Analyzing single-lead short ECG recordings using dense convolutional neural networks and feature-based post-processing to detect atrial fibrillation

A lightweight 1D convolutional neural network model for arrhythmia diagnosis from electrocardiogram signal

Electrocardiogram (ECG) is used by cardiologist to diagnose heart diseases. The use of ECG signal in an artificial intelligence system can permit to automatically analyze these signals and thereby improve diagnosis quality. For this purpose, many models have been proposed in the literature. But many of these models are complex enough for implementation in an embedded system dedicated to medical diagnosis. Still others have performances that remain to be improved. To solve this problem of complexity, while improving performance, we propose a simple 1D convolutional neural network model for cardiac arrhythmia diagnosis. The proposed model combines two convolution layers, two max pooling layers, three dense layers, two dropout layers and a flatten layer. We apply the proposed model on the public MIT-BIH database for inter-patient classification of five distinct types of heartbeat rhythms which are consistent with the association for advancement of medical instrumentation (AAMI) standard. We also apply our model on the PTB database in order to evaluate its generalization capability. On the MIT-BIH database, the results provide an accuracy of 0.9842, a precision of 0.9523, a sensitivity of 0.8760, a specificity of 0.9869, a negative predictive value (NPV) of 0.9936, an average area under the ROC curve (AUC) of 0.99 and a F1-measure of 0.9095. The accuracy, precision, sensitivity, specificity, NPV, and AUC on the PTB dataset are 0.9924, 0.9938, 0.9957, 0.9844, 0.9892, and 1, respectively. Compared to other existing models, for unbalanced data, the performances obtained by our model are quite interesting for an inter-patient classification.

Research on atrial fibrillation diagnosis in electrocardiograms based on CLA-AF model

Aims

The electrocardiogram (ECG) is the primary method for diagnosing atrial fibrillation (AF), but interpreting ECGs can be time-consuming and labour-intensive, which deserves more exploration.

Methods and results

We collected ECG data from 6590 patients as YY2023, classified as Normal, AF, and Other. Convolutional Neural Network (CNN), bidirectional Long Short-Term Memory (BiLSTM), and Attention construct the AF recognition model CNN BiLSTM Attention-Atrial Fibrillation (CLA-AF). The generalization ability of the model is validated on public datasets CPSC2018, PhysioNet2017, and PTB-XL, and we explored the performance of oversampling, resampling, and hybrid datasets. Finally, additional PhysioNet2021 was added to validate the robustness and applicability in different clinical settings. We employed the SHapley Additive exPlanations (SHAP) method to interpret the model's predictions. The F1-score, Precision, and area under the ROC curve (AUC) of the CLA-AF model on YY2023 are 0.956, 0.970, and 1.00, respectively. Similarly, the AUC on CPSC2018, PhysioNet2017, and PTB-XL reached above 0.95, demonstrating its strong generalization ability. After oversampling PhysioNet2017, F1-score and Recall improved by 0.156 and 0.260. Generalization ability varied with sampling frequency. The model trained from the hybrid dataset has the most robust generalization ability, achieving an AUC of 0.96 or more. The AUC of PhysioNet2021 is 1.00, which proves the applicability of CLA-AF. The SHAP values visualization results demonstrate that the model's interpretation of AF aligns with the diagnostic criteria of AF.

Conclusion

The CLA-AF model demonstrates a high accuracy in recognizing AF from ECG, exhibiting remarkable applicability and robustness in diverse clinical settings.

Pre-Processing techniques and artificial intelligence algorithms for electrocardiogram (ECG) signals analysis: A comprehensive review

