methods with a few algorithms handling the vast variety of data need to be understood and analyzed precisely [ 84 ]. CNN was used to train the system with the available clinical data and to comprehend the complex structure. Moreover, it was suggested to study radiomics and expansion of CADx to get the tumor signs using a CAD system. Breast cancer disease was classified using the parameters like Area Under Curve (AUC), sensitivity, and specificity [ 100 ]. A CAD system was developed using CNN where a large number of features were required, using multiview features. These features provide the maximum details of the image data to be extracted for the accuracy of detection and classification.

DL was used for analyzing medical images and also, the limitations along with success of DL techniques for medical imaging were discussed in [ 86 ]. Recent ML and DL technologies were reviewed for the classification and detection of medical imaging modalities [ 39 ]. It provided an insight on the progress of the technology used in the medical field. Various ML techniques used for image processing and DL techniques with the architecture of the algorithm were discussed. To study the technologies, the evaluation of various images such as histological images, thermography images, mammography, ultrasound and MRI using the CAD system was explored. Moreover, the system included ML techniques like SVM, ANN, DT, Naive bayes, K-Nearest Neighbor (KNN), etc.

Brain disease

The concept of TL was used for image segmentation where the MRI scan of the brain was segmented using voxel wise classification [ 7 ]. ML classifiers were applied for the classification of multiple diseases. Later, the results obtained were compared with the existing results to detect the disease.

A brief introduction of DNN in medical image analysis to diagnose the brain tumor using brain tissues is provided in [ 56 ]. It indicated the ways for applying DL to the entire process of MRI scanning, image retrieval, segmentation and disease prediction. It also focused on image acquisition to image retrieval, and from feature segmentation to prediction of disease. The entire process was divided into two parts: (i) the signal processing of MRI including the image restoration and image registration, and (ii) usage of DL for disease detection and prediction-based reports in the form of text and images. Also, the influence of DL in medical imaging was discussed in [ 82 ]. Image segmentation approaches using DL included tumor segmentation, brain and lung’s structure with bone tissues or cells. Patches were taken as input and 2-Dimensional Convolutional Neural Network (2D-CNN) was used to preprocess these at a later stage.

Lung disease

DL has the ability to automate the process of image interpretation which enhances the clinical decision making, identifying the disease and predicting the best treatment for the patient by reviewing the pros and cons of the DL techniques [ 51 ]. These techniques were used for the cardiovascular medication, following are the steps for implementing DL model: (i) problem identification, (ii) data selection, (iii) hardware and software selection, (iv) data preparation, (v) feature selection, and (vi) splitting of data for training as well as validation process. In [ 13 ], a disease was analyzed automatically using labeled data and achieved the accuracy by processing medical images using DL models. The automatic prediction of the disease using ML techniques and the concept of big data was summarized to detect the patterns [ 23 ]. The advantages and disadvantages for each algorithm were also discussed.

A comparative analysis of the classification algorithms based on iris images, using an iridology chart, was done for the diagnosis of diabetes [ 76 ]. Type-2 diabetes was detected by identifying the center of the pupil of an eye at the early stage using the I-Scan-2. Also, a filter-based feature selection method was used with the combination of five classifiers namely binary tree, SVM, neural network model, Random Forest (RF) and adaptive boosting model. Later, in [ 77 ] a study was compiled using the textural, statistical and various features (62 features of iris) to detect the same disease, however, an iridology chart was not used. ML and DL techniques were used to diagnose the errors in existing diagnostic systems [ 81 ]. These techniques were used to analyze the medical images and extract the features which are required for the diagnosis of errors in existing diagnostic systems. Both supervised and unsupervised algorithms were used for the prediction of the disease in specific datasets.

It was observed that DL technique is a way more powerful to investigate medical images [ 65 ]. Various techniques such as image classification, object detection, pattern recognition, etc. were used for the proper decision-making. It improved medical treatments by predicting the early symptoms of a disease. Moreover, an overview of ML and DL techniques used in the medical field was given for providing knowledge to the future researchers. In [ 78 ], techniques such as rubber sheet normalization, ML classifiers, PCA, etc. were used with self-created data and computed six parameters (i) accuracy, (ii) sensitivity, (iii) specificity, (iv)AUC, (v) precision, and (vi) F-score for accurate prediction of Type-2 diabetes.

Multiple disease detection

Multiple diseases were identified with different radiology techniques like MRI imaging for breast cancer along with brain tumor, CAD for breast cancer along with skin lesions, and X-Rays for chest analysis [ 46 ]. Also, ML techniques were used to attain better accuracy with denoising techniques including homographic wavelet, soft thresholding, non-homomorphic and wavelet thresholding. A CAD system using CNN was proposed to diagnose breast lesions as benign and malignant to assist the radiologists [ 100 ]. It was implemented using Inception-v3 architecture to extract the multiview features from Automated Breast Ultrasound (ABUS) images. For the implementation of the model, 316 breast lesions data were trained and evaluated. ML feature extraction scheme was compared with the given method, resulting in 10% increase in AUC value.

A review on image fusion was presented in [ 42 ], it reduced the randomness and improved the quality of available images. Various methods and challenges related to image fusion were also summarized. In [ 44 ], ML and DL techniques focusing on small labeled dataset were discussed as it was considered one of the important factors in decision making. Further, noisy data in medical images was analyzed with pros and cons of various ML algorithms.

In [ 4 ], data augmentation techniques were used to evaluate the dermatology diseases such as acne, atopic dermatitis, impetigo, psoriasis, and rosacea. To diagnose the mentioned diseases, the model was retrained in two phases: (i) with data augmentation, and (ii) without data augmentation using TensorFlow Inception V3. For statistical analysis, both the models were then compared and six parameters namely: (i) Positive Predictive Value (PPV), (ii) Negative Predictive Value (NPV), (iii) Matthew’s Correlation Coefficient (MCC), (iv) sensitivity, (v) specificity, and (vi) F1 score were calculated resulting in an increase of 7.7% average correlation coefficient.

Multiple diseases like diabetes, heart disease, liver disease, dengue and hepatitis were identified by recognizing the pattern in the available data and classifying them further using ML classifiers [ 29 , 27 , 47 ]. It used high-dimensional and multimodal dataset to predict the diseases accurately. The deteriorating condition of a patient was predicted using ML techniques like ML pipelines, classifiers (SVM and 5-fold cross-validation) with the baseline variables from MRI imaging [ 79 ]. AI applications in medical imaging, DL tools for the prediction and pattern recognition were described in [ 87 ]. In addition, apart from AI techniques, ANN and CNN were also useful for predicting the disease by analyzing the image pattern and classification of the disease can be carried out with the help of classifiers [ 62 , 63 ].

Various algorithms were reviewed to detect the error in the diagnosis system implying the importance of ML and DL for early diagnosis of the disease [ 81 ]. Whereas, [ 104 ] discussed the three main challenges: (i) to cope up with image variations, (ii) learning from weak labels, and (iii) interpreting the results with accuracy for the diagnosis of cancer through given medical images. It concluded that TL was used to cope up with image variations. The concept of Multiple Instance Learning (MIL) and weighted TL were used to overcome the weakly labeled data and improve the accuracy of the disease classification for better medical results, respectively. It was suggested to comprehend the relation between image label and image collection instead of learning about the individual instance. The main advantage of the used technique is that it does not require the local manual annotations.

Table ​ Table3 3 represents the current ML and DL techniques for medical imaging, various parameters considered while selecting the classifiers, identified disease and evaluation metrics. Also, early tumor detection can assist clinicians to treat patients timely.

Summary of existing works related to ML and DL techniques for medical imaging

CNN  Convolutional Neural Network, SVM  Support Vector Machine, ML  Machine Learning, DL  Deep Learning, MRI  Magnetic Resonance Imaging, PCA  Principal Component Analysis, BT  Binary Tree, RF  Random Forest, NN  Neural Network, AB  Adaptive Boosting, CAD  Computer-aided Diagnosis System, ANN  Artificial Neural Network, AUC  Area Under Curve, RMSE  Root Mean Square Error, 2D-DWT  Two-Dimensional Discrete Wavelet Transform, MAE  Mean Absolute Error, QSSIM  Quaternion Structure Similarity Index Metric, SSIM  Structure Similarity Index Metric, PCC  Pearson Correlation Coefficient, MoE  Mixture of Experts, MIL  Multiple Instance Learning, PPV  Positive Predictive Value, NPV  Negative Predictive Value, MCC  Matthew’s Correlation Coefficient, TL  Transfer Learning.

Modalities for medical image

Research question 2 (refer subsection 3.1 ) is addressed in this section, various medical image modalities (I-Scan-2, CT-Scan, MRI, X-Ray, Mammogram and Electrocardiogram (ECG)) used for classifying the diseases in the primary studies are shown in Table ​ Table4. 4 . As observed, following modalities were used for the evaluation of medical data using ML and DL techniques.

