Research on arrhythmia classification based on domain adaptation

Electro cardiogram (ECG) is a common and powerful tool to study cardiac function and diagnose several kinds of abnormal arrhythmias. It takes time and is incredi-bly hard for cardiologists to accurately detect long-term ECG records. Convolutional neural network can automatically detect arrhythmia by learning deep representations. However, due to the particularity of medical data, its labeling cost is high, and many data are incompletely labeled, the model cannot obtain enough training data; due to individual variances, varying data sources, and varying data collection methodologies, ECG data inevitably have data distribution differences. At the same time, with the development of long-term monitoring wearable devices, efficient models are needed to meet the needs in real-time detection scenarios. In light of the above issues, this paper proposes an end-to-end ECG signal classification method based on domain adaptation to achieve automatic arrhythmia detection and identification and diagnosis of arrhythmia, and reduces the distribution difference between the two domains through semantic space and feature space adaptation. We evaluated the proposed method on the MIT-BIH Arrhythmia Database and verified its generalization on the other two public ECG datasets.