Electrocardiogram (ECG) are the physiological signals and a standard test to measure the heart's electrical activity that depicts the movement of cardiac muscles. A review study has been conducted on ECG signals analysis with the help of artificial intelligence (AI) methods over the last ten years i.e., 2012-22. Primarily, the method of ECG analysis by software systems was divided into classical signal processing (e.g. spectrograms or filters), machine learning (ML) and deep learning (DL), including recursive models, transformers and hybrid. Secondly, the data sources and benchmark datasets were depicted. Authors grouped resources by ECG acquisition methods into hospital-based portable machines and wearable devices. Authors also included new trends like advanced pre-processing, data augmentation, simulations and agent-based modeling. The study found improvement in ECG examination perfection made each year through ML, DL, hybrid models, and transformers. Convolutional neural networks and hybrid models were more targeted and proved efficient. The transformer model extended the accuracy from 90% to 98%. The Physio-Net library helps acquire ECG signals, including the popular benchmark databases such as MIT-BIH, PTB, and challenging datasets. Similarly, wearable devices have been established as a appropriate option for monitoring patient health without the time and place limitations and are also helpful for AI model calibration with so far accuracy of 82%-83% on Samsung smartwatch. In the pre-processing signals, spectrogram generation through Fourier and wavelet transformations are erected leading approaches promoting on average accuracy of 90%-95%. Likewise, data enhancement using geometrical techniques is well-considered; however, extraction and concatenation-based methods need attention. As the what-if analysis in healthcare or cardiac issues can be performed using a complex simulation, the study reviews agent-based modeling and simulation approaches for cardiovascular risk event assessment.

Artificial intelligence in cardiovascular prevention: new ways will open new doors

Prevention and effective treatment of cardiovascular disease are progressive issues that grow in tandem with the average age of the world population. Over recent decades, the potential role of artificial intelligence in cardiovascular medicine has been increasingly recognized because of the incredible amount of real-world data (RWD) regarding patient health status and healthcare delivery that can be collated from a variety of sources wherein patient information is routinely collected, including patient registries, clinical case reports, reimbursement claims and billing reports, medical devices, and electronic health records. Like any other (health) data, RWD can be analysed in accordance with high-quality research methods, and its analysis can deliver valuable patient-centric insights complementing the information obtained from conventional clinical trials. Artificial intelligence application on RWD has the potential to detect a patient's health trajectory leading to personalized medicine and tailored treatment. This article reviews the benefits of artificial intelligence in cardiovascular prevention and management, focusing on diagnostic and therapeutic improvements without neglecting the limitations of this new scientific approach.

Atrial fibrillation classification based on the 2D representation of minimal subset ECG and a non-deep neural network

Atrial fibrillation (AF) is the most common cardiac arrhythmia, and its early detection is critical for preventing complications and optimizing treatment. In this study, a novel AF prediction method is proposed, which is based on investigating a subset of the 12-lead ECG data using a recurrent plot and ParNet-adv model. The minimal subset of ECG leads (II &V1) is determined via a forward stepwise selection procedure, and the selected 1D ECG data is transformed into 2D recurrence plot (RP) images as an input to train a shallow ParNet-adv Network for AF prediction. In this study, the proposed method achieved F1 score of 0.9763, Precision of 0.9654, Recall of 0.9875, Specificity of 0.9646, and Accuracy of 0.9760, which significantly outperformed solutions based on single leads and complete 12 leads. When studying several ECG datasets, including the CPSC and Georgia ECG databases of the PhysioNet/Computing in Cardiology Challenge 2020, the new method achieved F1 score of 0.9693 and 0.8660, respectively. The results suggested a good generalization of the proposed method. Compared with several state-of-art frameworks, the proposed model with a shallow network of only 12 depths and asymmetric convolutions achieved the highest average F1 score. Extensive experimental studies proved that the proposed method has a high potential for AF prediction in clinical and particularly wearable applications.

An accurate and efficient method to train classifiers for atrial fibrillation detection in ECGs: Learning by asking better questions

BACKGROUND

An increasing number of wearables are capable of measuring electrocardiograms (ECGs), which may help in early detection of atrial fibrillation (AF). Therefore, many studies focus on automated detection of AF in ECGs. A major obstacle is the required amount of manually labelled data. This study aimed to provide an efficient and reliable method to train a classifier for AF detection using large datasets of real-life ECGs.