Modalities for medical imaging and digital signal

CT-SCAN  Computed Tomography Scan, MRI  Magnetic Resonance Imaging, X-Ray  X Radiation, ECG  Electrocardiogram. “+” and “-” signify that the article does and does not support the corresponding parameter, respectively

• MRI : It uses magnetic resonance for obtaining electromagnetic signals. These signals are generated from human organs, which further reconstructs information about human organ structure [ 91 ]. MRIs with high resolution have more structural details which are required to locate lesions and disease diagnosis.
• CT-Scan : It is a technology which generates 3-D images from 2-D X-Ray images using digital geometry [ 88 ].
• Mammogram : For the effective breast cancer screening and early detection of abnormalities in the body, mammograms are used. Calcifications and masses are considered as the most common abnormalities resulting in breast cancer [ 5 ].
• ECG : It is used to measure the heart activity electrically and to detect the cardiac problems in humans [ 8 , 9 , 105 ].

Tools and techniques

This section addresses research question 3 (refer subsection 3.1 ). After a thorough analysis of primary studies, various techniques (refer Table ​ Table6) 6 ) and tools (refer Fig.  4 ) related to ML and DL techniques for healthcare were identified [ 67 , 89 ]. It was observed that techniques have used scanned images with the help of image modalities such as MRI, CT-Scan, X-Rays, and so on. Also, in order to automate the process of image segmentation and classification, programming languages like R, MATLAB and Python were used to obtain accurate results. The subsections 6.1  and 6.2  precisely explain the tools and techniques used in primary studies for medical images, respectively.

Tools used for medical image analysis

ML and DL techniques used for medical imaging

DT  Decision Tree, AB  Adaptive Boosting, ANN  Artificial Neural Network, CNN  Convolutional Neural Network, SVM  Support Vector Machine, RF  Random Forest, ML  Machine Learning, TL  Transfer Learning, PCA  Principal Component Analysis, ICA  Independent Component Analysis.

Tools used for medical images

Figure ​ Figure4 4 depicts the percentage of various tools (Table ​ (Table5) 5 ) used in the primary studies for the implementation of ML and DL models where MATLAB and NumPy have the percentage of 38 and 37, respectively, which signify the popularity of these tools among researchers. R and TensorFlow are the second most used tools with a percentage of 13 and 12, respectively.

Tool description

Techniques used for medical images

This subsection includes the description and identification of the most common ML and DL techniques (i) used for disease classification, detection and diagnosis, (ii) based on type of disease, and (iii) used for EEG and MEG data processing.

Description of techniques

• Convolutional Layer: It is responsible to apply the filters systematically to create feature maps for summarizing features present in the input image.
• Pooling Layer: It is used for ordering the repeated layers in a model. It operates on each feature map, received from the convolutional layer, to produce a new set of feature maps pooled together. Pooling operation is used to reduce the feature map size with required pixels or values in each feature map, hence, reducing the overfitting problem. It consists of two main functions namely, average pooling and maximum pooling.
• Fully-Connected Layer: It is simply the feed-forward neural network where input is received from the final pooling layer. Based on the extracted features, a fully connected layer predicts the image class.

CNN architecture

ANN architecture

TL architecture

RF architecture

DT architecture

SVM architecture

ML and DL techniques

Table ​ Table6 6 summarizes ML and DL techniques such as Naïve bayes [ 43 , 69 ], KNN [ 6 ], DTs [ 36 , 48 ], neural networks, and SVM [ 59 , 73 , 90 ] which are used for medical imaging in primary studies. Here, column 1 represents articles and row 1 represents various techniques. Further, “+” and “-” signify that the article does and does not support the corresponding technique, respectively. The most reliable ML and DL techniques based on the type of disease are shown in Table ​ Table7. 7 . The most significant ML and DL techniques for EEG and MEG data processing are shown in Table ​ Table8 8 .

ML and DL techniques based on the type of disease

ML  Machine Learning, DL  Deep Learning, RF  Random Forest, CNN  Convolutional Neural Network, SVM  Support Vector Machine, DNN  Deep Neural Network, LR  Linear Regression, TL  Transfer Learning

ML and DL techniques for EEG and MEG data processing

SVM  Support Vector Machine, LS-SVM  Least Square-SVM, FD  Fractal Dimensions, RF  Random Forest, KNN  K-Nearest Neighbor, SCNN  Spatial Summary Convolutional Neural Network, Ra-SCNN  SCNN model augmented with attention focused Recurrence, ANN  Artificial Neural Network, DNN  Deep Neural Network, LSTM  Long Short-Term Memory Networks, CNN-RNN  Convolutional Neural Network-Recurrent Neural Network, MLP  Multi-Layer Perceptron, EEG  Electroencephalogram, MEG  Magnetoencephalography

Dataset description

Following section addresses the research question 4 (refer subsection 3.1 ) by providing the details of the datasets used in primary studies for implementing ML and DL algorithms. Table ​ Table9 9 summarizes the description of dataset(s) such as MRI, X-Rays, lesion data, infra-red images and CT-Scan. The accessibility to a dataset is divided as (i) public (available at online repositories), and (ii) own created (created by the authors).

MRI  Magnetic Resonance Imaging, CT  Computer Tomography, DCE  Dynamic Contrast-Enhanced, PET  Positron Emission Tomography. Dermatology dataset (32): 32 features from dermatology dataset, Heart-C dataset (15): 15 features from Heart-C dataset, Lung cancer dataset (55): 55 features from lung cancer dataset, Pima Indian dataset (9): 9 features from Pima Indian dataset, Hepatitis dataset (18): 18 features from hepatitis dataset, Iris dataset (5): 5 features from iris dataset, Wisconsin cancer dataset (10): 10 features from Wisconsin cancer dataset, Lympho dataset (17): 17 features from Lympho dataset, Diabetes dataset (8): 8 features from diabetes dataset, Stalog disease dataset (12): 12 features from stalog disease dataset

Experimental description

Research question 5 (refer subsection 3.1 ) is addressed in this section. MRI dataset is used for the experiments to show the comparative analysis of ML classifiers and DL models. Dataset¹ description and experimental setup are discussed in subsections 8.1  and 8.2 , respectively. Similarly, the methodology and results are discussed in subsections 8.3  and 8.4 , respectively.

The experiments to classify the brain tumor include the publicly available tumor dataset. ( https://www.kaggle.com/datasets/masoudnickparvar/brain-tumor-mri-dataset ). The MRI dataset contains the 711 images of meningioma tumor and no tumor. Dataset is divided into two parts: testing and training with different image resolutions.

Experimental setup

The whole series of experiments were performed on a 64-bit computer with an Intel(R) 221 Core(TM) i3-10110U CPU @ 2.10 GHz 2.59 GHz, 8GB RAM. To train and validate the model, code was implemented in python language in Google colab platform.

Methodology

Figure ​ Figure11 11 depicts the methodology used in the experiments for disease classification. It is described as follows:

Methodology used for disease prediction

• Import dataset : Dataset¹ is retrieved from the public website which is divided into two categories namely: no tumor and meningioma tumor. The dimensions of images given in the dataset were different from one another, which was further resized to 200 × 200.
• Label dataset : Dataset is labeled in the form of 0 and 1, where 0 and 1 indicate the data having no tumor and data having meningioma tumor, respectively.
• Split dataset : Further, the dataset is splitted in the ratio of 80:20 for training (80%) and testing (20%) dataset.
• Feature scaling and feature selection : ML algorithms work on numbers without knowing what the number represents. Feature scaling helps to resolve the given problem by scaling the features into a specific defined range, so that one feature does not dominate the other one. In this experiment, PCA technique is used to reduce the feature count and select the required features.
• Apply ML classifiers : For this experiment, ML classifiers (SVM, RF, DT, LR) and DL models (CNN, ResNet50V2) are used, which further classified the dataset into two categories i.e., 0 and 1.
• Prediction and testing the model : The model was tested with testing data (20% of the dataset) and predicted the disease accurately for the given dataset.

Publisher by number of citations for a  ML articles b  DL articles

This subsection discusses the results obtained by ML classifiers as shown in Fig.  12 ; Table ​ Table10. 10 . In Fig.  12a , ​ ,b, b , ​ ,c, c , and ​ andd d illustrate the confusion matrix obtained from SVM, LR, RF, and DT, respectively. Table ​ Table10 10 shows the values of accuracy obtained after implementing the considered ML classifiers and DL models for the MRI dataset. The results show that CNN and RF have better accuracy with 97.6% and 96.93%, respectively.

Confusion matrix for a  SVM, b  LR, c  RF, and d  DT

Accuracy results for MRI images using ML classifier/DL model

SVM  Support Vector Machine, LR  Logistic Regression, RF  Random Forest, DT  Decision Tree, CNN  Convolutional Neural Network.

Analytical discussion

The primary studies were analyzed based on the publisher citation count, year wise publications, keywords, various diseases, techniques, imaging modalities and type of publication.

Publisher by citations

A schematic view of the influential publishers in the concerned domain is presented by the citations of the articles published in it. Figure ​ Figure13 13 shows all the publishers considered for this review in between 2014 and 2022. Moreover, it depicts the number of citations of ML and DL articles with respect to the publishers in Fig.  14a and ​ andb, b , respectively. Due to many types of indexing procedures along with time, there is a variation in the count of citations in Google Scholar. It was observed that most of the articles for ML and DL were published in ScienceDirect and IEEE publishers with the maximum citation 2425 and 42,866, respectively.

