Combined deep CNN-LSTM network-based multitasking learning architecture for noninvasive continuous blood pressure estimation using difference in ECG-PPG features

Prototype Learning for Medical Time Series Classification via Human-Machine Collaboration

Deep neural networks must address the dual challenge of delivering high-accuracy predictions and providing user-friendly explanations. While deep models are widely used in the field of time series modeling, deciphering the core principles that govern the models' outputs remains a significant challenge. This is crucial for fostering the development of trusted models and facilitating domain expert validation, thereby empowering users and domain experts to utilize them confidently in high-risk decision-making contexts (e.g., decision-support systems in healthcare). In this work, we put forward a deep prototype learning model that supports interpretable and manipulable modeling and classification of medical time series (i.e., ECG signal). Specifically, we first optimize the representation of single heartbeat data by employing a bidirectional long short-term memory and attention mechanism, and then construct prototypes during the training phase. The final classification outcomes (i.e., normal sinus rhythm, atrial fibrillation, and other rhythm) are determined by comparing the input with the obtained prototypes. Moreover, the proposed model presents a human-machine collaboration mechanism, allowing domain experts to refine the prototypes by integrating their expertise to further enhance the model's performance (contrary to the human-in-the-loop paradigm, where humans primarily act as supervisors or correctors, intervening when required, our approach focuses on a human-machine collaboration, wherein both parties engage as partners, enabling more fluid and integrated interactions). The experimental outcomes presented herein delineate that, within the realm of binary classification tasks-specifically distinguishing between normal sinus rhythm and atrial fibrillation-our proposed model, albeit registering marginally lower performance in comparison to certain established baseline models such as Convolutional Neural Networks (CNNs) and bidirectional long short-term memory with attention mechanisms (Bi-LSTMAttns), evidently surpasses other contemporary state-of-the-art prototype baseline models. Moreover, it demonstrates significantly enhanced performance relative to these prototype baseline models in the context of triple classification tasks, which encompass normal sinus rhythm, atrial fibrillation, and other rhythm classifications. The proposed model manifests a commendable prediction accuracy of 0.8414, coupled with macro precision, recall, and F1-score metrics of 0.8449, 0.8224, and 0.8235, respectively, achieving both high classification accuracy as well as good interpretability.

Bayesian and deep-learning models applied to the early detection of ovarian cancer using multiple longitudinal biomarkers

BACKGROUND

Ovarian cancer is the most lethal of all gynecological cancers. Cancer Antigen 125 (CA125) is the best-performing ovarian cancer biomarker which however is still not effective as a screening test in the general population. Recent literature reports additional biomarkers with the potential to improve on CA125 for early detection when using longitudinal multimarker models.

METHODS

Our data comprised 180 controls and 44 cases with serum samples sourced from the multimodal arm of UK Collaborative Trial of Ovarian Cancer Screening (UKCTOCS). Our models were based on Bayesian change-point detection and recurrent neural networks.

RESULTS

We obtained a significantly higher performance for CA125-HE4 model using both methodologies (AUC 0.971, sensitivity 96.7% and AUC 0.987, sensitivity 96.7%) with respect to CA125 (AUC 0.949, sensitivity 90.8% and AUC 0.953, sensitivity 92.1%) for Bayesian change-point model (BCP) and recurrent neural networks (RNN) approaches, respectively. One year before diagnosis, the CA125-HE4 model also ranked as the best, whereas at 2 years before diagnosis no multimarker model outperformed CA125.

CONCLUSIONS

Our study identified and tested different combination of biomarkers using longitudinal multivariable models that outperformed CA125 alone. We showed the potential of multivariable models and candidate biomarkers to increase the detection rate of ovarian cancer.