METHOD

Human-controlled semi-supervised learning was applied, consisting of two phases: the pre-training phase and the semi-automated training phase. During pre-training, an initial classifier was trained, which was used to predict the classes of new ECG segments in the semi-automated training phase. Based on the degree of certainty, segments were added to the training dataset automatically or after human validation. Thereafter, the classifier was retrained and this procedure was repeated. To test the model performance, a real-life telemetry dataset containing 3,846,564 30-s ECG segments of hospitalized patients (n = 476) and the CinC Challenge 2017 database were used.

RESULTS

After pre-training, the average F1-score on a hidden testing dataset was 89.0%. Furthermore, after the pre-training phase 68.0% of all segments in the hidden test set could be classified with an estimated probability of successful classification of 99%, providing an F1-score of 97.9% for these segments. During the semi-automated training phase, this F1-score showed little variation (97.3%-97.9% in the hidden test set), whilst the number of segments which could be automatically classified increased from 68.0% to 75.8% due to the enhanced training dataset. At the same time, the overall F1-score increased from 89.0% to 91.4%.

CONCLUSIONS

Human-validated semi-supervised learning makes training a classifier more time efficient without compromising on accuracy, hence this method might be valuable in the automated detection of AF in real-life ECGs.

Deep learning-based computer-aided heart sound analysis in children with left-to-right shunt congenital heart disease

OBJECTIVE

The purpose of this study was to explore a new algorithm model capable of leverage deep learning to screen and diagnose specific types of left-to-right shunt congenital heart disease (CHD) in children.

METHODS

Using deep learning, screening models were constructed to identify 884 heart sound recordings from children with left-to-right shunt CHD. The most suitable model for each type was summarized and compared with expert auscultation. An exploratory analysis was conducted to assess whether there were correlations between heart sounds and left ventricular ejection fraction (LVEF), pulmonary artery pressure, and malformation size.

RESULTS

The residual convolution recurrent neural network (RCRnet) classification model had higher accuracy than other models with respect to atrial septal defect (ASD), ventricular septum defect (VSD), patent ductus arteriosus (PDA) and combined CHD, and the best auscultation sites were determined to be the 4th, 5th, 2nd and 3rd auscultation areas, respectively. The diagnostic results of this model were better than those derived from expert auscultation, with sensitivity values of 0.932-1.000, specificity values of 0.944-0.997, precision values of 0.888-0.997 and accuracy values of 0.940-0.994. Absolute Pearson correlation coefficient values between heart sounds of the four types of CHD and LVEF, right ventricular systolic pressure (RVSP) and malformation size were all less than 0.3.

CONCLUSIONS

The RCRnet model can preliminarily determine types of left-to-right shunt CHD and improve diagnostic efficiency, which may provide a new choice algorithmic CHD screening in children.

Application of Dense Neural Networks for Detection of Atrial Fibrillation and Ranking of Augmented ECG Feature Set

Considering the significant burden to patients and healthcare systems globally related to atrial fibrillation (AF) complications, the early AF diagnosis is of crucial importance. In the view of prominent perspectives for fast and accurate point-of-care arrhythmia detection, our study optimizes an artificial neural network (NN) classifier and ranks the importance of enhanced 137 diagnostic ECG features computed from time and frequency ECG signal representations of short single-lead strips available in 2017 Physionet/CinC Challenge database. Based on hyperparameters' grid search of densely connected NN layers, we derive the optimal topology with three layers and 128, 32, 4 neurons per layer (DenseNet-3@128-32-4), which presents maximal F1-scores for classification of Normal rhythms (0.883, 5076 strips), AF (0.825, 758 strips), Other rhythms (0.705, 2415 strips), Noise (0.618, 279 strips) and total F1 relevant to the CinC Challenge of 0.804, derived by five-fold cross-validation. DenseNet-3@128-32-4 performs equally well with 137 to 32 features and presents tolerable reduction by about 0.03 to 0.06 points for limited input sets, including 8 and 16 features, respectively. The feature reduction is linked to effective application of a comprehensive method for computation of the feature map importance based on the weights of the activated neurons through the total path from input to specific output in DenseNet. The detailed analysis of 20 top-ranked ECG features with greatest importance to the detection of each rhythm and overall of all rhythms reveals DenseNet decision-making process, noticeably corresponding to the cardiologists' diagnostic point of view.

