Automatic Triage of 12-Lead ECGs Using Deep Convolutional Neural Networks

Comparing multi-class classifier performance by multi-class ROC analysis: A nonparametric approach

The area under the Receiver Operating Characteristic (ROC) curve (AUC) is a standard metric for quantifying and comparing binary classifiers. Real world applications often require classification into multiple (more than two) classes. For multi-class classifiers that produce class membership scores, a popular multi-class AUC (MAUC) variant is to average the pairwise AUC values [1]. Due to the complicated correlation patterns, the variance of MAUC is often estimated numerically using resampling techniques. This work is a generalization of DeLong's nonparameteric approach for binary AUC analysis [2] to MAUC. We first derive the closed-form expression of the covariance matrix of the pairwise AUCs within a single MAUC. Then by dropping higher order terms, we obtain an approximate covariance matrix with a compact, matrix factorization form, which then serves as the basis for variance estimation of a single MAUC. We further extend this approach to estimate the covariance of correlated MAUCs that arise from multiple competing classifiers. For the special case of binary correlated AUCs, our results coincide with that of DeLong. Our numerical studies confirm the accuracy of the variance and covariance estimates. We provide the source code of the proposed covariance estimation of correlated MAUCs on GitHub (https://tinyurl.com/euj6wvsz) for its easy adoption by machine learning and statistical analysis packages to quantify and compare multi-class classifiers.

Detection of acute coronary occlusion with a novel mobile electrocardiogram device: a pilot study

Aims

Many portable electrocardiogram (ECG) devices have been developed to monitor patients at home, but the majority of these devices are single lead and only intended for rhythm disorders. We developed the miniECG, a smartphone-sized portable device with four dry electrodes capable of recording a high-quality multi-lead ECG by placing the device on the chest. The aim of our study was to investigate the ability of the miniECG to detect occlusive myocardial infarction (OMI) in patients with chest pain.

Methods and results

Patients presenting with acute chest pain at the emergency department of the University Medical Center Utrecht or Meander Medical Center, between May 2021 and February 2022, were included in the study. The clinical 12-lead ECG and the miniECG before coronary intervention were recorded. The recordings were evaluated by cardiologists and compared the outcome of the coronary angiography, if performed. A total of 369 patients were measured with the miniECG, 46 of whom had OMI. The miniECG detected OMI with a sensitivity and specificity of 65 and 92%, compared with 83 and 90% for the 12-lead ECG. Sensitivity of the miniECG was similar for different culprit vessels.

Conclusion

The miniECG can record a multi-lead ECG and rule-in ST-segment deviation in patients with occluded or near-occluded coronary arteries from different culprit vessels without many false alarms. Further research is required to add automated analysis to the recordings and to show feasibility to use the miniECG by patients at home.

Automatic triage of twelve-lead electrocardiograms using deep convolutional neural networks: a first implementation study

Aims

Expert knowledge to correctly interpret electrocardiograms (ECGs) is not always readily available. An artificial intelligence (AI)-based triage algorithm (DELTAnet), able to support physicians in ECG prioritization, could help reduce current logistic burden of overreading ECGs and improve time to treatment for acute and life-threatening disorders. However, the effect of clinical implementation of such AI algorithms is rarely investigated.

Methods and results

Adult patients at non-cardiology departments who underwent ECG testing as a part of routine clinical care were included in this prospective cohort study. DELTAnet was used to classify 12-lead ECGs into one of the following triage classes: normal, abnormal not acute, subacute, and acute. Performance was compared with triage classes based on the final clinical diagnosis. Moreover, the associations between predicted classes and clinical outcomes were investigated. A total of 1061 patients and ECGs were included. Performance was good with a mean concordance statistic of 0.96 (95% confidence interval 0.95-0.97) when comparing DELTAnet with the clinical triage classes. Moreover, zero ECGs that required a change in policy or referral to the cardiologist were missed and there was a limited number of cases predicted as acute that did not require follow-up (2.6%).

Conclusion

This study is the first to prospectively investigate the impact of clinical implementation of an ECG-based AI triage algorithm. It shows that DELTAnet is efficacious and safe to be used in clinical practice for triage of 12-lead ECGs in non-cardiology hospital departments.

Deep learning-based regional ECG diagnosis platform

OBJECTIVE

To enable the intelligent diagnosis of a variety of common Electrocardiogram (ECG), we investigate the deep learning-based ECG diagnosis system.

METHODS

From January 2015 to December 2019, four consecutive years of 100,120 conventional 12-lead ECG data were collected in our hospital. Utilizing this dataset, we constructed a deep learning model designed to intelligently diagnose prevalent ECG anomalies by employing a multi-task learning framework. The system performance was evaluated using various metrics, including sensitivity, specificity, negative predictive value, positive predictive value, and so forth. Additionally, we employed an ECG intelligent diagnostic platform for clinical application to undertake real-time online analysis of 2500 conventional 12-lead ECG samples in June 2020, aiming to validate our model. At this stage, we compared the performance of our model against the traditional manual identification method.