Year wise publication of ML and DL in healthcare

Scholarly articles published between 2014 and 2022

In this subsection, Fig.  14 depicts that out of 40 primary studies, the most published articles for ML were from the year 2020 with a count of 10, which is equivalent to 25% of the total. Followed by the year 2021 with 8 (20%), 2019 with 6 (15%), and 2022 with 4 (10%). Other years like 2017 and 2018 have the same count of 3 with 7.5%, 2014 and 2016 have the same count of 2 with 5%, and 2015 has the count of 1 with 2.5%. Thus, it can be observed that the maximum number of articles for primary study was considered from the year 2020 and minimum from 2014 to 2017.

Most commonly used keywords in the primary studies

Word cloud is the simple way to identify the relevant terms and themes utilized in the referenced research articles. Figure ​ Figure15 15 depicts the word cloud which represents larger font for the most often used keywords and smaller font for less frequent keywords.

Word cloud for frequently used keywords

Disease types

Figure ​ Figure16 16 depicts the percentage of multiple diseases diagnosed in the primary studies. As observed, breast disease is the most common disease with the highest percentage (21%) among all. Brain tumor took the second place (18%) followed by diabetes (16%) and lung disease (16%). Also, other diseases such as eye, liver, skin, hepatitis and cancer were diagnosed using various techniques.

Percentage of diseases

Publication by techniques and statistical analysis of techniques

It was observed that researchers have used multiple techniques to attain better results as shown in Table ​ Table5. 5 . For classification, ML classifiers like SVM, RF and Naïve bayes were combinedly used for the same. Detection was performed using neural networks such as ANN or CNN, and TL was performed frequently due to its capability of breaking down the large datasets. Figure ​ Figure17 17 depicts the percentage of various techniques used in primary studies. It summarizes that SVM (20%) is the most widely used technique for medical image classification.

Percentage of ML and DL techniques in healthcare

The statistical analysis of ML and DL techniques for medical diagnosis is represented in Fig.  18 .

Statistical analysis of ML and DL techniques for medical diagnosis

Imaging modalities

Figure ​ Figure19 19 demonstrates the multiple image modalities used for the evaluation of medical images. However, MRI/X-Ray dominates the subject area with 45%. The second most used modality is CT-Scan (30%), followed by mammogram (10%) and I-Scan-2 (10%). Moreover, to automate the process of retrieving and analyzing the features, computer modalities such as CAD was included for the detection of hepatitis and cancer [ 55 , 60 ].

Percentage of modalities used in medical imaging

Type of publication

Figure ​ Figure20 20 illustrates the distribution of articles according to the type of publications considered for this review. Majority of the articles were considered from journals with 70%, book Chaps. (8%), conference proceeding papers (7%), workshop articles (2%) and others (13%) including the society articles, online database articles, articles from publications like Bentham Science, springer archives and the transcript.

Percentage of type of publication

From this study, it was observed that the variability in the literature occurred due to uncertainty of the evaluated data and models (refer Fig.  21 ). Data uncertainty was caused due to the multiple sources such as transmission noise, missing values and measurement noise. Whereas, model uncertainty was observed due to the less understanding of architecture and prediction of future data with parameters. The observed uncertainty was helpful to attain different results with various methods. Recently, many advanced technologies were introduced to attain enormous amounts of raw data in different scenarios.

Data and model uncertainties

Further, while reviewing the literature, it has been observed that focusing on every aspect of data (noisy or clear) is important as it impacts the results. The utilization of an appropriate algorithm to analyze images can be used for increasing the success ratio. Thus, variation in expected standard results is due to the use of raw data which may incorporate a certain amount of noise (refer Fig.  22 ). CNN is not much sensitive to the noise due to which it can extract information from noisy data [ 44 ]. Moreover, Hermitian basis functions were used for extracting the accumulated data from the ECG data which reduce the effects of Gaussian noise.

Noisy data [ 11 ]

Therefore, dealing with the uncertainty of data and models with ML along with DL techniques is the most important issue to be handled by researchers. These techniques are useful for obtaining accurate and better results for decision making in every respective domain [ 2 , 3 ,  45 , 64 , 75 , 93 ]. Therefore, there is a need to deal with the variance in ML and DL algorithms such as RF, Rubber Sheet Normalization, DT, bagging-boosting, ANN, CNN, SVM, TL, Bayes Net, and GLCM. Further, such strategies can be used to deal with ambiguity in medical data for achieving high performance. Based on this review, it has been observed that medical professionals may be able to treat tumors promptly if they are identified early.

Conclusions and future work

This study provides an overview of various ML and DL approaches for the disease diagnosis along with classification, imaging modalities, tools, techniques, datasets and challenges in the medical domain. MRI and X-Ray scans are the most commonly used modalities for the disease diagnosis. Further, among all the tools and techniques studied, MATLAB and SVM dominated, respectively. It was observed that MRI dataset is widely used by researchers. Also, a series of experiments using MRI dataset has provided a comparative analysis of ML classifiers and DL models where CNN (97.6%) and RF (96.93%) have outperformed other algorithms. This study indicates that there is a need to include denoising techniques with DL models in the healthcare domain. It also concludes that various classical ML and DL techniques are extensively applied to deal with data uncertainty. Due to the superior performance, DL approaches have recently become quite popular among researchers. This review will assist healthcare community, physicians, clinicians and medical practitioners to choose an appropriate ML and DL technique for the diagnosis of disease with reduced time and high accuracy.

Future work will incorporate DL approaches for the diagnosis of all diseases considering noise removal from any given dataset. The additional aspects and properties of DL models for medical images can be explored. To increase the accuracy, enormous amount of data is required, therefore, the potential of the model should be improved to deal with large datasets. Also, different data augmentation techniques along with required features of the dataset can be explored to attain better accuracy.

Data availability

Declarations.

The authors declare that they have no conflict of interest.

Publisher’s note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Contributor Information

Meghavi Rana, Email: [email protected] .

Megha Bhushan, Email: [email protected] .

Review: Developments and challenges of advanced flexible electronic materials for medical monitoring applications

• Published: 10 September 2024
• Volume 7 , article number  141 , ( 2024 )

Cite this article

• Tao Zeng 1 ,
• Yufeng Wu 2 &
• Ming Lei 3  

Flexible sensors, made from flexible electronic materials, are of great importance in the medical field due to the rising prevalence of cardiovascular and cerebrovascular diseases. Studies have demonstrated that timely diagnosis and continuous monitoring of relevant physiological signals can be beneficial in preventing such conditions. Although traditional rigid monitoring sensors are still widely used for medical monitoring, the EMG, ECG, and EEG signals they obtain are often significantly affected by motion artifacts and noise. Therefore, the significance of wearable smart monitoring devices based on flexible electronic materials cannot be overstated. Numerous researchers have been working tirelessly for this purpose, exploring solutions from various angles, including material choice, circuit design, and algorithmic processing. This paper begins by analyzing the causes of motion artifacts in medical smart monitoring devices. Next, it introduces the application of flexible materials and flexible electronic materials in several aspects, along with the work of some representative flexible sensors. Following this, it discusses materials selection and device designs (e.g., accelerometers, gyroscopes, differential circuits, etc.) and algorithmic approaches for eliminating motion artifacts. Finally, an outlook on motion artifact removal techniques from the perspectives of more in-depth material development, structural design, and machine learning is provided. The purpose of this paper is to offer a comprehensive overview of current motion artifact removal techniques and materials, aiming to encourage further research and effectively address the key problem of signal acquisition accuracy in smart biomonitoring.

Graphical Abstract

TOC: Motion artifact occurrence state [40, 120]. Two methods of motion artifact removal or attenuation states are now commonly used: device design [30, 109] and algorithm development [115]. Future development focusing on machine learning and AI. [136]

This is a preview of subscription content, log in via an institution to check access.

Access this article

Subscribe and save.

• Get 10 units per month
• Download Article/Chapter or eBook
• 1 Unit = 1 Article or 1 Chapter
• Cancel anytime

Price includes VAT (Russian Federation)

Instant access to the full article PDF.

Rent this article via DeepDyve

Institutional subscriptions

Explore related subjects

Data availability.

No datasets were generated or analysed during the current study.