Quantifying the Suitability of Biosignals Acquired During Surgery for Multimodal Analysis

Goal: Recently, large datasets of biosignals acquired during surgery became available. As they offer multiple physiological signals measured in parallel, multimodal analysis - which involves their joint analysis - can be conducted and could provide deeper insights than unimodal analysis based on a single signal. However, it is unclear what percentage of intraoperatively acquired data is suitable for multimodal analysis. Due to the large amount of data, manual inspection and labelling into suitable and unsuitable segments are not feasible. Nevertheless, multimodal analysis is performed successfully in sleep studies since many years as their signals have proven suitable. Hence, this study evaluates the suitability to perform multimodal analysis on a surgery dataset (VitalDB) using a multi-center sleep dataset (SIESTA) as reference. Methods: We applied widely known algorithms entitled "signal quality indicators" to the common biosignals in both datasets, namely electrocardiography, electroencephalography, and respiratory signals split in segments of 10 s duration. As there are no multimodal methods available, we used only unimodal signal quality indicators. In case, all three signals were determined as being adequate by the indicators, we assumed that the whole signal segment was suitable for multimodal analysis. Results: 82% of SIESTA and 72% of VitalDB are suitable for multimodal analysis. Unsuitable signal segments exhibit constant or physiologically unreasonable values. Histogram examination indicated similar signal quality distributions between the datasets, albeit with potential statistical biases due to different measurement setups. Conclusions: The majority of data within VitalDB is suitable for multimodal analysis.

Contactless Blood Pressure Measurement Via Remote Photoplethysmography With Synthetic Data Generation Using Generative Adversarial Networks

Remote photoplethysmography (rPPG) has been used to measure vital signs such as heart rate, heart rate variability, blood pressure (BP), and blood oxygen. Recent studies adopt features developed with photoplethysmography (PPG) to achieve contactless BP measurement via rPPG. These features can be classified into two groups: time or phase differences from multiple signals, or waveform feature analysis from a single signal. Here we devise a solution to extract the time difference information from the rPPG signal captured at 30 FPS. We also propose a deep learning model architecture to estimate BP from the extracted features. To prevent overfitting and compensate for the lack of data, we leverage a multi-model design and generate synthetic data. We also use subject information related to BP to assist in model learning. For real-world usage, the subject information is replaced with values estimated from face images, with performance that is still better than the state-of-the-art. To our best knowledge, the improvements can be achieved because of: 1) the model selection with estimated subject information, 2) replacing the estimated subject information with the real one, 3) the InfoGAN assistance training (synthetic data generation), and 4) the time difference features as model input. To evaluate the performance of the proposed method, we conduct a series of experiments, including dynamic BP measurement for many single subjects and nighttime BP measurement with infrared lighting. Our approach reduces the MAE from 15.49 to 8.78 mmHg for systolic blood pressure (SBP) and 10.56 to 6.16 mmHg for diastolic blood pressure(DBP) on a self-constructed rPPG dataset. On the Taipei Veterans General Hospital(TVGH) dataset for nighttime applications, the MAE is reduced from 21.58 to 11.12 mmHg for SBP and 9.74 to 7.59 mmHg for DBP, with improvement ratios of 48.47% and 22.07% respectively.

TRCCBP: Transformer Network for Radar-Based Contactless Continuous Blood Pressure Monitoring

Contactless continuous blood pressure (BP) monitoring is of great significance for daily healthcare. Radar-based continuous monitoring methods typically extract time-domain features manually such as pulse transit time (PTT) to calculate the BP. However, breathing and slight body movements usually distort the features extracted from pulse-wave signals, especially in long-term continuous monitoring, and manually extracted features may have limited performance for BP estimation. This article proposes a Transformer network for Radar-based Contactless Continuous Blood Pressure monitoring (TRCCBP). A heartbeat signal-guided single-beat pulse wave extraction method is designed to obtain pure pulse-wave signals. A transformer network-based blood pressure estimation network is proposed to estimate BP, which utilizes convolutional layers with different scales, a gated recurrent unit (GRU) to capture time-dependence in continuous radar signal and multi-head attention modules to capture deep temporal domain characteristics. A radar signal dataset captured in an indoor environment containing 31 persons and a real medical situation containing five persons is set up to evaluate the performance of TRCCBP. Compared with the state-of-the-art method, the average accuracy of diastolic blood pressure (DBP) and systolic blood pressure (SBP) is 4.49 mmHg and 4.73 mmHg, improved by 12.36 mmHg and 8.80 mmHg, respectively. The proposed TRCCBP source codes and radar signal dataset have been made open-source online for further research.

