Classification of 12-lead ECGs: the PhysioNet/Computing in Cardiology Challenge 2020

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'.

Transfer learning for ECG classification

Remote monitoring devices, which can be worn or implanted, have enabled a more effective healthcare for patients with periodic heart arrhythmia due to their ability to constantly monitor heart activity. However, these devices record considerable amounts of electrocardiogram (ECG) data that needs to be interpreted by physicians. Therefore, there is a growing need to develop reliable methods for automatic ECG interpretation to assist the physicians. Here, we use deep convolutional neural networks (CNN) to classify raw ECG recordings. However, training CNNs for ECG classification often requires a large number of annotated samples, which are expensive to acquire. In this work, we tackle this problem by using transfer learning. First, we pretrain CNNs on the largest public data set of continuous raw ECG signals. Next, we finetune the networks on a small data set for classification of Atrial Fibrillation, which is the most common heart arrhythmia. We show that pretraining improves the performance of CNNs on the target task by up to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$6.57\%$$\end{document}6.57%, effectively reducing the number of annotations required to achieve the same performance as CNNs that are not pretrained. We investigate both supervised as well as unsupervised pretraining approaches, which we believe will increase in relevance, since they do not rely on the expensive ECG annotations. The code is available on GitHub at https://github.com/kweimann/ecg-transfer-learning.

Deep Learning Algorithm Classifies Heartbeat Events Based on Electrocardiogram Signals

Cardiovascular diseases (CVDs) have become the number 1 threat to human health. Their numerous complications mean that many countries remain unable to prevent the rapid growth of such diseases, although significant health resources have been invested toward their prevention and management. Electrocardiogram (ECG) is the most important non-invasive physiological signal for CVD screening and diagnosis. For exploring the heartbeat event classification model using single- or multiple-lead ECG signals, we proposed a novel deep learning algorithm and conducted a systemic comparison based on the different methods and databases. This new approach aims to improve accuracy and reduce training time by combining the convolutional neural network (CNN) with the bidirectional long short-term memory (BiLSTM). To our knowledge, this approach has not been investigated to date. In this study, Database I with single-lead ECG and Database II with 12-lead ECG were used to explore a practical and viable heartbeat event classification model. An evolutionary neural system approach (Method I) and a deep learning approach (Method II) that combines CNN with BiLSTM network were compared and evaluated in processing heartbeat event classification. Overall, Method I achieved slightly better performance than Method II. However, Method I took, on average, 28.3 h to train the model, whereas Method II needed only 1 h. Method II achieved an accuracy of 80, 82.6, and 85% compared with the China Physiological Signal Challenge 2018, PhysioNet Challenge 2017, and Massachusetts Institute of Technology-Beth Israel Hospital (MIT-BIH) Arrhythmia datasets, respectively. These results are impressive compared with the performance of state-of-the-art algorithms used for the same purpose.