Recurrence Plot-Based Approach for Cardiac Arrhythmia Classification Using Inception-ResNet-v2

The present study addresses the cardiac arrhythmia (CA) classification problem using the deep learning (DL)-based method for electrocardiography (ECG) data analysis. Recently, various DL techniques have been utilized to classify arrhythmias, with one typical approach to developing a one-dimensional (1D) convolutional neural network (CNN) model to handle the ECG signals in the time domain. Although the CA classification in the time domain is very prevalent, current methods' performances are still not robust or satisfactory. This study aims to develop a solution for CA classification in two dimensions by introducing the recurrence plot (RP) combined with an Inception-ResNet-v2 network. The proposed method for nine types of CA classification was tested on the 1st China Physiological Signal Challenge 2018 dataset. During implementation, the optimal leads (lead II and lead aVR) were selected, and then 1D ECG segments were transformed into 2D texture images by the RP approach. These RP-based images as input signals were passed into the Inception-ResNet-v2 for CA classification. In the CPSC, Georgia, and the PTB_XL ECG databases of the PhysioNet/Computing in Cardiology Challenge 2020, the RP-based method achieved an average F1-score of 0.8521, 0.8529, and 0.8862, respectively. The results suggested the excellent generalization ability of the proposed method. To further assess the performance of the proposed method, we compared the 2D RP-image-based solution with the published 1D ECG-based works on the same dataset. Also, it was compared with two traditional ECG transform into 2D image methods, including the time waveform of the ECG recordings and time-frequency images based on continuous wavelet transform (CWT). The proposed method achieved the highest average F1-score of 0.844, with only two leads of the 12-lead ECG original data, which outperformed other works. Therefore, the promising results indicate that the 2D RP-based method has a high clinical potential for CA classification using fewer lead ECG signals.

How machine learning is impacting research in atrial fibrillation: implications for risk prediction and future management

There has been an exponential growth of artificial intelligence (AI) and machine learning (ML) publications aimed at advancing our understanding of atrial fibrillation (AF), which has been mainly driven by the confluence of two factors: the advances in deep neural networks (DeepNNs) and the availability of large, open access databases. It is observed that most of the attention has centred on applying ML for dvsetecting AF, particularly using electrocardiograms (ECGs) as the main data modality. Nearly a third of them used DeepNNs to minimize or eliminate the need for transforming the ECGs to extract features prior to ML modelling; however, we did not observe a significant advantage in following this approach. We also found a fraction of studies using other data modalities, and others centred in aims, such as risk prediction, AF management, and others. From the clinical perspective, AI/ML can help expand the utility of AF detection and risk prediction, especially for patients with additional comorbidities. The use of AI/ML for detection and risk prediction into applications and smart mobile health (mHealth) technology would enable ‘real time’ dynamic assessments. AI/ML could also adapt to treatment changes over time, as well as incident risk factors. Incorporation of a dynamic AI/ML model into mHealth technology would facilitate ‘real time’ assessment of stroke risk, facilitating mitigation of modifiable risk factors (e.g. blood pressure control). Overall, this would lead to an improvement in clinical care for patients with AF.

Digital biomarkers and algorithms for detection of atrial fibrillation using surface electrocardiograms: A systematic review

AIMS

Automated detection of atrial fibrillation (AF) in continuous rhythm registrations is essential in order to prevent complications and optimize treatment of AF. Many algorithms have been developed to detect AF in surface electrocardiograms (ECGs) during the past few years. The aim of this systematic review is to gain more insight into these available classification methods by discussing previously used digital biomarkers and algorithms and make recommendations for future research.

METHODS

On the 14^th of September 2020, the PubMed database was searched for articles focusing on algorithms for AF detection in ECGs using the MeSH terms Atrial Fibrillation, Electrocardiography and Algorithms. Articles which solely focused on differentiation of types of rhythm disorders or prediction of AF termination were excluded.

RESULTS

The search resulted in 451 articles, of which 130 remained after full-text screening. Not only did the amount of research on methods for AF detection increase over the past years, but a trend towards more complex classification methods is observed. Furthermore, three different types of features can be distinguished: atrial features, ventricular features, and signal features. Although AF is an atrial disease, only 22% of the described methods use atrial features.