A Multi-modal Teacher-student Framework for Improved Blood Pressure Estimation

Blood pressure (BP) is a critical vital sign that hypertensive patients regularly measure. In this study, we propose a novel BP estimation framework to distill the knowledge from a multi-modal model to a uni-modal BP estimation model through teacher-student training. The multi-modal BP estimation model consists of three components: first, a gated recurrent unit network that generates features from photoplethysmogram, electrocardiogram, age, height, and weight; second, an attention mechanism that integrates each feature into joint embeddings; and third, a regression layer that estimates BP from the joint embeddings. The uni-modal BP estimation model has similar components to the multi-modal model but uses only PPG signal. BP is predicted by the embeddings extracted from the uni-modal model, and these embeddings are trained to be as similar as possible to the joint embeddings extracted from the multi-modal model. The proposed method demonstrates absolute prediction errors of 5.24±6.41 and 3.49±3.85 mmHg for systolic blood pressure and diastolic blood pressure in the MIMIC-III dataset, satisfying the AAMI standard.

Boosting Algorithms based Cuff-less Blood Pressure Estimation from Clinically Relevant ECG and PPG Morphological Features

Blood Pressure (BP) is often coined as a critical physiological marker for cardiovascular health. Multiple studies have explored either Photoplethysmogram (PPG) or ECG-PPG derived features for continuous BP estimation using machine learning (ML); deep learning (DL) techniques. Majority of those derived features often lack a stringent biological explanation and are not significantly correlated with BP. In this paper, we identified several clinically relevant (bio-inspired) ECG and PPG features; and exploited them to estimate Systolic (SBP), and Diastolic Blood Pressure (DBP) values using CatBoost, and AdaBoost algorithms. The estimation performance was then compared against popular ML algorithms. SBP and DBP achieved a Pearson's correlation coefficient of 0.90 and 0.83 between estimated and target BP values. The estimated mean absolute error (MAE) values are 3.81 and 2.22 mmHg with a Standard Deviation of 6.24 and 3.51 mmHg, respectively, for SBP and DBP using CatBoost. The results surpassed the Advancement of Medical Instrumentation (AAMI) standards. For the British Hypertension Society (BHS) protocol, the results achieved for all the BP categories resided in Grade A. Further investigation reveals that bio-inspired features along with tuned ML models can produce comparable results w.r.t parameter-intensive DL networks. ln(HR × mNPV), HR, BMI index, ageing index, and PPG-K point were identified as the top five key features for estimating BP. The group-based analysis further concludes that a trade-off lies between the number of features and MAE. Increasing the no. of features beyond a certain threshold saturates the reduction in MAE.

Optimizing the input feature sets and machine learning algorithms for reliable and accurate estimation of continuous, cuffless blood pressure

The advent of mobile devices, wearables and digital healthcare has unleashed a demand for accurate, reliable, and non-interventional ways to measure continuous blood pressure (BP). Many consumer products claim to measure BP with a cuffless device, but their lack of accuracy and reliability limit clinical adoption. Here, we demonstrate how multimodal feature datasets, comprising: (i) pulse arrival time (PAT); (ii) pulse wave morphology (PWM), and (iii) demographic data, can be combined with optimized Machine Learning (ML) algorithms to estimate Systolic BP (SBP), Diastolic BP (DBP) and Mean Arterial Pressure (MAP) within a 5 mmHg bias of the gold standard Intra-Arterial BP, well within the acceptable limits of the IEC/ANSI 80601-2-30 (2018) standard. Furthermore, DBP's calculated using 126 datasets collected from 31 hemodynamically compromised patients had a standard deviation within 8 mmHg, while SBP's and MAP's exceeded these limits. Using ANOVA and Levene's test for error means and standard deviations, we found significant differences in the various ML algorithms but found no significant differences amongst the multimodal feature datasets. Optimized ML algorithms and key multimodal features obtained from larger real-world data (RWD) sets could enable more reliable and accurate estimation of continuous BP in cuffless devices, accelerating wider clinical adoption.

