Graph-CMMC Enhances 12-Lead ECG Analysis with Pseudo-Multimodal Learning

This paper introduces Graph-CMMC, a novel graph-based pseudo-multimodal contrastive learning framework designed to enhance the analysis of 12-lead electrocardiograms (ECGs). Traditional ECG analysis often struggles to capture the crucial inter-lead dependencies and global waveform patterns that clinicians use for diagnosis. Graph-CMMC addresses this by converting 12-lead ECG waveforms into Gramian Angular Difference Field (GADF) images, creating complementary representations of the same cardiac activity. The framework then aligns these waveform and GADF representations in a self-supervised manner. A key innovation is the integration of a graph-based relational module, which explicitly models the dependencies between the 12 leads and enforces structural consistency during the contrastive learning process. Experimental evaluations on a multi-label coronary artery occlusion classification task demonstrate that Graph-CMMC achieves competitive performance, even surpassing some supervised learning methods. This highlights the effectiveness of using GADF as a complementary representation and the benefits of explicit graph-based modeling for learning robust 12-lead ECG representations.