Multi-label Classification of Arrhythmia Using Dynamic Graph Convolutional Network Based on Encoder-Decoder Framework

Yuhao Cheng,Wenliang Zhu,Deyin Li,Lirong Wang
DOI: https://doi.org/10.1016/j.bspc.2024.106348
IF: 5.1
2024-01-01
Biomedical Signal Processing and Control
Abstract:Background: Cardiac arrhythmia is a condition characterized by abnormal heart electrical activity, posing a significant health risk to patients. Despite significant progress in existing arrhythmia classification methods, the accuracy of network-based classification still requires improvement. Methods: In the paper, we propose an effective classification algorithm called Encoder-Decoder Dynamic Graph Convolutional Network (ED-DGCN). In the encoder stage, a convolutional neural network(CNN) is utilized to extract feature information from electrocardiogram signals. The decoder stage employs a bidirectional long short-term memory network to define semantic regions and generate corresponding class relationships. A dynamic graph convolutional network module is constructed to capture label dependencies by obtaining a dynamic adjacency matrix. Additionally, transfer learning is employed by pre-training the proposed network on a large dataset to further enhance convergence performance and classification accuracy. Results: The proposed method is verified on the China Physiological Signal Challenge 2018 (CPSC2018) dataset. Experimental results demonstrate that the average F1-score of our method achieved 0.843, which is 1.6 % improved over existing methods. Conclusion: The proposed method effectively enhances the classification performance of the network and provides an effective solution for multi-label classification tasks in cardiac arrhythmias. The method holds promising prospects for clinical applications.
What problem does this paper attempt to address?