[A non-contact continuous blood pressure measurement method based on video stream]

Hypertension is the primary disease that endangers human health. A convenient and accurate blood pressure measurement method can help to prevent the hypertension. This paper proposed a continuous blood pressure measurement method based on facial video signal. Firstly, color distortion filtering and independent component analysis were used to extract the video pulse wave of the region of interest in the facial video signal, and the multi-dimensional feature extraction of the pulse wave was preformed based on the time-frequency domain and physiological principles; Secondly, an integrated feature selection method was designed to extract the universal optimal feature subset; After that, we compared the single person blood pressure measurement models established by Elman neural network based on particle swarm optimization, support vector machine (SVM) and deep belief network; Finally, we used SVM algorithm to build a general blood pressure prediction model, which was compared and evaluated with the real blood pressure value. The experimental results showed that the blood pressure measurement results based on facial video were in good agreement with the standard blood pressure values. Comparing the estimated blood pressure from the video with standard blood pressure value, the mean absolute error (MAE) of systolic blood pressure was 4.9 mm Hg with a standard deviation (STD) of 5.9 mm Hg, and the MAE of diastolic blood pressure was 4.6 mm Hg with a STD of 5.0 mm Hg, which met the AAMI standards. The non-contact blood pressure measurement method based on video stream proposed in this paper can be used for blood pressure measurement.

Multi-Task Classification of Physical Activity and Acute Psychological Stress for Advanced Diabetes Treatment

Wearable sensor data can be integrated and interpreted to improve the treatment of chronic conditions, such as diabetes, by enabling adjustments in treatment decisions based on physical activity and psychological stress assessments. The challenges in using biological analytes to frequently detect physical activity (PA) and acute psychological stress (APS) in daily life necessitate the use of data from noninvasive sensors in wearable devices, such as wristbands. We developed a recurrent multi-task deep neural network (NN) with long-short-term-memory architecture to integrate data from multiple sensors (blood volume pulse, skin temperature, galvanic skin response, three-axis accelerometers) and simultaneously detect and classify the type of PA, namely, sedentary state, treadmill run, stationary bike, and APS, such as non-stress, emotional anxiety stress, mental stress, and estimate the energy expenditure (EE). The objective was to assess the feasibility of using the multi-task recurrent NN (RNN) rather than independent RNNs for detection and classification of AP and APS. The multi-task RNN achieves comparable performance to independent RNNs, with the multi-task RNN having F1 scores of 98.00% for PA and 98.97% for APS, and a root mean square error (RMSE) of 0.728 calhr.kg for EE estimation for testing data. The independent RNNs have F1 scores of 99.64% for PA and 98.83% for APS, and an RMSE of 0.666 calhr.kg for EE estimation. The results indicate that a multi-task RNN can effectively interpret the signals from wearable sensors. Additionally, we developed individual and multi-task extreme gradient boosting (XGBoost) for separate and simultaneous classification of PA types and APS types. Multi-task XGBoost achieved F1 scores of 99.89% and 98.31% for the classification of PA types and APS types, respectively, while the independent XGBoost achieved F1 scores of 99.68% and 96.77%, respectively. The results indicate that both multi-task RNN and XGBoost can be used for the detection and classification of PA and APS without loss of performance with respect to individual separate classification systems. People with diabetes can achieve better outcomes and quality of life by including physical activity and psychological stress assessments in treatment decision-making.

PulseDB: A large, cleaned dataset based on MIMIC-III and VitalDB for benchmarking cuff-less blood pressure estimation methods

There has been a growing interest in developing cuff-less blood pressure (BP) estimation methods to enable continuous BP monitoring from electrocardiogram (ECG) and/or photoplethysmogram (PPG) signals. The majority of these methods have been evaluated using publicly-available datasets, however, there exist significant discrepancies across studies with respect to the size, the number of subjects, and the applied pre-processing steps for the data that is eventually used for training and testing the models. Such differences make conducting performance comparison across models largely unfair, and mask the generalization capability of various BP estimation methods. To fill this important gap, this paper presents "PulseDB," the largest cleaned dataset to date, for benchmarking BP estimation models that also fulfills the requirements of standardized testing protocols. PulseDB contains 1) 5,245,454 high-quality 10 -s segments of ECG, PPG, and arterial BP (ABP) waveforms from 5,361 subjects retrieved from the MIMIC-III waveform database matched subset and the VitalDB database; 2) subjects' identification and demographic information, that can be utilized as additional input features to improve the performance of BP estimation models, or to evaluate the generalizability of the models to data from unseen subjects; and 3) positions of the characteristic points of the ECG/PPG signals, making PulseDB directly usable for training deep learning models with minimal data pre-processing. Additionally, using this dataset, we conduct the first study to provide insights about the performance gap between calibration-based and calibration-free testing approaches for evaluating generalizability of the BP estimation models. We expect PulseDB, as a user-friendly, large, comprehensive and multi-functional dataset, to be used as a reliable source for the evaluation of cuff-less BP estimation methods.