Vasan Ramachandran S, Larson Martin G, Leip Eric P, Evans Jane C, O’Donnell Christopher J, Kannel William B, Levy D (2001) Impact of high-normal blood pressure on the risk of cardiovascular disease. N Engl J Med 345:1291–1297. https://doi.org/10.1056/NEJMoa003417

Article   Google Scholar  

Dalen JE, Alpert JS, Goldberg RJ, Weinstein RS (2014) The epidemic of the 20th century: coronary heart disease. Am J Med 127:807–812. https://doi.org/10.1016/j.amjmed.2014.04.015

Article   PubMed   Google Scholar  

DeVon HA, Vuckovic K, Ryan CJ, Barnason S, Zerwic JJ, Pozehl B, Schulz P, Seo Y, Zimmerman L (2017) Systematic review of symptom clusters in cardiovascular disease. Eur J Cardiovasc Nurs 16:6–17. https://doi.org/10.1177/1474515116642594

Lu L-Q, Tian J, Luo X-J, Peng J (2021) Targeting the pathways of regulated necrosis: a potential strategy for alleviation of cardio-cerebrovascular injury. Cell Mol Life Sci 78:63–78. https://doi.org/10.1007/s00018-020-03587-8

Article   PubMed   CAS   Google Scholar  

Luo M-Y, Su J-H, Gong S-X, Liang N, Huang W-Q, Chen W, Wang A-P, Tian Y (2021) Ferroptosis: new dawn for overcoming the cardio-cerebrovascular diseases. Front Cell Dev Biol 9:733908. https://doi.org/10.3389/fcell.2021.733908

Article   PubMed   PubMed Central   Google Scholar  

Yu J-N, Wu B-B, Yang J, Lei X-L, Shen W-Q (2020) Cardio-cerebrovascular disease is associated with severity and mortality of COVID-19: a systematic review and meta-analysis. Biol Res Nurs 23:258–269. https://doi.org/10.1177/1099800420951984

Zhang P, Dong G, Sun B, Zhang L, Chen X, Ma N, Yu F, Guo H, Huang H, Lee YL, Tang N, Chen J (2011) Long-term exposure to ambient air pollution and mortality due to cardiovascular disease and cerebrovascular disease in Shenyang. China PLoS ONE 6:e20827. https://doi.org/10.1371/journal.pone.0020827

Meng K, Xiao X, Wei W, Chen G, Nashalian A, Shen S, Xiao X, Chen J (2022) Wearable pressure sensors for pulse wave monitoring. Adv Mater 34:2109357. https://doi.org/10.1002/adma.202109357

Article   CAS   Google Scholar  

Lin G, Yuhui W, Yifei W, Gang W, Quan Z, Jianbao Z, Xiang C, Xiangguo Y (2022) Hybrid motion artifact detection and correction approach for functional near-infrared spectroscopy measurements. J Biomed Opt 27:025003. https://doi.org/10.1117/1.JBO.27.2.025003

Ray D, Collins T, Woolley SI, Ponnapalli PVS (2023) A review of wearable multi-wavelength photoplethysmography. IEEE Rev Biomed Eng 16:136–151. https://doi.org/10.1109/RBME.2021.3121476

Gibbs P, Asada HH (2005) Reducing motion artifact in wearable biosensors using mems accelerometers for active noise cancellation. In: Proceedings of the 2005, American control conference, Portland, OR, USA, pp 1581–1586. https://doi.org/10.1109/ACC.2005.1470193

Kumar A, Komaragiri R, Kumar M (2018) From pacemaker to wearable: techniques for ECG detection systems. J Med Syst 42:34. https://doi.org/10.1007/s10916-017-0886-1

McGeehan M, Hahn ME, Karipott S, Ong KG (2022) A wearable shear force transducer based on color spectrum analysis. Meas Sci Technol 34:015106. https://doi.org/10.20944/preprints202202.0294.v1

Zhang Z, Yu Y, Yu H, Feng Y, Feng W (2022) Water-resistant conductive organogels with sensation and actuation functions for artificial neuro-sensory muscular systems. SmartMat 3:632–643. https://doi.org/10.1002/smm2.1108

Lee J, Kim M, Park HK, Kim IY (2020) Motion artifact reduction in wearable photoplethysmography based on multi-channel sensors with multiple wavelengths. Sensors 20:1493. https://doi.org/10.3390/s20051493

Eladio AC, Miguel BZ, Daniel CG, Marco RC, Roberto LA (2021) Removal of ECG baseline wander using a resting cycle template. Eng Proc 10:45. https://doi.org/10.3390/ecsa-8-11289

Zhou K, Dai K, Liu C, Shen C (2020) Flexible conductive polymer composites for smart wearable strain sensors. SmartMat 1:e1010. https://doi.org/10.1002/smm2.1010

Kwon S, Kwon Y-T, Kim Y-S, Lim H-R, Mahmood M, Yeo W-H (2020) Skin-conformal, soft material-enabled bioelectronic system with minimized motion artifacts for reliable health and performance monitoring of athletes. Biosens Bioelectron 151:111981. https://doi.org/10.1016/j.bios.2019.111981

Yiu C, Liu Y, Zhang C, Zhou J, Jia H, Wong TH, Huang X, Li J, Yao K, Yau MK, Zhao L, Li H, Zhang B, Park W, Zhang Y, Wang Z, Yu X (2022) Soft, stretchable, wireless intelligent three-lead electrocardiograph monitors with feedback functions for warning of potential heart attack. SmartMat 3:668–684. https://doi.org/10.1002/smm2.1114

Strasser F, Muma M, Zoubir AM (2012) Motion artifact removal in ECG signals using multiresolution thresholding. In: 2012 proceedings of the 20th european signal processing conference (EUSIPCO), Bucharest, Romania, pp 899–903. https://ieeexplore.ieee.org/abstract/document/6333855

Wang C, He K, Li J, Chen X (2021) Conformal electrodes for on-skin digitalization SmartMat 2:252–262. https://doi.org/10.1002/smm2.1068

Lilienthal J, Dargie W (2021) Spectral characteristics of motion artifacts in wireless ECG and their correlation with reference motion sensors. In: 2021 43rd annual international conference of the IEEE engineering in medicine & biology society (EMBC), Mexico, pp 517–521. https://doi.org/10.1109/EMBC46164.2021.9630394

Ding Q, Wang H, Zhou Z, Wu Z, Tao K, Gui X, Liu C, Shi W, Wu J (2023) Stretchable, self-healable, and breathable biomimetic iontronics with superior humidity-sensing performance for wireless respiration monitoring. SmartMat 4:e1147. https://doi.org/10.1002/smm2.1147

Lin S, Liu J, Li W, Wang D, Huang Y, Jia C, Li Z, Murtaza M, Wang H, Song J, Liu Z, Huang K, Zu D, Lei M, Hong B, Wu H (2019) A flexible, robust, and gel-free electroencephalogram electrode for noninvasive brain-computer interfaces. Nano Lett 19:6853–6861. https://doi.org/10.1021/acs.nanolett.9b02019

Liu J, Lin S, Li W, Zhao Y, Liu D, He Z, Wang D, Lei M, Hong B, Wu H (2022) Ten-hour stable noninvasive brain-computer interface realized by semidry hydrogel-based electrodes. Research 2022. https://doi.org/10.34133/2022/9830457

Tian L, Zimmerman B, Akhtar A, Yu KJ, Moore M, Wu J, Larsen RJ, Lee JW, Li J, Liu Y, Metzger B, Qu S, Guo X, Mathewson KE, Fan JA, Cornman J, Fatina M, Xie Z, Ma Y, Zhang J, Zhang Y, Dolcos F, Fabiani M, Gratton G, Bretl T, Hargrove LJ, Braun PV, Huang Y, Rogers JA (2019) Large-area MRI-compatible epidermal electronic interfaces for prosthetic control and cognitive monitoring. Nat Biomed Eng 3:194–205. https://doi.org/10.1038/s41551-019-0347-x

Wang C, Wang C, Huang Z, Xu S (2018) Materials and structures toward soft electronics. Adv Mater 30:1801368. https://doi.org/10.1002/adma.201801368

Servati A, Zou L, Wang ZJ, Ko F, Servati P (2017) Novel flexible wearable sensor materials and signal processing for vital sign and human activity monitoring. Sensors 17:1622. https://doi.org/10.3390/s17071622

Maity AK, Veeraraghavan A, Sabharwal A (2022) PPGMotion: model-based detection of motion artifacts in photoplethysmography signals. Biomed Signal Proces 75:103632. https://doi.org/10.1016/j.bspc.2022.103632

Ottenbacher J, Kirst M, Jatoba L, Huflejt M, Grossmann U, Stork W (2008) Reliable motion artifact detection for ecg monitoring systems with dry electrodes. In: 2008 30th annual international conference of the IEEE engineering in medicine and biology society, Vancouver, BC, Canada, pp 1695–1698. https://doi.org/10.1109/IEMBS.2008.4649502

Dao D, Salehizadeh SMA, Noh Y, Chong JW, Cho CH, McManus D, Darling CE, Mendelson Y, Chon KH (2017) A robust motion artifact detection algorithm for accurate detection of heart rates from photoplethysmographic signals using time–frequency spectral features. IEEE J Biomed Health 21:1242–1253. https://doi.org/10.1109/JBHI.2016.2612059

Suraj H, Takashi S (2021) Motion robust remote photoplethysmography via frequency domain motion artifact reduction. In: 2021 IEEE biomedical circuits and systems conference (BioCAS), Berlin, Germany, pp 1–4. https://doi.org/10.1109/BioCAS49922.2021.9644650

Qureshi R, Uzair M, Khurshid K (2017) Multistage adaptive filter for ECG signal processing. In: 2017 international conference on communication, computing and digital systems (C-CODE), Islamabad, Pakistan, pp 363–368. https://doi.org/10.1109/C-CODE.2017.7918958

