Diagnostic of Multiple Cardiac Disorders from 12-lead ECGs Using Graph Convolutional Network Based Multi-label Classification
Zheheng Jiang, Tiago P. Almeida, Fernando S. Schlindwein, G. André Ng, Huiyu Zhou, Xin Li
Abstract
Automated detection and classification of clinical electrocardiogram (ECG) play a critical role in the analysis of cardiac disorders. Deep learning is effective for automated feature extraction and has shown promising results in ECG classification. Most of these methods, however, assume that multiple cardiac disorders are mutually exclusive. In this work, we have created and trained a novel deep learning architecture for addressing the multi-label classification of 12-lead ECGs. It contains an ECG representation work for extracting features from raw ECG recordings and a Graph Convolutional Network (GCN) for modelling and capturing label dependencies. In the Phy-sioNet/Computing in Cardiology Challenge 2020 [1], our team, Leicester-Fox, reached a challenge validation score of 0.395, and full test score of -0.012, placing us 34 out of 41 in the official ranking.