CONCLUSION

More and more studies focus on improving accuracy of classification methods for AF in ECGs. As a result, algorithms become increasingly complex and less well interpretable. Only a few studies focus on detecting atrial activity in the ECG. Developing innovative methods focusing on detection of atrial activity might provide accurate classifiers without compromising on transparency.

Novel Analgesic Index for Postoperative Pain Assessment Based on a Photoplethysmographic Spectrogram and Convolutional Neural Network: Observational Study

BACKGROUND

Although commercially available analgesic indices based on biosignal processing have been used to quantify nociception during general anesthesia, their performance is low in conscious patients. Therefore, there is a need to develop a new analgesic index with improved performance to quantify postoperative pain in conscious patients.

OBJECTIVE

This study aimed to develop a new analgesic index using photoplethysmogram (PPG) spectrograms and a convolutional neural network (CNN) to objectively assess pain in conscious patients.

METHODS

PPGs were obtained from a group of surgical patients for 6 minutes both in the absence (preoperatively) and in the presence (postoperatively) of pain. Then, the PPG data of the latter 5 minutes were used for analysis. Based on the PPGs and a CNN, we developed a spectrogram-CNN index for pain assessment. The area under the curve (AUC) of the receiver-operating characteristic curve was measured to evaluate the performance of the 2 indices.

RESULTS

PPGs from 100 patients were used to develop the spectrogram-CNN index. When there was pain, the mean (95% CI) spectrogram-CNN index value increased significantly-baseline: 28.5 (24.2-30.7) versus recovery area: 65.7 (60.5-68.3); P<.01. The AUC and balanced accuracy were 0.76 and 71.4%, respectively. The spectrogram-CNN index cutoff value for detecting pain was 48, with a sensitivity of 68.3% and specificity of 73.8%.

CONCLUSIONS

Although there were limitations to the study design, we confirmed that the spectrogram-CNN index can efficiently detect postoperative pain in conscious patients. Further studies are required to assess the spectrogram-CNN index's feasibility and prevent overfitting to various populations, including patients under general anesthesia.

TRIAL REGISTRATION

Clinical Research Information Service KCT0002080; https://cris.nih.go.kr/cris/search/search_result_st01.jsp?seq=6638.

Path to precision: prevention of post-operative atrial fibrillation

Development of post-operative atrial fibrillation (POAF) following open-heart surgery is a significant clinical and economic burden. Despite advancements in medical therapies, the incidence of POAF remains elevated at 25-40%. Early work focused on detecting arrhythmias from electrocardiograms as well as identifying pre-operative risk factors from medical records. However, further progress has been stagnant, and a deeper understanding of pathogenesis and significant influences is warranted. With the advent of more complex machine learning (ML) algorithms and high-throughput sequencing, we have an unprecedented ability to capture and predict POAF in real-time. Integration of multimodal heterogeneous data and application of ML can generate a paradigm shift for diagnosis and treatment. This will require a concerted effort to consolidate and streamline real-time data. Herein, we will review the current literature and emerging opportunities aimed at predictive targets and new insights into the mechanisms underlying long-term sequelae of POAF.

Opportunities and challenges of deep learning methods for electrocardiogram data: A systematic review

BACKGROUND

The electrocardiogram (ECG) is one of the most commonly used diagnostic tools in medicine and healthcare. Deep learning methods have achieved promising results on predictive healthcare tasks using ECG signals.

OBJECTIVE

This paper presents a systematic review of deep learning methods for ECG data from both modeling and application perspectives.

METHODS

We extracted papers that applied deep learning (deep neural network) models to ECG data that were published between January 1st of 2010 and February 29th of 2020 from Google Scholar, PubMed, and the Digital Bibliography & Library Project. We then analyzed each article according to three factors: tasks, models, and data. Finally, we discuss open challenges and unsolved problems in this area.