Cömert A, Hyttinen J (2015) Investigating the possible effect of electrode support structure on motion artifact in wearable bioelectric signal monitoring. Biomed Eng Online 14:44. https://doi.org/10.1186/s12938-015-0044-2

Yang X, Zhang M, Gong S, Sun M, Xie M, Niu P, Pang W (2023) Chest-laminated and sweat-permeable e-skin for speech and motion artifact-insensitive cough detection. Adv Mater Technol 8:2201043. https://doi.org/10.1002/admt.202201043

Nordin AD, Hairston WD, Ferris DP (2018) Dual-electrode motion artifact cancellation for mobile electroencephalography. J Neural Eng 15:056024. https://doi.org/10.1088/1741-2552/aad7d7

Sirtoli VG, Liamini M, Lins LT, Lessard-Tremblay M, Cowan GER, Zednik RJ, Gagnon G (2023) Removal of motion artifacts in capacitive electrocardiogram acquisition: a review. IEEE T Biomed Circ S 17:394–412. https://doi.org/10.1109/TBCAS.2023.3270661

Wartzek T, Lammersen T, Eilebrecht B, Walter M, Leonhardt S (2011) Triboelectricity in capacitive biopotential measurements. IEEE Trans Biomed Eng 58:1268–1277. https://doi.org/10.1109/TBME.2010.2100393

Ottenbacher J, Jatoba L, Großmann U, Stork W, Müller-Glaser K (2006) ECG electrodes for a context-aware cardiac permanent monitoring system. In: World congress on medical physics and biomedical engineering 2006, Berlin, Heidelberg, pp 672–675. https://doi.org/10.1007/978-3-540-36841-0_156

Xu J, Büsze B, Hoof CV, Makinwa KAA, Yazicioglu RF (2015) A 15-channel digital active electrode system for multi-parameter biopotential measurement. IEEE J of Solid-St Circ 50:2090–2100. https://doi.org/10.1109/JSSC.2015.2422798

Zhu W, Ma G, Chen Y, Qiu L, Zheng L, Wang L (2021) A novel method for detecting noise segments in ECG signals. In: 2021 IEEE 20th international conference on trust, security and privacy in computing and communications (TrustCom), Shenyang, China, pp 1570–1574. https://doi.org/10.1109/TrustCom53373.2021.00228

Yang C, Tavassolian N (2019) An independent component analysis approach to motion noise cancelation of cardio-mechanical signals. IEEE Trans Biomed Eng 66:784–793. https://doi.org/10.1109/TBME.2018.2856700

Shimazaki T, Hara S, Okuhata H, Nakamura H, Kawabata T (2014) Cancellation of motion srtifact induced by exercise for PPG-based heart rate sensing. In: 2014 36th annual international conference of the IEEE engineering in medicine and biology society, Chicago, IL, USA, pp 3216–3219. https://doi.org/10.1109/EMBC.2014.6944307

Helleputte NV, Kim S, Kim H, Kim JP, Hoof CV, Yazicioglu RF (2012) A 160 um biopotential acquisition IC with fully integrated IA and motion artifact suppression. IEEE Trans Biomed Circuits Syst 6:552–561. https://doi.org/10.1109/TBCAS.2012.2224113

Xie X, Liu H, Shu M, Zhu Q, Huang A, Kong X, Wang Y (2021) A multi-stage denoising framework for ambulatory ECG signal based on domain knowledge and motion artifact detection. Future Gener Comp Sy 116:103–116. https://doi.org/10.1016/j.future.2020.10.024

Fakroul RH, Lykourgos P, John S, Sairul IS (2012) Wavelet based motion artifact removal for ECG signals. In: 2012 IEEE-EMBS conference on biomedical engineering and sciences, Langkawi, Malaysia, pp 339–342. https://doi.org/10.1109/IECBES.2012.6498019

Mithun P, Pandey Prem C, Sebastian Toney, Mishra Prashant, Pandey Vinod K (2011) A wavelet based technique for suppression of EMG noise and motion artifact in ambulatory ECG. In: 2011 annual international conference of the IEEE engineering in medicine and biology society, Boston, MA, USA, pp 7087–7090. https://doi.org/10.1109/IEMBS.2011.6091791

Richer N, Downey RJ, Hairston WD, Ferris DP, Nordin AD (2020) Motion and muscle artifact removal validation using an electrical head phantom, robotic motion platform, and dual layer mobile EEG. IEEE Trans Neural Syst Rehabil Eng 28:1825–1835. https://doi.org/10.1109/TNSRE.2020.3000971

Berset T, Geng D, Romero I (2012) An optimized DSP implementation of adaptive filtering and ICA for motion artifact reduction in ambulatory ECG monitoring. In: 2012 annual international conference of the IEEE engineering in medicine and biology society, San Diego, CA, USA, pp 6496–6499. https://doi.org/10.1126/scitranslmed.aay4

Jayakumar A, Jose VK, Lee J-M (2020) Hydrogels for medical and environmental applications. Small Methods 4:1900735. https://doi.org/10.1002/smtd.201900735

Poustchi F, Amani H, Ahmadian Z, Niknezhad SV, Mehrabi S, Santos HA, Shahbazi M-A (2021) Combination therapy of killing diseases by injectable hydrogels: from concept to medical applications. Adv Healthc Mater 10:2001571. https://doi.org/10.1002/adhm.202001571

Yang KR, Han Q, Chen B, Zheng Y, Zhang K, Li Q, Wang JC (2018) Antimicrobial hydrogels: promising materials for medical application. Int J Nanomedicine 13:2217–2263. https://doi.org/10.2147/IJN.S154748

Article   PubMed   PubMed Central   CAS   Google Scholar  

Teng Y, Li S, Tang H, Tao X, Fan Y, Huang Y (2023) Medical applications of hydrogels in skin infections: a review. Infect Drug Resist 16:391–401. https://doi.org/10.2147/idr.S396990

Liang Y, He J, Guo B (2021) Functional hydrogels as wound dressing to enhance wound healing. ACS Nano 15:12687–12722. https://doi.org/10.1021/acsnano.1c04206

Xu Z, Han S, Gu Z, Wu J (2020) Advances and impact of antioxidant hydrogel in chronic wound healing. Adv Healthc Mater 9:1901502. https://doi.org/10.1002/adhm.201901502

Zhao Z, Wang Z, Li G, Cai Z, Wu J, Wang L, Deng L, Cai M, Cui W (2021) Injectable microfluidic hydrogel microspheres for cell and drug delivery. Adv Funct Mater 31:2103339. https://doi.org/10.1002/adfm.202103339

Zhao X, Liu S, Yildirimer L, Zhao H, Ding R, Wang H, Cui W, Weitz D (2016) Injectable stem cell-laden photocrosslinkable microspheres fabricated using microfluidics for rapid generation of osteogenic tissue constructs. Adv Funct Mater 26:2809–2819. https://doi.org/10.1002/adfm.201504943

Fiorica C, Mauro N, Pitarresi G, Scialabba C, Palumbo FS, Giammona G (2017) Double-network-structured graphene oxide-containing nanogels as photothermal agents for the treatment of colorectal cancer. Biomacromol 18:1010–1018. https://doi.org/10.1021/acs.biomac.6b01897

Liu J, Qu M, Wang C, Xue Y, Huang H, Chen Q, Sun W, Zhou X, Xu G, Jiang X (2022) A dual-cross-linked hydrogel patch for promoting diabetic wound healing. Small 18:2106172. https://doi.org/10.1002/smll.202106172

Wang M, Ma C, Uzabakiriho PC, Chen X, Chen Z, Cheng Y, Wang Z, Zhao G (2021) Stencil printing of liquid metal upon electrospun nanofibers enables high-performance flexible electronics. ACS Nano 15:19364–19376. https://doi.org/10.1021/acsnano.1c05762

Zhang D, Huang T, Duan L (2020) Emerging self-emissive technologies for flexible displays. Adv Mater 32:1902391. https://doi.org/10.1002/adma.201902391

Koo JH, Kim DC, Shim HJ, Kim T-H, Kim D-H (2018) Flexible and stretchable smart display: materials, fabrication, device design, and system integration. Adv Funct Mater 28:1801834. https://doi.org/10.1002/adfm.201801834

Wei XY, Wang HL, Wang Y, Kuang SY, Zhu XX, Zou J, Wang L, Zeng X, Rao F, Zhu G (2019) Fully-integrated motion-driven electroluminescence enabled by triboelectrification for customized flexible display. Nano Energy 61:158–164. https://doi.org/10.1016/j.nanoen.2019.04.005

Su R, Park SH, Ouyang X, Ahn SI, McAlpine MC (2022) 3D-printed flexible organic light-emitting diode displays. Sci Adv 8:eabl8798. https://doi.org/10.1126/sciadv.abl8798

Zhu Y, Yang M, Huang Q, Wang D, Yu R, Wang J, Zheng Z, Wang D (2020) V 2 O 5 textile cathodes with high capacity and stability for flexible lithium-ion batteries. Adv Mater 32:1906205. https://doi.org/10.1002/adma.201906205