Applied machine learning for blood pressure estimation using a small, real-world electrocardiogram and photoplethysmogram dataset

Applying machine learning techniques to electrocardiography and photoplethysmography signals and their multivariate-derived waveforms is an ongoing effort to estimate non-occlusive blood pressure. Unfortunately, real ambulatory electrocardiography and photoplethysmography waveforms are inevitably affected by motion and noise artifacts, so established machine learning architectures perform poorly when trained on data of the Multiparameter Intelligent Monitoring in Intensive Care II type, a publicly available ICU database. Our study addresses this problem by applying four well-established machine learning methods, i.e., random forest regression, support vector regression, Adaboost regression and artificial neural networks, to a small, self-sampled electrocardiography-photoplethysmography dataset (n = 54) to improve the robustness of machine learning to real-world BP estimates. We evaluated the performance using a selection of optimal feature morphologies of waveforms by using pulse arrival time, morphological and frequency photoplethysmography parameters and heart rate variability as characterization data. On the basis of the root mean square error and mean absolute error, our study showed that support vector regression gave the best performance for blood pressure estimation from noisy data, achieving an mean absolute error of 6.97 mmHg, which meets the level C criteria set by the British Hypertension Society. We demonstrate that ambulatory electrocardiography- photoplethysmography signals acquired by mobile discrete devices can be used to estimate blood pressure.

Advancement in the Cuffless and Noninvasive Measurement of Blood Pressure: A Review of the Literature and Open Challenges

Hypertension is a chronic condition that is one of the prominent reasons behind cardiovascular disease, brain stroke, and organ failure. Left unnoticed and untreated, the deterioration in a health condition could even result in mortality. If it can be detected early, with proper treatment, undesirable outcomes can be avoided. Until now, the gold standard is the invasive way of measuring blood pressure (BP) using a catheter. Additionally, the cuff-based and noninvasive methods are too cumbersome or inconvenient for frequent measurement of BP. With the advancement of sensor technology, signal processing techniques, and machine learning algorithms, researchers are trying to find the perfect relationships between biomedical signals and changes in BP. This paper is a literature review of the studies conducted on the cuffless noninvasive measurement of BP using biomedical signals. Relevant articles were selected using specific criteria, then traditional techniques for BP measurement were discussed along with a motivation for cuffless measurement use of biomedical signals and machine learning algorithms. The review focused on the progression of different noninvasive cuffless techniques rather than comparing performance among different studies. The literature survey concluded that the use of deep learning proved to be the most accurate among all the cuffless measurement techniques. On the other side, this accuracy has several disadvantages, such as lack of interpretability, computationally extensive, standard validation protocol, and lack of collaboration with health professionals. Additionally, the continuing work by researchers is progressing with a potential solution for these challenges. Finally, future research directions have been provided to encounter the challenges.

Classification of Blood Pressure Levels Based on Photoplethysmogram and Electrocardiogram Signals with a Concatenated Convolutional Neural Network

Hypertension is a severe public health issue worldwide that significantly increases the risk of cardiac vascular disease, stroke, brain hemorrhage, and renal dysfunction. Early screening of blood pressure (BP) levels is essential to prevent the dangerous complication associated with hypertension as the leading cause of death. Recent studies have focused on employing photoplethysmograms (PPG) with machine learning to classify BP levels. However, several studies claimed that electrocardiograms (ECG) also strongly correlate with blood pressure. Therefore, we proposed a concatenated convolutional neural network which integrated the features extracted from PPG and ECG signals. This study used the MIMIC III dataset, which provided PPG, ECG, and arterial blood pressure (ABP) signals. A total of 14,298 signal segments were obtained from 221 patients, which were divided into 9150 signals of train data, 2288 signals of validation data, and 2860 signals of test data. In the training process, five-fold cross-validation was applied to select the best model with the highest classification performance. The proposed concatenated CNN architecture using PPG and ECG obtained the highest test accuracy of 94.56-95.15% with a 95% confidence interval in classifying BP levels into hypotension, normotension, prehypertension, hypertension stage 1, and hypertension stage 2. The result shows that the proposed method is a promising solution to categorize BP levels effectively, assisting medical personnel in making a clinical diagnosis.

