LeadWeaver: Topology-Aware Masked-Lead Modeling for Robust 12-Lead ECG Diagnosis

LeadWeaver, a topology-aware masked-lead ECG representation model, improves robust 12-lead diagnosis on PTB-XL and CPSC-style settings under missing, corrupted, and device-shifted leads.

• Why it matters: Most ECG classifiers assume all twelve leads are present and comparable across devices, leaving a gap for real-world cases with absent, corrupted, or shifted lead signals.
• What they did: The method tokenizes each lead as a temporal stream, links leads via a fixed-plus-learned physiological lead graph, and uses masked-lead latent completion with a reliability router and a consistency objective.
• The result: Across complete-input diagnosis, arbitrary missing-lead subsets, lead corruption, cross-dataset transfer, calibration, and selective prediction, it supports the hypothesis that modeling inter-lead topology and uncertainty about missing evidence is key for robustness.

AI summary of the abstract · may contain errors · not medical advice · read the paper.