Chao D, Lai C-H, Liang P, Wei Q, Wang Y-S, Zhu C, Deng G, Doan-Nguyen VVT, Lin J, Mai L, Fan HJ, Dunn B, Shen ZX (2018) Sodium vanadium fluorophosphates (NVOPF) array cathode designed for high-rate full sodium ion storage device. Adv Energy Mater 8:1800058. https://doi.org/10.1002/aenm.201800058

Yang M, Shu X, Pan W, Zhang J (2021) Toward flexible zinc–air batteries with self-supported air electrodes. Small 17:2006773. https://doi.org/10.1002/smll.202006773

Dong H, Li J, Guo J, Lai F, Zhao F, Jiao Y, Brett DJL, Liu T, He G, Parkin IP (2021) Insights on flexible zinc-ion batteries from lab research to commercialization. Adv Mater 33:2007548. https://doi.org/10.1002/adma.202007548

Sidheekha MP, Shabeeba A, Rajan L, Thayyil MS, Ismail YA (2023) Conducting polymer/hydrogel hybrid free-standing electrodes for flexible supercapacitors capable of self-sensing working conditions: large-scale fabrication through facile and low-cost route. Eng Sci 23:890. https://doi.org/10.30919/es890

Xue L, Savilov SV, Lunin VV, Xia H (2018) Self-standing porous LiCoO 2 nanosheet arrays as 3D Cathodes for flexible Li-Ion batteries. Adv Funct Mater 28:1705836. https://doi.org/10.1002/adfm.201705836

Suryavanshi VB, Bobade RG, Lokh BJ, Ambare RC (2023) Electro-synthesized bismuth oxide nanomaterials on flexible substrate electrode for supercapacitor application. ES Energy Environ 21:944. https://doi.org/10.30919/esee944

Zhang X, Li Z, Luo L, Fan Y, Du Z (2022) A review on thermal management of lithium-ion batteries for electric vehicles. Energy 238:121652. https://doi.org/10.1016/j.energy.2021.121652

Zhao S, Zuo Y, Liu T, Zhai S, Dai Y, Guo Z, Wang Y, He Q, Xia L, Zhi C, Bae J, Wang K, Ni M (2021) Multi-functional hydrogels for flexible zinc-based batteries working under extreme conditions. Adv Energy Mater 11:2101749. https://doi.org/10.1002/aenm.202101749

Zhang P, Wang K, Pei P, Zuo Y, Wei M, Liu X, Xiao Y, Xiong J (2021) Selection of hydrogel electrolytes for flexible zinc–air batteries. Mater Today Chem 21:100538. https://doi.org/10.1016/j.mtchem.2021.100538

Fan X, Liu J, Song Z, Han X, Deng Y, Zhong C, Hu W (2019) Porous nanocomposite gel polymer electrolyte with high ionic conductivity and superior electrolyte retention capability for long-cycle-life flexible zinc–air batteries. Nano Energy 56:454–462. https://doi.org/10.1016/j.nanoen.2018.11.057

Liu L, Liu Y, Tang R, Ai J, Ma Y, Chen Y, Feng X (2022) Stable and low-resistance polydopamine methacrylamide-polyacrylamide hydrogel for brain-computer interface. Sci China Mater 65:2298–2308. https://doi.org/10.1007/s40843-022-2145-3

Mahmood M, Kim N, Mahmood M, Kim H, Kim H, Rodeheaver N, Sang M, Yu KJ, Yeo W-H (2022) VR-enabled portable brain-computer interfaces via wireless soft bioelectronics. Biosens Bioelectron 210:114333. https://doi.org/10.1016/j.bios.2022.114333

Tang X, Shen H, Zhao S, Li N, Liu J (2023) Flexible brain–computer interfaces. Nat Electron 6:109–118. https://doi.org/10.1038/s41928-022-00913-9

Wu N, Wan S, Su S, Huang H, Dou G, Sun L (2021) Electrode materials for brain–machine interface: A review. InfoMat 3:1174–1194. https://doi.org/10.1002/inf2.12234

Yao C, Zhao W, Liu L, Liu Q, Li J (2023) Flexible, programable sensing system with poly(AAm-HEMA-SA) for human motion detection. ES Mater Manuf 20:818. https://doi.org/10.30919/esmm5f818

Kim JU, Seo SG, Rogers JA (2023) Compound semiconductor devices for the skin. Nat Mater 22:155–156. https://doi.org/10.1038/s41563-022-01441-9

Lee K, Ni X, Lee JY, Arafa H, Pe DJ, Xu S, Avila R, Irie M, Lee JH, Easterlin RL, Kim DH, Chung HU, Olabisi OO, Getaneh S, Chung E, Hill M, Bell J, Jang H, Liu C, Park JB, Kim J, Kim SB, Mehta S, Pharr M, Tzavelis A, Reeder JT, Huang I, Deng Y, Xie Z, Davies CR, Huang Y, Rogers JA (2020) Mechano-acoustic sensing of physiological processes and body motions via a soft wireless device placed at the suprasternal notch. Nat Biomed Eng 4:148–158. https://doi.org/10.1038/s41551-019-0480-6

Yoo S, Yang T, Park M, Jeong H, Lee YJ, Cho D, Kim J, Kwak SS, Shin J, Park Y, Wang Y, Miljkovic N, King WP, Rogers JA (2023) Responsive materials and mechanisms as thermal safety systems for skin-interfaced electronic devices. Nat Commun 14:1024. https://doi.org/10.1038/s41467-023-36690-y

Song E, Chiang C-H, Li R, Jin X, Zhao J, Hill M, Xia Y, Li L, Huang Y, Won SM, Yu KJ, Sheng X, Fang H, Alam MA, Huang Y, Viventi J, Chang J-K, Rogers JA (2019) Flexible electronic/optoelectronic microsystems with scalable designs for chronic biointegration. Proc Natl Acad Sci 116:15398–15406. https://doi.org/10.1073/pnas.1907697116

Chiang C-H, Won SM, Orsborn AL, Yu KJ, Trumpis M, Bent B, Wang C, Xue Y, Min S, Woods V, Yu C, Kim BH, Kim SB, Huq R, Li J, Seo KJ, Vitale F, Richardson A, Fang H, Huang Y, Shepard K, Pesaran B, Rogers JA, Viventi J (2020) Development of a neural interface for high-definition, long-term recording in rodents and nonhuman primates. Sci Transl Med 12:eaay4682. https://doi.org/10.1126/scitranslmed.aay4682

Yu X, Xie Z, Yu Y, Lee J, Vazquez-Guardado A, Luan H, Ruban J, Ning X, Akhtar A, Li D, Ji B, Liu Y, Sun R, Cao J, Huo Q, Zhong Y, Lee C, Kim S, Gutruf P, Zhang C, Xue Y, Guo Q, Chempakasseril A, Tian P, Lu W, Jeong J, Yu Y, Cornman J, Tan C, Kim B, Lee K, Feng X, Huang Y, Rogers JA (2019) Skin-integrated wireless haptic interfaces for virtual and augmented reality. Nature 575:473–479. https://doi.org/10.1038/s41586-019-1687-0

Shin J, Wang H, Kwon K, Ostojich D, Christiansen Z, Berkovich J, Park Y, Li Z, Lee G, Nasif R, Chung TS, Su C-J, Lim J, Kubota H, Ikoma A, Lu Y-A, Lin DH, Xu S, Banks A, Chang J-K, Rogers JA (2023) Wireless, soft sensors of skin hydration with designs optimized for rapid, accurate diagnostics of dermatological health. Adv Healthc Mater 12:2202021. https://doi.org/10.1002/adhm.202202021

Chung HU, Kim BH, Lee JY, Lee J, Xie Z, Ibler EM, Lee K, Banks A, Jeong JY, Kim J, Ogle C, Grande D, Yu Y, Jang H, Assem P, Ryu D, Kwak JW, Namkoong M, Park JB, Lee Y, Kim DH, Ryu A, Jeong J, You K, Ji B, Liu Z, Huo Q, Feng X, Deng Y, Xu Y, Jang K-I, Kim J, Zhang Y, Ghaffari R, Rand CM, Schau M, Hamvas A, Weese-Mayer DE, Huang Y, Lee SM, Lee CH, Shanbhag NR, Paller AS, Xu S, Rogers JA (2019) Binodal, wireless epidermal electronic systems with in-sensor analytics for neonatal intensive care. Science 363:eaau0780. https://doi.org/10.1126/science.aau0780

Chung HU, Rwei AY, Hourlier-Fargette A, Xu S, Lee K, Dunne EC, Xie Z, Liu C, Carlini A, Kim DH, Ryu D, Kulikova E, Cao J, Odland IC, Fields KB, Hopkins B, Banks A, Ogle C, Grande D, Park JB, Kim J, Irie M, Jang H, Lee J, Park Y, Kim J, Jo HH, Hahm H, Avila R, Xu Y, Namkoong M, Kwak JW, Suen E, Paulus MA, Kim RJ, Parsons BV, Human KA, Kim SS, Patel M, Reuther W, Kim HS, Lee SH, Leedle JD, Yun Y, Rigali S, Son T, Jung I, Arafa H, Soundararajan VR, Ollech A, Shukla A, Bradley A, Schau M, Rand CM, Marsillio LE, Harris ZL, Huang Y, Hamvas A, Paller AS, Weese-Mayer DE, Lee JY, Rogers JA (2020) Skin-interfaced biosensors for advanced wireless physiological monitoring in neonatal and pediatric intensive-care units. Nat Med 26:418–429. https://doi.org/10.1038/s41591-020-0792-9