RESULTS

The efficacy of the ECG intelligent diagnostic model was notably high for common and straightforward ECG patterns, such as sinus rhythm (F1 = 98.01%), sinus tachycardia (F1 = 96.26%), sinus bradycardia (F1 = 94.88%), and a normal electrocardiogram (F1 = 91.71%), as well as for Premature Ventricular Contractions (F1 = 91.62%). Nevertheless, when diagnosing rarer and more intricate ECG anomalies, the system requires an increased number of samples to refine the deep learning models. During the validation stage, our model exhibited better efficiency in terms of accuracy, labor time and labor cost when compared to the manual identification approach.

CONCLUSIONS

Our deep learning-driven intelligent ECG diagnostic model clearly demonstrates significant clinical utility. The integrated artificial intelligence diagnosis system not only has the potential to augment physicians in their diagnostic processes but also offers a viable avenue to reduce associated labor costs.

Machine learning in cardiology: Clinical application and basic research

Machine learning is a subfield of artificial intelligence. The quality and versatility of machine learning have been rapidly improving and playing a critical role in many aspects of social life. This trend is also observed in the medical field. Generally, there are three main types of machine learning: supervised, unsupervised, and reinforcement learning. Each type of learning is adequately selected for the purpose and type of data. In the field of medicine, various types of information are collected and used, and research using machine learning is becoming increasingly relevant. Many clinical studies are conducted using electronic health and medical records, including in the cardiovascular area. Machine learning has also been applied in basic research. Machine learning has been widely used for several types of data analysis, such as clustering of microarray analysis and RNA sequence analysis. Machine learning is essential for genome and multi-omics analyses. This review summarizes the recent advancements in the use of machine learning in clinical applications and basic cardiovascular research.

Electrocardiogram monitoring as a predictor of neurological and survival outcomes in patients with out-of-hospital cardiac arrest: a single-center retrospective observational study

Introduction

This study hypothesized that monitoring electrocardiogram (ECG) waveforms in patients with out-of-hospital cardiac arrest (OHCA) could have predictive value for survival or neurological outcomes. We aimed to establish a new prognostication model based on the single variable of monitoring ECG waveforms in patients with OHCA using machine learning (ML) techniques.

Methods

This observational retrospective study included successfully resuscitated patients with OHCA aged ≥ 18 years admitted to an intensive care unit in Japan between April 2010 and April 2020. Waveforms from ECG monitoring for 1 h after admission were obtained from medical records and examined. Based on the open-access PTB-XL dataset, a large publicly available 12-lead ECG waveform dataset, we built an ML-supported premodel that transformed the II-lead waveforms of the monitoring ECG into diagnostic labels. The ECG diagnostic labels of the patients in this study were analyzed for prognosis using another model supported by ML. The endpoints were favorable neurological outcomes (cerebral performance category 1 or 2) and survival to hospital discharge.

Results

In total, 590 patients with OHCA were included in this study and randomly divided into 3 groups (training set, n = 283; validation set, n = 70; and test set, n = 237). In the test set, our ML model predicted neurological and survival outcomes, with the highest areas under the receiver operating characteristic curves of 0.688 (95% CI: 0.682-0.694) and 0.684 (95% CI: 0.680-0.689), respectively.

Conclusion

Our ML predictive model showed that monitoring ECG waveforms soon after resuscitation could predict neurological and survival outcomes in patients with OHCA.

Optimization of Arrhythmia-based ECG-lead Selection for Computer-interpreted Heart Rhythm Classification

The 12-lead ECG only has 8 independent ECG leads, which leads to diagnostic redundancy when using all 12 leads for heart arrhythmias classification. We have previously developed a deep learning (DL)-based computer-interpreted ECG (CIE) approach to identify an optimal 4-lead ECG subset for classifying heart arrhythmias. However, the clinical diagnostic criteria of cardiac arrhythmia types are often lead-specific, so this study is going to explore the selection of arrhythmia-based ECG-lead subsets rather than one general optimal ECG-lead subset, which could improve the classification performance for the CIE. The DL-based CIE model previously developed was used to learn 4 common types of heart arrhythmias (LBBB, RBBB, AF, and I-AVB) for identifying corresponding optimal ECG-lead subsets. A public dataset that splits into training (approx. 70%), validation (approx. 15%), and test (approx. 15%) sets from the PhysioNet Cardiology Challenge 2020 was used to explore the study. The results demonstrated that the DL-based CIE model identified an optimal ECG-lead subset for each arrhythmia: I, II, aVR, aVL, V1, V3, and V5 for I-AVB; I, II, aVR, and V3 for AF; I, II, aVR, aVF, V1, V3, and V4 for LBBB; and I, II, III, aVR, V1, V4, and V6 for RBBB. For each arrhythmia classification, the DL-based CIE model using the optimal ECG-lead subset significantly outperformed the model using the full 12-lead ECG set on the validation set and on the external test dataset.The results support the hypothesis that using an optimal ECG-lead subset instead of the full 12-lead ECG set can improve the classification performance of a specific arrhythmia when using the DL-based CIE approach.Clinical Relevance- Using an arrhythmia-based optimal ECG-lead subset, the classification performance of a deep-learning-based model can be achieved without loss of accuracy in comparison with the full 12-lead set (p<0.05).

