Domain Knowledge Enhances ECG Recognition with Graph Networks

Researchers have developed a new approach for Electrocardiogram (ECG) recognition that integrates domain-specific medical knowledge into a graph convolution network. This method moves beyond traditional end-to-end convolutional neural networks by explicitly incorporating key ECG landmark points, such as PRQST, which are vital for expert interpretation. The core of the system is a double-stream directed graph that models relationships both within a single ECG cycle (spatial dependencies among key points) and between adjacent cycles in extended ECG sequences (temporal dependencies). This dual modeling allows for a more nuanced and informed analysis of the complex ECG signals. Evaluated on the First Chinese ECG Intelligent Competition dataset, which classifies ECGs into nine categories, the proposed model achieved an overall average F1 score of 88.1%. Crucially, it significantly improved the detection performance for rare categories, reaching an average F1 score of 76.3%, thereby outperforming existing state-of-the-art models and demonstrating the efficacy of incorporating domain knowledge.