Graph Neural Networks Enhance CTRCD Detection from 12-Lead ECG by Modeling Inter-Lead Relationships: A Preliminary Study

Cancer therapy–related cardiac dysfunction (CTRCD) is a serious complication, posing significant risks to cancer survivors’ cardiovascular health. While early detection is crucial, current imaging-based methods are expensive and impractical for routine screening. This study proposes a deep learning framework that combines convolutional neural networks (CNNs) and graph convolutional networks (GCNs) to analyze 12-lead electrocardiography (ECG) data for CTRCD detection. By representing ECG leads as graph nodes and modeling their anatomical and physiological relationships through different adjacency matrices, our approach captures inter-lead dependencies overlooked by conventional methods. Experimental results demonstrate that the physiological function-based graph structure outperforms the conventional CNN approach, particularly in sensitivity and F1-score. Interpretability analysis reveals distinct lead-specific patterns, enhancing clinical understanding.