Convolution Neural Network Algorithm for Shockable Arrhythmia Classification Within a Digitally Connected Automated External Defibrillator

Background Diagnosis of shockable rhythms leading to defibrillation remains integral to improving out-of-hospital cardiac arrest outcomes. New machine learning techniques have emerged to diagnose arrhythmias on ECGs. In out-of-hospital cardiac arrest, an algorithm within an automated external defibrillator is the major determinant to deliver defibrillation. This study developed and validated the performance of a convolution neural network (CNN) to diagnose shockable arrhythmias within a novel, miniaturized automated external defibrillator. Methods and Results There were 26 464 single-lead ECGs that comprised the study data set. ECGs of 7-s duration were retrospectively adjudicated by 3 physician readers (N=18 total readers). After exclusions (N=1582), ECGs were divided into training (N=23 156), validation (N=721), and test data sets (N=1005). CNN performance to diagnose shockable and nonshockable rhythms was reported with area under the receiver operating characteristic curve analysis, F1, and sensitivity and specificity calculations. The duration for the CNN to output was reported with the algorithm running within the automated external defibrillator. Internal and external validation analyses included CNN performance among arrhythmias, often mistaken for shockable rhythms, and performance among ECGs modified with noise to mimic artifacts. The CNN algorithm achieved an area under the receiver operating characteristic curve of 0.995 (95% CI, 0.990-1.0), sensitivity of 98%, and specificity of 100% to diagnose shockable rhythms. The F1 scores were 0.990 and 0.995 for shockable and nonshockable rhythms, respectively. After input of a 7-s ECG, the CNN generated an output in 383±29 ms (total time of 7.383 s). The CNN outperformed adjudicators in classifying atrial arrhythmias as nonshockable (specificity of 99.3%-98.1%) and was robust against noise artifacts (area under the receiver operating characteristic curve range, 0.871-0.999). Conclusions We demonstrate high diagnostic performance of a CNN algorithm for shockable and nonshockable rhythm arrhythmia classifications within a digitally connected automated external defibrillator. Registration URL: https://clinicaltrials.gov/ct2/show/NCT03662802; Unique identifier: NCT03662802.

CardioLabelNet: An uncertainty estimation using fuzzy for abnormalities detection in ECG

Electrocardiography (ECG) abnormalities are evaluated through several automatic detection methods. Primarily, real-world ECG data are digital signals those are stored in the form of images in hospitals. Also, the existing automated detection technique eliminates the cardiac pattern that is abnormal and it is difficult to multiple abnormalities at some instances. To address those issues in this paper conventional ECG image automated techniques CardioLabelNet model is proposed. The proposed model incorporates two stages for image abnormality detection. At first fuzzy membership is performed in the image for computation of uncertainty. In second stage, classification is performed for computation of abnormal activity. The proposed CardioLabelNet collect ECG image data set for the uncertainty estimation while taking the account of various image classes which includes the global and local entropy of image pixels. For each waveform, uncertainties are calculated on the basis of global entropy. The computation of uncertainty in the images is performed with the fuzzy membership function. The spatial likelihood estimation of a fuzzy weighted membership function is used to calculate local entropy. Upon completion of fuzzification, classification is performed for the detection of normal and abnormal patterns in the ECG signal images. Through integration of stacked architecture model classification is performed for ECG images. The proffered algorithm performance is calculated in terms of accuracy for segmentation, Dice similarity coefficient, and partition entropy. Additionally, classification parameters accuracy sensitivity, specificity, and AUC are evaluated. The proposed approach outperforms the existing methodology, according to the results of a comparative analysis.

Improving explainability of deep neural network-based electrocardiogram interpretation using variational auto-encoders. EUROPEAN HEART JOURNAL. DIGITAL HEALTH 2022;3:390-404. [PMID: 36712164 PMCID: PMC9707974 DOI: 10.1093/ehjdh/ztac038]

Aims

Deep neural networks (DNNs) perform excellently in interpreting electrocardiograms (ECGs), both for conventional ECG interpretation and for novel applications such as detection of reduced ejection fraction (EF). Despite these promising developments, implementation is hampered by the lack of trustworthy techniques to explain the algorithms to clinicians. Especially, currently employed heatmap-based methods have shown to be inaccurate.

Methods and results

We present a novel pipeline consisting of a variational auto-encoder (VAE) to learn the underlying factors of variation of the median beat ECG morphology (the FactorECG), which are subsequently used in common and interpretable prediction models. As the ECG factors can be made explainable by generating and visualizing ECGs on both the model and individual level, the pipeline provides improved explainability over heatmap-based methods. By training on a database with 1.1 million ECGs, the VAE can compress the ECG into 21 generative ECG factors, most of which are associated with physiologically valid underlying processes. Performance of the explainable pipeline was similar to 'black box' DNNs in conventional ECG interpretation [area under the receiver operating curve (AUROC) 0.94 vs. 0.96], detection of reduced EF (AUROC 0.90 vs. 0.91), and prediction of 1-year mortality (AUROC 0.76 vs. 0.75). Contrary to the 'black box' DNNs, our pipeline provided explainability on which morphological ECG changes were important for prediction. Results were confirmed in a population-based external validation dataset.

