Diagnostic of multiple cardiac disorders from 12-lead ECGs using Graph Convolutional Network based multi-label classification

Automated detection and classification of clinical elec-trocardiogram (ECG) play a critical role in the analysisof cardiac disorders. Deep learning is effective for auto-mated feature extraction and has shown promising resultsin ECG classification. Most of these methods, however,assume that multiple cardiac disorders are mutually exclu-sive. In this work, we have created and trained a noveldeep learning architecture for addressing the multi-labelclassification of 12-lead ECGs. It contains an ECG rep-resentation work for extracting features from raw ECGrecordings and a Graph Convolutional Network (GCN) formodelling and capturing label dependencies. In the Phys-ioNet/Computing in Cardiology Challenge 2020, our team,Leicester-Fox, reached a score of0.627±0.054using 5-fold cross-validation on the full training data.