RESULTS

The total number of papers extracted was 191. Among these papers, 108 were published after 2019. Different deep learning architectures have been used in various ECG analytics tasks, such as disease detection/classification, annotation/localization, sleep staging, biometric human identification, and denoising.

CONCLUSION

The number of works on deep learning for ECG data has grown explosively in recent years. Such works have achieved accuracy comparable to that of traditional feature-based approaches and ensembles of multiple approaches can achieve even better results. Specifically, we found that a hybrid architecture of a convolutional neural network and recurrent neural network ensemble using expert features yields the best results. However, there are some new challenges and problems related to interpretability, scalability, and efficiency that must be addressed. Furthermore, it is also worth investigating new applications from the perspectives of datasets and methods.

SIGNIFICANCE

This paper summarizes existing deep learning research using ECG data from multiple perspectives and highlights existing challenges and problems to identify potential future research directions.

Fully Convolutional Deep Neural Networks with Optimized Hyperparameters for Detection of Shockable and Non-Shockable Rhythms

Deep neural networks (DNN) are state-of-the-art machine learning algorithms that can be learned to self-extract significant features of the electrocardiogram (ECG) and can generally provide high-output diagnostic accuracy if subjected to robust training and optimization on large datasets at high computational cost. So far, limited research and optimization of DNNs in shock advisory systems is found on large ECG arrhythmia databases from out-of-hospital cardiac arrests (OHCA). The objective of this study is to optimize the hyperparameters (HPs) of deep convolutional neural networks (CNN) for detection of shockable (Sh) and nonshockable (NSh) rhythms, and to validate the best HP settings for short and long analysis durations (2-10 s). Large numbers of (Sh + NSh) ECG samples were used for training (720 + 3170) and validation (739 + 5921) from Holters and defibrillators in OHCA. An end-to-end deep CNN architecture was implemented with one-lead raw ECG input layer (5 s, 125 Hz, 2.5 uV/LSB), configurable number of 5 to 23 hidden layers and output layer with diagnostic probability p ∈ [0: Sh,1: NSh]. The hidden layers contain N convolutional blocks × 3 layers (Conv1D (filters = Fi, kernel size = Ki), max-pooling (pool size = 2), dropout (rate = 0.3)), one global max-pooling and one dense layer. Random search optimization of HPs = {N, Fi, Ki}, i = 1, … N in a large grid of N = [1, 2, … 7], Fi = [5;50], Ki = [5;100] was performed. During training, the model with maximal balanced accuracy BAC = (Sensitivity + Specificity)/2 over 400 epochs was stored. The optimization principle is based on finding the common HPs space of a few top-ranked models and prediction of a robust HP setting by their median value. The optimal models for 1-7 CNN layers were trained with different learning rates LR = [10-5; 10-2] and the best model was finally validated on 2-10 s analysis durations. A number of 4216 random search models were trained. The optimal models with more than three convolutional layers did not exhibit substantial differences in performance BAC = (99.31-99.5%). Among them, the best model was found with {N = 5, Fi = {20, 15, 15, 10, 5}, Ki = {10, 10, 10, 10, 10}, 7521 trainable parameters} with maximal validation performance for 5-s analysis (BAC = 99.5%, Se = 99.6%, Sp = 99.4%) and tolerable drop in performance (<2% points) for very short 2-s analysis (BAC = 98.2%, Se = 97.6%, Sp = 98.7%). DNN application in future-generation shock advisory systems can improve the detection performance of Sh and NSh rhythms and can considerably shorten the analysis duration complying with resuscitation guidelines for minimal hands-off pauses.

A Comprehensive Study of Complexity and Performance of Automatic Detection of Atrial Fibrillation: Classification of Long ECG Recordings Based on the PhysioNet Computing in Cardiology Challenge 2017

OBJECTIVE

The 2017 PhysioNet/CinC Challenge focused on automatic classification of atrial fibrillation (AF) in short ECGs. This study aimed to evaluate the use of the data and results from the challenge for detection of AF in longer ECGs, taken from three other PhysioNet datasets.