Blood Pressure Measurement: From Cuff-Based to Contactless Monitoring

Blood pressure (BP) determines whether a person has hypertension and offers implications as to whether he or she could be affected by cardiovascular disease. Cuff-based sphygmomanometers have traditionally provided both accuracy and reliability, but they require bulky equipment and relevant skills to obtain precise measurements. BP measurement from photoplethysmography (PPG) signals has become a promising alternative for convenient and unobtrusive BP monitoring. Moreover, the recent developments in remote photoplethysmography (rPPG) algorithms have enabled new innovations for contactless BP measurement. This paper illustrates the evolution of BP measurement techniques from the biophysical theory, through the development of contact-based BP measurement from PPG signals, and to the modern innovations of contactless BP measurement from rPPG signals. We consolidate knowledge from a diverse background of academic research to highlight the importance of multi-feature analysis for improving measurement accuracy. We conclude with the ongoing challenges, opportunities, and possible future directions in this emerging field of research.

Beat-to-Beat Blood Pressure Estimation by Photoplethysmography and Its Interpretation

Blood pressure (BP) is among the most important vital signals. Estimation of absolute BP solely using photoplethysmography (PPG) has gained immense attention over the last years. Available works differ in terms of used features as well as classifiers and bear large differences in their results. This work aims to provide a machine learning method for absolute BP estimation, its interpretation using computational methods and its critical appraisal in face of the current literature. We used data from three different sources including 273 subjects and 259,986 single beats. We extracted multiple features from PPG signals and its derivatives. BP was estimated by xgboost regression. For interpretation we used Shapley additive values (SHAP). Absolute systolic BP estimation using a strict separation of subjects yielded a mean absolute error of 9.456mmHg and correlation of 0.730. The results markedly improve if data separation is changed (MAE: 6.366mmHg, r: 0.874). Interpretation by means of SHAP revealed four features from PPG, its derivation and its decomposition to be most relevant. The presented approach depicts a general way to interpret multivariate prediction algorithms and reveals certain features to be valuable for absolute BP estimation. Our work underlines the considerable impact of data selection and of training/testing separation, which must be considered in detail when algorithms are to be compared. In order to make our work traceable, we have made all methods available to the public.

Recent Advances in Non-Invasive Blood Pressure Monitoring and Prediction Using a Machine Learning Approach

Blood pressure (BP) monitoring can be performed either invasively via arterial catheterization or non-invasively through a cuff sphygmomanometer. However, for conscious individuals, traditional cuff-based BP monitoring devices are often uncomfortable, intermittent, and impractical for frequent measurements. Continuous and non-invasive BP (NIBP) monitoring is currently gaining attention in the human health monitoring area due to its promising potentials in assessing the health status of an individual, enabled by machine learning (ML), for various purposes such as early prediction of disease and intervention treatment. This review presents the development of a non-invasive BP measuring tool called sphygmomanometer in brief, summarizes state-of-the-art NIBP sensors, and identifies extended works on continuous NIBP monitoring using commercial devices. Moreover, the NIBP predictive techniques including pulse arrival time, pulse transit time, pulse wave velocity, and ML are elaborated on the basis of bio-signals acquisition from these sensors. Additionally, the different BP values (systolic BP, diastolic BP, mean arterial pressure) of the various ML models adopted in several reported studies are compared in terms of the international validation standards developed by the Advancement of Medical Instrumentation (AAMI) and the British Hypertension Society (BHS) for clinically-approved BP monitors. Finally, several challenges and possible solutions for the implementation and realization of continuous NIBP technology are addressed.