Boutry CM, Kaizawa Y, Schroeder BC, Chortos A, Legrand A, Wang Z, Chang J, Fox P, Bao Z (2018) A stretchable and biodegradable strain and pressure sensor for orthopaedic application. Nat Electron 1:314–321. https://doi.org/10.1038/s41928-018-0071-7

Abramson A, Chan CT, Khan Y, Mermin-Bunnell A, Matsuhisa N, Fong R, Shad R, Hiesinger W, Mallick P, Gambhir SS, Bao Z (2022) A flexible electronic strain sensor for the real-time monitoring of tumor regression. Sci Adv 8:eabn6550. https://doi.org/10.1126/sciadv.abn6550

Li G, Liu Y, Chen Y, Xia Y, Qi X, Wan X, Jin Y, Liu J, He Q, Li K, Tang J (2024) Robust, self-adhesive, and low-contact impedance polyvinyl alcohol/polyacrylamide dual-network hydrogel semidry electrode for biopotential signal acquisition. SmartMat 5:e1173. https://doi.org/10.1002/smm2.1173

Ham J, Huh TM, Kim J, Kim J-O, Park S, Cutkosky MR, Bao Z (2023) Porous dielectric elastomer based flexible multiaxial tactile sensor for dexterous robotic or prosthetic hands. Adv Mater Technol 8:2200903. https://doi.org/10.1002/admt.202200903

Ham J, Han AK, Cutkosky MR, Bao Z (2022) UV-laser-machined stretchable multi-modal sensor network for soft robot interaction. NPJ Flex Electron 6:94. https://doi.org/10.1038/s41528-022-00225-0

Ruth SRA, Beker L, Tran H, Feig VR, Matsuhisa N, Bao Z (2020) Rational design of capacitive pressure sensors based on pyramidal microstructures for specialized monitoring of biosignals. Adv Funct Mater 30:1903100. https://doi.org/10.1002/adfm.201903100

Jk O’Neill S, Gong H, Matsuhisa N, Chen S, Moon H, Wu H-C, Chen X, Chen X, Bao Z (2020) A carbon flower based flexible pressure sensor made from large-area coating. Adv Mater Interfaces 7:2000875. https://doi.org/10.1002/admi.202000875

Kotlowski C, Aspermair P, Khan HU, Reiner-Rozman C, Breu J, Szunerits S, Kim J-J, Bao Z, Kleber C, Pelosi P, Knoll W (2018) Electronic biosensing with flexible organic transistor devices. Flex Print Electron 3:034003. https://doi.org/10.1088/2058-8585/aad433

Oh JY, Son D, Katsumata T, Lee Y, Kim Y, Lopez J, Wu H-C, Kang J, Park J, Gu X, Mun J, Wang NG-J, Yin Y, Cai W, Yun Y, Tok JBH, Bao Z (2019) Stretchable self-healable semiconducting polymer film for active-matrix strain-sensing array. Sci Adv 5:eaav3097. https://doi.org/10.1126/sciadv.aav3097

Lee G-H, Moon H, Kim H, Lee GH, Kwon W, Yoo S, Myung D, Yun SH, Bao Z, Hahn SK (2020) Multifunctional materials for implantable and wearable photonic healthcare devices. Nat Rev Mater 5:149–165. https://doi.org/10.1038/s41578-019-0167-3

Kim S-K, Lee G-H, Jeon C, Han HH, Kim S-J, Mok JW, Joo C-K, Shin S, Sim J-Y, Myung D, Bao Z, Hahn SK (2022) Bimetallic nanocatalysts immobilized in nanoporous hydrogels for long-term robust continuous glucose monitoring of smart contact lens. Adv Mater 34:2110536. https://doi.org/10.1002/adma.202110536

Boutry CM, Beker L, Kaizawa Y, Vassos C, Tran H, Hinckley AC, Pfattner R, Niu S, Li J, Claverie J, Wang Z, Chang J, Fox PM, Bao Z (2019) Biodegradable and flexible arterial-pulse sensor for the wireless monitoring of blood flow. Nat Biomed Eng 3:47–57. https://doi.org/10.1038/s41551-018-0336-5

Deng M, Du C, Fang J, Xu C, Guo C, Huang J, Li K, Chen L, Zhang Y-T, Chang Y, Pan T (2024) Flexible adaptive sensing tonometry for medical-grade multi-parametric hemodynamic monitoring. npj Flex Electron 8:45. https://doi.org/10.1038/s41528-024-00329-9

Wang X, Deng Y, Chen X, Jiang P, Cheung YK, Yu H (2021) An ultrafast-response and flexible humidity sensor for human respiration monitoring and noncontact safety warning. Microsystems Nanoeng 7:99. https://doi.org/10.1038/s41378-021-00324-4

Zhong M, Zhang L, Liu X, Zhou Y, Zhang M, Wang Y, Yang L, Wei D (2021) Wide linear range and highly sensitive flexible pressure sensor based on multistage sensing process for health monitoring and human-machine interfaces. Chem Eng J 412:128649. https://doi.org/10.1016/j.cej.2021.128649

Li Y, Zhang M, Hu X, Yu L, Fan X, Huang C, Li Y (2021) Graphdiyne-based flexible respiration sensors for monitoring human health. Nano Today 39:101214. https://doi.org/10.1016/j.nantod.2021.101214

Tam H, Webster JG (1977) Minimizing electrode motion artifact by skin abrasion. IEEE Trans Biomed Eng BME 24:134–139. https://doi.org/10.1109/TBME.1977.326117

Pandia K, Ravindran S, Cole R, Kovacs G, Giovangrandi L (2010) Motion artifact cancellation to obtain heart sounds from a single chest-worn accelerometer. 2010 IEEE International Conference on Acoustics, Speech and Signal Processing, Dallas, TX, USA, pp. 590–593. https://doi.org/10.1109/ICASSP.2010.5495553

Lee H, Chung H, Lee J (2019) Motion artifact cancellation in wearable photoplethysmography using gyroscope. IEEE Sens J 19:1166–1175. https://doi.org/10.1109/JSEN.2018.2879970

Ershad F, Thukral A, Yue J, Comeaux P, Lu Y, Shim H, Sim K, Kim N-I, Rao Z, Guevara R, Contreras L, Pan F, Zhang Y, Guan Y-S, Yang P, Wang X, Wang P, Wu X, Yu C (2020) Ultra-conformal drawn-on-skin electronics for multifunctional motion artifact-free sensing and point-of-care treatment. Nat Commun 11:3823. https://doi.org/10.1038/s41467-020-17619-1

Kim J, Ouh H, Johnston ML (2021) Multi-channel biopotential acquisition system using frequency-division multiplexing with cable motion artifact suppression. IEEE T Biomed Circ S 15:1419–1429. https://doi.org/10.1109/TBCAS.2021.3131642

Khateb F, Prommee P, Kulej T (2022) MIOTA-based filters for noise and motion artifact reductions in biosignal acquisition. IEEE Access 10:14325–14338. https://doi.org/10.1109/ACCESS.2022.3147665

Wiese SR, Anheier P, Connemara RD, Mollner AT, Neils TF, Kahn JA, Webster JG (2005) Electrocardiographic motion artifact versus electrode impedance. IEEE Trans Biomed Eng 52:136–139. https://doi.org/10.1109/TBME.2004.836503

Buxi D, Kim S, Helleputte NV, Altini M, Wijsman J, Yazicioglu RF, Penders J, Hoof CV (2012) Correlation between electrode-tissue impedance and motion artifact in biopotential recordings. IEEE Sens J 12:3373–3383. https://doi.org/10.1109/JSEN.2012.2221163

Addison PS (2005) Wavelet transforms and the ECG: a review. Physiol Meas 26:R155. https://doi.org/10.1088/0967-3334/26/5/R01

Zhang Y, Song S, Vullings R, Biswas D, Simões-Capela N, Helleputte N, Hoof C, Groenendaal W (2019) Motion artifact reduction for wrist-worn photoplethysmograph sensors based on different wavelengths. Sensors 19:673. https://doi.org/10.3390/s19030673

Gajbhiye P, Tripathy RK, Bhattacharyya A, Pachori RB (2019) Novel approaches for the removal of motion artifact from EEG recordings. IEEE Sens J 19:10600–10608. https://doi.org/10.1109/JSEN.2019.2931727

Kumar S, Yadav AS, Yadav S, Pani NK, Cotrina-Aliaga JCC, Sernaqué MAC, Carranza CPM, Kapila D (2023) A novel speech steganography mechanism to securing data through shift invariant continuous wavelet transform with speech activity and message detection. Eng Sci 25:950. https://doi.org/10.30919/es950