Conclusions

Future studies on DNNs for ECGs should employ pipelines that are explainable to facilitate clinical implementation by gaining confidence in artificial intelligence and making it possible to identify biased models.

Life-threatening ventricular arrhythmia prediction in patients with dilated cardiomyopathy using explainable electrocardiogram-based deep neural networks

Aims

While electrocardiogram (ECG) characteristics have been associated with life-threatening ventricular arrhythmias (LTVA) in dilated cardiomyopathy (DCM), they typically rely on human-derived parameters. Deep neural networks (DNNs) can discover complex ECG patterns, but the interpretation is hampered by their ‘black-box’ characteristics. We aimed to detect DCM patients at risk of LTVA using an inherently explainable DNN.

Methods and results

In this two-phase study, we first developed a variational autoencoder DNN on more than 1 million 12-lead median beat ECGs, compressing the ECG into 21 different factors (F): FactorECG. Next, we used two cohorts with a combined total of 695 DCM patients and entered these factors in a Cox regression for the composite LTVA outcome, which was defined as sudden cardiac arrest, spontaneous sustained ventricular tachycardia, or implantable cardioverter-defibrillator treated ventricular arrhythmia. Most patients were male (n = 442, 64%) with a median age of 54 years [interquartile range (IQR) 44–62], and median left ventricular ejection fraction of 30% (IQR 23–39). A total of 115 patients (16.5%) reached the study outcome. Factors F₈ (prolonged PR-interval and P-wave duration, P < 0.005), F₁₅ (reduced P-wave height, P = 0.04), F₂₅ (increased right bundle branch delay, P = 0.02), F₂₇ (P-wave axis P < 0.005), and F₃₂ (reduced QRS-T voltages P = 0.03) were significantly associated with LTVA.

Conclusion

Inherently explainable DNNs can detect patients at risk of LTVA which is mainly driven by P-wave abnormalities.

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.

Identification of Characteristic Points in Multivariate Physiological Signals by Sensor Fusion and Multi-Task Deep Networks

Identification of characteristic points in physiological signals, such as the peak of the R wave in the electrocardiogram and the peak of the systolic wave of the photopletismogram, is a fundamental step for the quantification of clinical parameters, such as the pulse transit time. In this work, we presented a novel neural architecture, called eMTUnet, to automate point identification in multivariate signals acquired with a chest-worn device. The eMTUnet consists of a single deep network capable of performing three tasks simultaneously: (i) localization in time of characteristic points (labeling task), (ii) evaluation of the quality of signals (classification task); (iii) estimation of the reliability of classification (reliability task). Preliminary results in overnight monitoring showcased the ability to detect characteristic points in the four signals with a recall index of about 1.00, 0.90, 0.90, and 0.80, respectively. The accuracy of the signal quality classification was about 0.90, on average over four different classes. The average confidence of the correctly classified signals, against the misclassifications, was 0.93 vs. 0.52, proving the worthiness of the confidence index, which may better qualify the point identification. From the achieved outcomes, we point out that high-quality segmentation and classification are both ensured, which brings the use of a multi-modal framework, composed of wearable sensors and artificial intelligence, incrementally closer to clinical translation.

Deep neural networks reveal novel sex-specific electrocardiographic features relevant for mortality risk

Aims

Incorporation of sex in study design can lead to discoveries in medical research. Deep neural networks (DNNs) accurately predict sex based on the electrocardiogram (ECG) and we hypothesized that misclassification of sex is an important predictor for mortality. Therefore, we first developed and validated a DNN that classified sex based on the ECG and investigated the outcome. Second, we studied ECG drivers of DNN-classified sex and mortality.

Methods and results

A DNN was trained to classify sex based on 131 673 normal ECGs. The algorithm was validated on internal (68 500 ECGs) and external data sets (3303 and 4457 ECGs). The survival of sex (mis)classified groups was investigated using time-to-event analysis and sex-stratified mediation analysis of ECG features. The DNN successfully distinguished female from male ECGs {internal validation: area under the curve (AUC) 0.96 [95% confidence interval (CI): 0.96, 0.97]; external validations: AUC 0.89 (95% CI: 0.88, 0.90), 0.94 (95% CI: 0.93, 0.94)}. Sex-misclassified individuals (11%) had a 1.4 times higher mortality risk compared with correctly classified peers. The ventricular rate was the strongest mediating ECG variable (41%, 95% CI: 31%, 56%) in males, while the maximum amplitude of the ST segment was strongest in females (18%, 95% CI: 11%, 39%). Short QRS duration was associated with higher mortality risk.