APPROACH

The used data-driven models were based on features extracted from ECG recordings, calculated according to three solutions from the challenge. A Random Forest classifier was trained with the data from the challenge. The performance was evaluated on all non-overlapping 30 s segments in all recordings from three MIT-BIH datasets. Fifty-six models were trained using different feature sets, both before and after applying three feature reduction techniques.

MAIN RESULTS

Based on rhythm annotations, the AF proportion was 0.00 in the MIT-BIH Normal Sinus Rhythm (N = 46083 segments), 0.10 in the MIT-BIH Arrhythmia (N = 2880), and 0.41 in the MIT-BIH Atrial Fibrillation (N = 28104) dataset. For the best performing model, the corresponding detected proportions of AF were 0.00, 0.11 and 0.36 using all features, and 0.01, 0.10 and 0.38 when using the 15 best performing features.

SIGNIFICANCE

The results obtained on the MIT-BIH datasets indicate that the training data and solutions from the 2017 Physionet/Cinc Challenge can be useful tools for developing robust AF detectors also in longer ECG recordings, even when using a low number of carefully selected features. The use of feature selection allows significantly reducing the number of features while preserving the classification performance, which can be important when building low-complexity AF classifiers on ECG devices with constrained computational and energy resources.

Cardiac arrhythmia detection using deep learning: A review

Due to its simplicity and low cost, analyzing an electrocardiogram (ECG) is the most common technique for detecting cardiac arrhythmia. The massive amount of ECG data collected every day, in home and hospital, may preclude data review by human operators/technicians. Therefore, several methods are proposed for either fully automatic arrhythmia detection or event selection for further verification by human experts. Traditional machine learning approaches have made significant progress in the past years. However, those methods rely on hand-crafted feature extraction, which requires in-depth domain knowledge and preprocessing of the signal (e.g., beat detection). This, plus the high variability in wave morphology among patients and the presence of noise, make it challenging for computerized interpretation to achieve high accuracy. Recent advances in deep learning make it possible to perform automatic high-level feature extraction and classification. Therefore, deep learning approaches have gained interest in arrhythmia detection. In this work, we reviewed the recent advancement of deep learning methods for automatic arrhythmia detection. We summarized existing literature from five aspects: utilized dataset, application, type of input data, model architecture, and performance evaluation. We also reported limitations of reviewed papers and potential future opportunities.

Ventricular ectopic beat detection using a wavelet transform and a convolutional neural network

OBJECTIVE

Ventricular contractions in healthy individuals normally follow the contractions of atria to facilitate more efficient pump action and cardiac output. With a ventricular ectopic beat (VEB), volume within the ventricles are pumped to the body's vessels before receiving blood from atria, thus causing inefficient blood circulation. VEBs tend to cause perturbations in the instantaneous heart rate time series, making the analysis of heart rate variability inappropriate around such events, or requiring special treatment (such as signal averaging). Moreover, VEB frequency can be indicative of life-threatening problems. However, VEBs can often mimic artifacts both in morphology and timing. Identification of VEBs is therefore an important unsolved problem. The aim of this study is to introduce a method of wavelet transform in combination with deep learning network for the classification of VEBs.

APPROACH

We proposed a method to automatically discriminate VEB beats from other beats and artifacts with the use of wavelet transform of the electrocardiogram (ECG) and a convolutional neural network (CNN). Three types of wavelets (Morlet wavelet, Paul wavelet and Gaussian derivative) were used to transform segments of single-channel (1D) ECG waveforms to two-dimensional (2D) time-frequency 'images'. The 2D time-frequency images were then passed into a CNN to optimize the convolutional filters and classification. Ten-fold cross validation was used to evaluate the approach on the MIT-BIH arrhythmia database (MIT-BIH). The American Heart Association (AHA) database was then used as an independent dataset to evaluate the trained network.

MAIN RESULTS

Ten-fold cross validation results on MIT-BIH showed that the proposed algorithm with Paul wavelet achieved an overall F1 score of 84.94% and accuracy of 97.96% on out of sample validation. Independent test on AHA resulted in an F1 score of 84.96% and accuracy of 97.36%.

SIGNIFICANCE

The trained network possessed exceptional transferability across databases and generalization to unseen data.