Uplaonkar DS, Virupakshappa V, Patil N (2021) An efficient discrete wavelet transform based partial hadamard feature extraction and hybrid neural network based monarch butterfly optimization for liver tumor classification. Eng Sci 16:354–365. https://doi.org/10.30919/es8d594

Shibata K, Nagai D, Nakagawa T (2021) Motion artifact removal in R-R interval estimation of ECG signal for wearable biomedical sensors using slope information and correlation matching. IEICE Proceedings Series 68:P3-21. https://doi.org/10.34385/proc.68.P3-21

Ranjan R, Sahana BC, Bhandari AK (2022) Motion artifacts suppression from EEG signals using an adaptive signal denoising method. IEEE T Instrum Meas 71:1–10. https://doi.org/10.1109/TIM.2022.3142037

Hamilton P S, Curley M G, Aimi R M, Sae-Hau C (2000) Comparison of methods for adaptive removal of motion artifact. Computers in Cardiology 2000 , Cambridge, MA, USA, pp. 383–386. https://doi.org/10.1109/CIC.2000.898537

Raya M A D, Sison L G (2022) Adaptive noise cancelling of motion artifact in stress ecg signals using accelerometer. Proceedings of the Second Joint 24th Annual Conference and the Annual Fall Meeting of the Biomedical Engineering Society,Engineering in Medicine and Biology, Houston, TX, USA, pp. 1756–1757. https://doi.org/10.1109/IEMBS.2002.1106637

Kim S H, Ryoo D W, Bae C (2007) Adaptive noise cancellation using accelerometers for the ppg signal from forehead. 2007 29th Annual International Conference of the IEEE Engineering in Medicine and Biology Society, Lyon, France, pp. 2564–2567. https://doi.org/10.1109/IEMBS.2007.4352852

Ram MR, Madhav KV, Krishna EH, Komalla NR, Reddy KA (2012) A novel approach for motion artifact reduction in PPG signals based on AS-LMS adaptive filter. IEEE T Instrum Meas 61:1445–1457. https://doi.org/10.1109/TIM.2011.2175832

Yousefi R, Nourani M, Panahi I (2012) Adaptive cancellation of motion artifact in wearable biosensors. 2012 Annual International Conference of the IEEE Engineering in Medicine and Biology Society, San Diego, CA, USA, pp. 2004–2008. https://doi.org/10.1109/EMBC.2012.6346350

Yousefi R, Nourani M, Ostadabbas S, Panahi I (2014) A motion-tolerant adaptive algorithm for wearable photoplethysmographic biosensors. IEEE J Biomed Health 18:670–681. https://doi.org/10.1109/JBHI.2013.2264358

Kilicarslan A, Contreras Vidal JL (2019) Characterization and real-time removal of motion artifacts from EEG signals. J Neural Eng 16:056027. https://doi.org/10.1088/1741-2552/ab2b61

Reddy K A, Kumar V J (2007) Motion artifact reduction in photoplethysmographic signals using singular value decomposition. 2007 IEEE Instrumentation & Measurement Technology Conference IMTC 2007, Warsaw, Poland, pp. 1–4. https://doi.org/10.1109/IMTC.2007.379467

Zheng Y, Hu X (2019) Interference removal From electromyography based on independent component analysis. IEEE Trans Neural Syst Rehabil Eng 27:887–894. https://doi.org/10.1109/TNSRE.2019.2910387

Kolda TG, Bader BW (2009) Tensor decompositions and applications. SIAM Rev 51:455–500. https://doi.org/10.1137/07070111X

Lilienthal J, Dargie W (2020) Application of tensor decomposition in removing motion artifacts from the measurements of a wireless electrocardiogram. 2020 IEEE 23rd International Conference on Information Fusion (FUSION), Rustenburg, South Africa, pp. 1–8. https://doi.org/10.23919/FUSION45008.2020.9190621

Mishra P, Singla SK (2013) Artifact removal from biosignal using fixed point ICA algorithm for pre-processing in biometric recognition. Meas Sci Rev 13:7–11

Arnold Y, Harinath G, Carolyn V T, Patrick M, Ozgur B, Scott M, Naznin V B (2015) Comparison of foam-based and spring-loaded dry eeg electrodes with wet electrodes in resting and moving conditions. 2015 37th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC), Milan, Italy, pp. 7131–7134. https://doi.org/10.1109/EMBC.2015.7320036

Reddy KA, George B, Kumar VJ (2009) Use of fourier series analysis for motion artifact reduction and data compression of photoplethysmographic signals. IEEE T Instrum Meas 58:1706–1711. https://doi.org/10.1109/TIM.2008.2009136

Pankaj KA, Komaragiri R, Kumar M (2022) Reference signal less Fourier analysis based motion artifact removal algorithm for wearable photoplethysmography devices to estimate heart rate during physical exercises. Comput Biol Med 141:105081. https://doi.org/10.1016/j.compbiomed.2021.105081

Maddirala AK, Shaik RA (2016) Motion artifact removal from single channel electroencephalogram signals using singular spectrum analysis. Biomed Signal Proces 30:79–85. https://doi.org/10.1016/j.bspc.2016.06.017

Mahmud S, Hossain MS, Chowdhury MEH, Reaz MBI (2023) MLMRS-net: electroencephalography (EEG) motion artifacts removal using a multi-layer multi-resolution spatially pooled 1D signal reconstruction network. Neural Comput Appl 35:8371–8388. https://doi.org/10.1007/s00521-022-08111-6

Stankoski S, Kiprijanovska I, Mavridou I, Nduka C, Gjoreski H, Gjoreski M (2022) Breathing rate estimation from head-worn photoplethysmography sensor data using machine learning. Sensors 22:2079. https://doi.org/10.3390/s22062079

Cömert A, Hyttinen J (2014) Impedance spectroscopy of changes in skin-electrode impedance induced by motion. Biomed Eng Online 13:149. https://doi.org/10.1186/1475-925X-13-149

Cömert A, Honkala M, Hyttinen J (2013) Effect of pressure and padding on motion artifact of textile electrodes. Biomed Eng Online 12:26. https://doi.org/10.1186/1475-925X-12-26

Dabbaghian A, Yousefi T, Fatmi SZ, Shafia P, Kassiri H (2019) A 9.2-g fully-flexible wireless ambulatory EEG monitoring and diagnostics headband with analog motion artifact detection and compensation. IEEE T Biomed Circ S 13:1141–1151. https://doi.org/10.1109/TBCAS.2019.2936327

Xu X, Kothapalli SVVN, Liu J, Kahali S, Gan W, Yablonskiy DA, Kamilov US (2022) Learning-based motion artifact removal networks for quantitative mapping. Magn Reson Med 88:106–119. https://doi.org/10.1002/mrm.29188

Hossain M-B, Posada-Quintero HF, Kong Y, McNaboe R, Chon KH (2022) Automatic motion artifact detection in electrodermal activity data using machine learning. Biomed Signal Proces 74:103483. https://doi.org/10.1016/j.bspc.2022.103483

Download references

This study was supported financially by the Fundamental Research Funds for the Central Universities (2021XD-A04-2), the National Natural Science Foundation of China (Nos. 61874014 and 61874013), and the Fund of State Key Laboratory of Information Photonics and Optical Communications (Beijing University of Posts and Telecommunications, People’s Republic of China), and BUPT Excellent Ph.D. Students Foundation (CX2022237).

Author information

Authors and affiliations.

College of Electronic and Optical Engineering College of Flexible Electronics (Future Technology), Nanjing University of Posts and Telecommunications, Nanjing, 210023, China

State Key Laboratory of Information Photonics and Optical Communications, School of Science, Beijing University of Posts and Telecommunications, Beijing, 100876, China

State Key Laboratory of Information Photonics and Optical Communications, School of Integrated Circuits, Beijing University of Posts and Telecommunications, Beijing, 100876, People’s Republic of China

You can also search for this author in PubMed   Google Scholar

Contributions

Tao Zeng, Yufeng Wu, and Ming Lei proposed the idea of the manuscript and discussed the framework of the paper fully. Tao Zeng and Yufeng Wu co-wrote the manuscript. Ming Lei added and corrected the contents of the manuscript and provided various research supports.

Corresponding author

Correspondence to Yufeng Wu .

Ethics declarations

Conflict of interest.

The authors declare no competing interests.

Additional information

Publisher's note.

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Rights and permissions

Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.

Reprints and permissions

About this article

Zeng, T., Wu, Y. & Lei, M. Review: Developments and challenges of advanced flexible electronic materials for medical monitoring applications. Adv Compos Hybrid Mater 7 , 141 (2024). https://doi.org/10.1007/s42114-024-00949-9

Download citation

Received : 29 May 2024

Revised : 06 August 2024

Accepted : 22 August 2024

Published : 10 September 2024

DOI : https://doi.org/10.1007/s42114-024-00949-9

Share this article

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

• Motion artifact
• Smart biosensors
• Monitoring devices
• Smart wearable flexible electronics
• Find a journal
• Publish with us
• Track your research