Learning Cardiac Latent Representations in Vectorcardiogram Space
This work addresses the redundancy and overfitting issues in cardiac representation learning for medical professionals by moving from ECG to the physically grounded VCG space, offering enhanced robustness and generalization.
This paper proposes LVCG, a self-supervised learning framework that learns cardiac representations in the vectorcardiogram (VCG) space instead of the standard electrocardiogram (ECG) signal space. By operating in the VCG space, LVCG minimizes redundancy and improves generalization, outperforming ECG-space baselines across various tasks, particularly in domain shift scenarios.
Electrocardiography (ECG) is a cornerstone of cardiac assessment, making the learning of informative ECG representations fundamental to tasks ranging from disease diagnosis to clinical report generation. However, existing methods operate almost exclusively in the observable ECG signal space. In practice, the standard twelve-lead ECG represents multiple projections of the same underlying cardiac electrical activity from different spatial orientations. Therefore, representation learning in the ECG space inevitably introduces substantial redundancy, which may lead to spurious correlations and increased risk of overfitting. To address this and motivated by the Frank vectorcardiogram (VCG) model, we propose learning a unified latent representation of cardiac electrical activity directly in the VCG space. We introduce LVCG, the first general self-supervised representation learning framework designed to operate in this physically grounded latent space. By learning view-invariant latent VCG representations rather than lead-specific artifacts, VCG minimizes redundancy and improves generalization. LVCG generally outperforms ECG-space baselines across tasks, demonstrating enhanced robustness and generalization, especially in domain shift settings.