Conclusion

Deep neural networks accurately classify sex based on ECGs. While the proportion of ECG-based sex misclassifications is low, it is an interesting biomarker. Investigation of the causal pathway between misclassification and mortality uncovered new ECG features that might be associated with mortality. Increased emphasis on sex as a biological variable in artificial intelligence is warranted.

Electrocardiogram-based mortality prediction in patients with COVID-19 using machine learning

Background and purpose

The electrocardiogram (ECG) is frequently obtained in the work-up of COVID-19 patients. So far, no study has evaluated whether ECG-based machine learning models have added value to predict in-hospital mortality specifically in COVID-19 patients.

Methods

Using data from the CAPACITY-COVID registry, we studied 882 patients admitted with COVID-19 across seven hospitals in the Netherlands. Raw format 12-lead ECGs recorded within 72 h of admission were studied. With data from five hospitals (n = 634), three models were developed: (a) a logistic regression baseline model using age and sex, (b) a least absolute shrinkage and selection operator (LASSO) model using age, sex and human annotated ECG features, and (c) a pre-trained deep neural network (DNN) using age, sex and the raw ECG waveforms. Data from two hospitals (n = 248) was used for external validation.

Results

Performances for models a, b and c were comparable with an area under the receiver operating curve of 0.73 (95% confidence interval [CI] 0.65–0.79), 0.76 (95% CI 0.68–0.82) and 0.77 (95% CI 0.70–0.83) respectively. Predictors of mortality in the LASSO model were age, low QRS voltage, ST depression, premature atrial complexes, sex, increased ventricular rate, and right bundle branch block.

Conclusion

This study shows that the ECG-based prediction models could be helpful for the initial risk stratification of patients diagnosed with COVID-19, and that several ECG abnormalities are associated with in-hospital all-cause mortality of COVID-19 patients. Moreover, this proof-of-principle study shows that the use of pre-trained DNNs for ECG analysis does not underperform compared with time-consuming manual annotation of ECG features.

Supplementary Information

The online version of this article (10.1007/s12471-022-01670-2) contains supplementary material, which is available to authorized users.

Patient contrastive learning: A performant, expressive, and practical approach to electrocardiogram modeling

Supervised machine learning applications in health care are often limited due to a scarcity of labeled training data. To mitigate the effect of small sample size, we introduce a pre-training approach, Patient Contrastive Learning of Representations (PCLR), which creates latent representations of electrocardiograms (ECGs) from a large number of unlabeled examples using contrastive learning. The resulting representations are expressive, performant, and practical across a wide spectrum of clinical tasks. We develop PCLR using a large health care system with over 3.2 million 12-lead ECGs and demonstrate that training linear models on PCLR representations achieves a 51% performance increase, on average, over six training set sizes and four tasks (sex classification, age regression, and the detection of left ventricular hypertrophy and atrial fibrillation), relative to training neural network models from scratch. We also compared PCLR to three other ECG pre-training approaches (supervised pre-training, unsupervised pre-training with an autoencoder, and pre-training using a contrastive multi ECG-segment approach), and show significant performance benefits in three out of four tasks. We found an average performance benefit of 47% over the other models and an average of a 9% performance benefit compared to best model for each task. We release PCLR to enable others to extract ECG representations at https://github.com/broadinstitute/ml4h/tree/master/model_zoo/PCLR.

ECGs are a rich source of cardiac health information. Many recent works have shown that deep learning can extract new information from ECGs when there are a sufficient number of labeled data. However, when there are not enough labeled data or a clinician scientist does not have the resources to train a deep learning model from scratch, options are limited. We introduce Patient Contrastive Learning of Representations (PCLR), an approach to train a neural network that extracts representations of ECGs. The only labels required to train PCLR are which ECG comes from which patient. The resulting ECG representations can be used directly in linear models for new tasks without needing to finetune the neural network. We show PCLR is better than using a set of handpicked features for four tasks, and better than three other deep learning approaches for three out of four tasks evaluated. Furthermore, PCLR is better than training a neural network from scratch when training data are limited. PCLR is one of the first attempts at releasing and evaluating a pre-trained ECG model with the purpose of accelerating deep learning ECG research.

Optimal ECG-lead selection increases generalizability of deep learning on ECG abnormality classification

Deep learning (DL) has achieved promising performance in detecting common abnormalities from the 12-lead electrocardiogram (ECG). However, diagnostic redundancy exists in the 12-lead ECG, which could impose a systematic overfitting on DL, causing poor generalization. We, therefore, hypothesized that finding an optimal lead subset of the 12-lead ECG to eliminate the redundancy would help improve the generalizability of DL-based models. In this study, we developed and evaluated a DL-based model that has a feature extraction stage, an ECG-lead subset selection stage and a decision-making stage to automatically interpret multiple common ECG abnormality types. The data analysed in this study consisted of 6877 12-lead ECG recordings from CPSC 2018 (labelled as normal rhythm or eight types of ECG abnormalities, split into training (approx. 80%), validation (approx. 10%) and test (approx. 10%) sets) and 3998 12-lead ECG recordings from PhysioNet/CinC 2020 (labelled as normal rhythm or four types of ECG abnormalities, used as external text set). The ECG-lead subset selection module was introduced within the proposed model to efficiently constrain model complexity. It detected an optimal 4-lead ECG subset consisting of leads II, aVR, V1 and V4. The proposed model using the optimal 4-lead subset significantly outperformed the model using the complete 12-lead ECG on the validation set and on the external test dataset. The results demonstrated that our proposed model successfully identified an optimal subset of 12-lead ECG; the resulting 4-lead ECG subset improves the generalizability of the DL model in ECG abnormality interpretation. This study provides an outlook on what channels are necessary to keep and which ones may be ignored when considering an automated detection system for cardiac ECG abnormalities. This article is part of the theme issue 'Advanced computation in cardiovascular physiology: new challenges and opportunities'.