State-of-the-art Deep Learning Methods on Electrocardiogram Data: A Systematic Review (Preprint)

Background

Electrocardiogram (ECG) is one of the most common noninvasive diagnostic tools that can provide useful information regarding a patient’s health status. Deep learning (DL) is an area of intense exploration that leads the way in most attempts to create powerful diagnostic models based on physiological signals.

Objective

This study aimed to provide a systematic review of DL methods applied to ECG data for various clinical applications.

Methods

The PubMed search engine was systematically searched by combining “deep learning” and keywords such as “ecg,” “ekg,” “electrocardiogram,” “electrocardiography,” and “electrocardiology.” Irrelevant articles were excluded from the study after screening titles and abstracts, and the remaining articles were further reviewed. The reasons for article exclusion were manuscripts written in any language other than English, absence of ECG data or DL methods involved in the study, and absence of a quantitative evaluation of the proposed approaches.

Results

We identified 230 relevant articles published between January 2020 and December 2021 and grouped them into 6 distinct medical applications, namely, blood pressure estimation, cardiovascular disease diagnosis, ECG analysis, biometric recognition, sleep analysis, and other clinical analyses. We provide a complete account of the state-of-the-art DL strategies per the field of application, as well as major ECG data sources. We also present open research problems, such as the lack of attempts to address the issue of blood pressure variability in training data sets, and point out potential gaps in the design and implementation of DL models.

Conclusions

We expect that this review will provide insights into state-of-the-art DL methods applied to ECG data and point to future directions for research on DL to create robust models that can assist medical experts in clinical decision-making.

A Flexible, Wearable, and Wireless Biosensor Patch with Internet of Medical Things Applications

Monitoring the vital signs and physiological responses of the human body in daily activities is particularly useful for the early diagnosis and prevention of cardiovascular diseases. Here, we proposed a wireless and flexible biosensor patch for continuous and longitudinal monitoring of different physiological signals, including body temperature, blood pressure (BP), and electrocardiography. Moreover, these modalities for tracking body movement and GPS locations for emergency rescue have been included in biosensor devices. We optimized the flexible patch design with high mechanical stretchability and compatibility that can provide reliable and long-term attachment to the curved skin surface. Regarding smart healthcare applications, this research presents an Internet of Things-connected healthcare platform consisting of a smartphone application, website service, database server, and mobile gateway. The IoT platform has the potential to reduce the demand for medical resources and enhance the quality of healthcare services. To further address the advances in non-invasive continuous BP monitoring, an optimized deep learning architecture with one-channel electrocardiogram signals is introduced. The performance of the BP estimation model was verified using an independent dataset; this experimental result satisfied the Association for the Advancement of Medical Instrumentation, and the British Hypertension Society standards for BP monitoring devices. The experimental results demonstrated the practical application of the wireless and flexible biosensor patch for continuous physiological signal monitoring with Internet of Medical Things-connected healthcare applications.

Assessment of Non-Invasive Blood Pressure Prediction from PPG and rPPG Signals Using Deep Learning

Exploiting photoplethysmography signals (PPG) for non-invasive blood pressure (BP) measurement is interesting for various reasons. First, PPG can easily be measured using fingerclip sensors. Second, camera based approaches allow to derive remote PPG (rPPG) signals similar to PPG and therefore provide the opportunity for non-invasive measurements of BP. Various methods relying on machine learning techniques have recently been published. Performances are often reported as the mean average error (MAE) on the data which is problematic. This work aims to analyze the PPG- and rPPG based BP prediction error with respect to the underlying data distribution. First, we train established neural network (NN) architectures and derive an appropriate parameterization of input segments drawn from continuous PPG signals. Second, we use this parameterization to train NNs with a larger PPG dataset and carry out a systematic evaluation of the predicted blood pressure. The analysis revealed a strong systematic increase of the prediction error towards less frequent BP values across NN architectures. Moreover, we tested different train/test set split configurations which underpin the importance of a careful subject-aware dataset assignment to prevent overly optimistic results. Third, we use transfer learning to train the NNs for rPPG based BP prediction. The resulting performances are similar to the PPG-only case. Finally, we apply different personalization techniques and retrain our NNs with subject-specific data for both the PPG-only and rPPG case. Whilst the particular technique is less important, personalization reduces the prediction errors significantly.