A Dual Arrhythmia Classification Algorithm Based on Deep Learning and Attention Mechanism Incorporating Morphological-temporal Information

In the context of increasing medical resource constraints and the global pandemic of COVID-19, the acquisition and automatic diagnosis of electrocardiogram (ECG) signal at home is becoming more and more important. In this paper, we propose a dual arrhythmia classification algorithm for edge-cloud collaboration. We first design a lightweight single-lead ECG signal binary classification model incorporating RR intervals that can be deployed at the edge, which achieves lightweight ECG feature extraction by using depthwise separable convolution and positional attention, and fuses RR interval features to the fully connected layer to achieve normal or abnormal classification of ECG heartbeats. For heartbeats classified as abnormal using the above model, we design a dual-branch arrhythmia multi-classification model with channel and spatial dual attention that integrates simple convolutional neural network (CNN) modules that can be deployed in a cloud artificial intelligence (AI) server to perform accurate classification of abnormal ECG heartbeats, where the input of one branch is a heartbeat signal and the input of the other branch is an ECG segment containing adjacent R-peaks. The experimental results based on the MIT-BIH arrhythmia database demonstrate that our binary classification model achieves an average accuracy of 99.80% and the multi-classification model achieves an average accuracy of 99.71%, and our method ensures a high enough accuracy while performing dual analysis to make the analysis results more reliable.