Detection and classification of arrhythmia using an explainable deep learning model

BACKGROUND

Early detection and intervention is the cornerstone for appropriate treatment of arrhythmia and prevention of complications and mortality. Although diverse deep learning models have been developed to detect arrhythmia, they have been criticized due to their unexplainable nature. In this study, we developed an explainable deep learning model (XDM) to classify arrhythmia, and validated its performance using diverse external validation data.

METHODS

In this retrospective study, the Sejong dataset comprising 86,802 electrocardiograms (ECGs) was used to develop and internally variate the XDM. The XDM based on a neural network-backed ensemble tree was developed with six feature modules that are able to explain the reasons for its decisions. The model was externally validated using data from 36,961 ECGs from four non-restricted datasets.

RESULTS

During internal and external validation of the XDM, the average area under the receiver operating characteristic curves (AUCs) using a 12‑lead ECG for arrhythmia classification were 0.976 and 0.966, respectively. The XDM outperformed a previous simple multi-classification deep learning model that used the same method. During internal and external validation, the AUCs of explainability were 0.925-0.991.

CONCLUSION

Our XDM successfully classified arrhythmia using diverse formats of ECGs and could effectively describe the reason for the decisions. Therefore, an explainable deep learning methodology could improve accuracy compared to conventional deep learning methods, and that the transparency of XDM can be enhanced for its application in clinical practice.

Detection of myocardial ischemia by intracoronary ECG using convolutional neural networks

INTRODUCTION

The electrocardiogram (ECG) is a valuable tool for the diagnosis of myocardial ischemia as it presents distinctive ischemic patterns. Deep learning methods such as convolutional neural networks (CNN) are employed to extract data-derived features and to recognize natural patterns. Hence, CNN enable an unbiased view on well-known clinical phenomenon, e.g., myocardial ischemia. This study tested a novel, hypothesis-generating approach using pre-trained CNN to determine the optimal ischemic parameter as obtained from the highly susceptible intracoronary ECG (icECG).

METHOD

This was a retrospective observational study in 228 patients with chronic coronary syndrome. Each patient had participated in clinical trials with icECG recording and ST-segment shift measurement at the beginning (i.e., non-ischemic) and the end (i.e., ischemic) of a one-minute proximal coronary artery balloon occlusion establishing the reference. Using these data (893 icECGs in total), two pre-trained, open-access CNN (GoogLeNet/ResNet101) were trained to recognize ischemia. The best performing CNN during training were compared with the icECG ST-segment shift for diagnostic accuracy in the detection of artificially induced myocardial ischemia.

RESULTS

Using coronary patency or occlusion as reference for absent or present myocardial ischemia, receiver-operating-characteristics (ROC)-analysis of manually obtained icECG ST-segment shift (mV) showed an area under the ROC-curve (AUC) of 0.903±0.043 (p<0.0001, sensitivity 80%, specificity 92% at a cut-off of 0.279mV). The best performing CNN showed an AUC of 0.924 (sensitivity 93%, specificity 92%). DeLong-Test of the ROC-curves showed no significant difference between the AUCs. The underlying morphology responsible for the network prediction differed between the trained networks but was focused on the ST-segment and the T-wave for myocardial ischemia detection.

CONCLUSIONS

When tested in an experimental setting with artificially induced coronary artery occlusion, quantitative icECG ST-segment shift and CNN using pathophysiologic prediction criteria detect myocardial ischemia with similarly high accuracy.

Explaining deep neural networks for knowledge discovery in electrocardiogram analysis

Deep learning-based tools may annotate and interpret medical data more quickly, consistently, and accurately than medical doctors. However, as medical doctors are ultimately responsible for clinical decision-making, any deep learning-based prediction should be accompanied by an explanation that a human can understand. We present an approach called electrocardiogram gradient class activation map (ECGradCAM), which is used to generate attention maps and explain the reasoning behind deep learning-based decision-making in ECG analysis. Attention maps may be used in the clinic to aid diagnosis, discover new medical knowledge, and identify novel features and characteristics of medical tests. In this paper, we showcase how ECGradCAM attention maps can unmask how a novel deep learning model measures both amplitudes and intervals in 12-lead electrocardiograms, and we show an example of how attention maps may be used to develop novel ECG features.

ECGAug: A novel method of generating augmented annotated electrocardiogram QRST complexes and rhythm strips

Applications of neural networks (NNs) in medicine have increased dramatically in recent years. In order to train a NN that performs ECG segmentation, it can be very time consuming, or even completely prohibitive, to manually annotate fiducial points on enough QRST complexes to reach a high level of performance. Existing methods for time series data augmentation risk creating non-physiological ECG signals that may hamper NN training, and are unable to provide accurate fiducial point locations in the augmented data. We therefore developed ECGAug, a new method which generates an augmented training set of QRST signals (single beats or rhythm strips) with accurate fiducial point annotations. Our algorithm recombines a library of existing, annotated QRS complexes and T waves in physiologic ways, and then performs additional physiological transformations to generate a set of new annotated QRST complexes or rhythm strips to be used for NN training or validation of ECG annotation algorithms. In experiments where we trained NNs to annotate QRST complexes with a limited training dataset, QRST complexes added to the training dataset by ECGAug significantly improved NN performance. We present the ECGAug process, demonstrate its efficacy, and provide links for downloading the open source ECGAug software.

Uncertainty estimation for deep learning-based automated analysis of 12-lead electrocardiograms

Aims

Automated interpretation of electrocardiograms (ECGs) using deep neural networks (DNNs) has gained much attention recently. While the initial results have been encouraging, limited attention has been paid to whether such results can be trusted, which is paramount for their clinical implementation. This study aims to systematically investigate uncertainty estimation techniques for automated classification of ECGs using DNNs and to gain insight into its utility through a clinical simulation.

Methods and results

On a total of 526 656 ECGs from three different datasets, six different methods for estimation of aleatoric and epistemic uncertainty were systematically investigated. The methods were evaluated based on ranking, calibration, and robustness against out-of-distribution data. Furthermore, a clinical simulation was performed where increasing uncertainty thresholds were applied to achieve a clinically acceptable performance. Finally, the correspondence between the uncertainty of ECGs and the lack of interpretational agreement between cardiologists was estimated. Results demonstrated the largest benefit when modelling both epistemic and aleatoric uncertainty. Notably, the combination of variational inference with Bayesian decomposition and ensemble with auxiliary output outperformed the other methods. The clinical simulation showed that the accuracy of the algorithm increased as uncertain predictions were referred to the physician. Moreover, high uncertainty in DNN-based ECG classification strongly corresponded with a lower diagnostic agreement in cardiologist's interpretation (P < 0.001).

Conclusion

Uncertainty estimation is warranted in automated DNN-based ECG classification and its accurate estimation enables intermediate quality control in the clinical implementation of deep learning. This is an important step towards the clinical applicability of automated ECG diagnosis using DNNs.

Study on Potential of Meridian Acupoints of Traditional Chinese Medicine

In recent years, Traditional Chinese Medicine (TCM) has attracted more and more attention due to its good therapeutic effect, low cost, and convenience. This research is also a part of the goal of the modernization of TCM. Based on the meridian electric potential acquisition system independently developed by our project team, in this paper, we designed the human body's meridian electric potential acquisition scheme. We use principal component analysis (PCA) to prove that the meridional potential signal is derived from the ECG signal. Then, Inception ResNet V2 was used to classify acupoints and nonacupoints. Finally, the classification accuracy rate reached 86.59045265, and the F1 score = 0.72161642. This shows that acupoints and nonacupoints can be distinguished by their surface potential.

Diagnosis and Risk Prediction of Dilated Cardiomyopathy in the Era of Big Data and Genomics

Dilated cardiomyopathy (DCM) is a leading cause of heart failure and life-threatening ventricular arrhythmias (LTVA). Work-up and risk stratification of DCM is clinically challenging, as there is great heterogeneity in phenotype and genotype. Throughout the last decade, improved genetic testing of patients has identified genotype-phenotype associations and enhanced evaluation of at-risk relatives leading to better patient prognosis. The field is now ripe to explore opportunities to improve personalised risk assessments. Multivariable risk models presented as "risk calculators" can incorporate a multitude of clinical variables and predict outcome (such as heart failure hospitalisations or LTVA). In addition, genetic risk scores derived from genome/exome-wide association studies can estimate an individual's lifetime genetic risk of developing DCM. The use of clinically granular investigations, such as late gadolinium enhancement on cardiac magnetic resonance imaging, is warranted in order to increase predictive performance. To this end, constructing big data infrastructures improves accessibility of data by using electronic health records, existing research databases, and disease registries. By applying methods such as machine and deep learning, we can model complex interactions, identify new phenotype clusters, and perform prognostic modelling. This review aims to provide an overview of the evolution of DCM definitions as well as its clinical work-up and considerations in the era of genomics. In addition, we present exciting examples in the field of big data infrastructures, personalised prognostic assessment, and artificial intelligence.

Deep learning and the electrocardiogram: review of the current state-of-the-art

In the recent decade, deep learning, a subset of artificial intelligence and machine learning, has been used to identify patterns in big healthcare datasets for disease phenotyping, event predictions, and complex decision making. Public datasets for electrocardiograms (ECGs) have existed since the 1980s and have been used for very specific tasks in cardiology, such as arrhythmia, ischemia, and cardiomyopathy detection. Recently, private institutions have begun curating large ECG databases that are orders of magnitude larger than the public databases for ingestion by deep learning models. These efforts have demonstrated not only improved performance and generalizability in these aforementioned tasks but also application to novel clinical scenarios. This review focuses on orienting the clinician towards fundamental tenets of deep learning, state-of-the-art prior to its use for ECG analysis, and current applications of deep learning on ECGs, as well as their limitations and future areas of improvement.

Discovering and Visualizing Disease-Specific Electrocardiogram Features Using Deep Learning: Proof-of-Concept in Phospholamban Gene Mutation Carriers

BACKGROUND

ECG interpretation requires expertise and is mostly based on physician recognition of specific patterns, which may be challenging in rare cardiac diseases. Deep neural networks (DNNs) can discover complex features in ECGs and may facilitate the detection of novel features which possibly play a pathophysiological role in relatively unknown diseases. Using a cohort of PLN (phospholamban) p.Arg14del mutation carriers, we aimed to investigate whether a novel DNN-based approach can identify established ECG features, but moreover, we aimed to expand our knowledge on novel ECG features in these patients.

METHODS

A DNN was developed on 12-lead median beat ECGs of 69 patients and 1380 matched controls and independently evaluated on 17 patients and 340 controls. Differentiating features were visualized using Guided Gradient Class Activation Mapping++. Novel ECG features were tested for their diagnostic value by adding them to a logistic regression model including established ECG features.

RESULTS

The DNN showed excellent discriminatory performance with a c-statistic of 0.95 (95% CI, 0.91-0.99) and sensitivity and specificity of 0.82 and 0.93, respectively. Visualizations revealed established ECG features (low QRS voltages and T-wave inversions), specified these features (eg, R- and T-wave attenuation in V2/V3) and identified novel PLN-specific ECG features (eg, increased PR-duration). The logistic regression baseline model improved significantly when augmented with the identified features (P<0.001).

CONCLUSIONS

A DNN-based feature detection approach was able to discover and visualize disease-specific ECG features in PLN mutation carriers and revealed yet unidentified features. This novel approach may help advance diagnostic capabilities in daily practice.

Big Data and Artificial Intelligence: Opportunities and Threats in Electrophysiology

The combination of big data and artificial intelligence (AI) is having an increasing impact on the field of electrophysiology. Algorithms are created to improve the automated diagnosis of clinical ECGs or ambulatory rhythm devices. Furthermore, the use of AI during invasive electrophysiological studies or combining several diagnostic modalities into AI algorithms to aid diagnostics are being investigated. However, the clinical performance and applicability of created algorithms are yet unknown. In this narrative review, opportunities and threats of AI in the field of electrophysiology are described, mainly focusing on ECGs. Current opportunities are discussed with their potential clinical benefits as well as the challenges. Challenges in data acquisition, model performance, (external) validity, clinical implementation, algorithm interpretation as well as the ethical aspects of AI research are discussed. This article aims to guide clinicians in the evaluation of new AI applications for electrophysiology before their clinical implementation.

Automatic Triage of 12-Lead ECGs Using Deep Convolutional Neural Networks

BACKGROUND

The correct interpretation of the ECG is pivotal for the accurate diagnosis of many cardiac abnormalities, and conventional computerized interpretation has not been able to reach physician‐level accuracy in detecting (acute) cardiac abnormalities. This study aims to develop and validate a deep neural network for comprehensive automated ECG triage in daily practice.

METHODS AND RESULTS

We developed a 37‐layer convolutional residual deep neural network on a data set of free‐text physician‐annotated 12‐lead ECGs. The deep neural network was trained on a data set with 336.835 recordings from 142.040 patients and validated on an independent validation data set (n=984), annotated by a panel of 5 cardiologists electrophysiologists. The 12‐lead ECGs were acquired in all noncardiology departments of the University Medical Center Utrecht. The algorithm learned to classify these ECGs into the following 4 triage categories: normal, abnormal not acute, subacute, and acute. Discriminative performance is presented with overall and category‐specific concordance statistics, polytomous discrimination indexes, sensitivities, specificities, and positive and negative predictive values. The patients in the validation data set had a mean age of 60.4 years and 54.3% were men. The deep neural network showed excellent overall discrimination with an overall concordance statistic of 0.93 (95% CI, 0.92–0.95) and a polytomous discriminatory index of 0.83 (95% CI, 0.79–0.87).

CONCLUSIONS

This study demonstrates that an end‐to‐end deep neural network can be accurately trained on unstructured free‐text physician annotations and used to consistently triage 12‐lead ECGs. When further fine‐tuned with other clinical outcomes and externally validated in clinical practice, the demonstrated deep learning–based ECG interpretation can potentially improve time to treatment and decrease healthcare burden.