Motion-Conditioned Multi-View Fusion for Myocardial Infarction Localization from Echocardiography

TL;DR - MCF-Net is a motion-guided, multi-view fusion framework that combines cardiac motion cues with echocardiography foundation-model features to localize myocardial infarction at the segment level, matters because it reduces annotation burden while improving reliability over single-view methods.

• Fuses EchoPrime (pretrained Echo foundation model) visual features across dual views with motion-derived priors for MI localization.
• Uses extremely sparse supervision: a single annotated template frame is transferred across videos to initialize point tracking, avoiding dense labels.
• Motion-derived segment-aware soft masks act as coarse spatial priors, selectively enhancing features for hard-to-read myocardial segments (notably apical views).
• Achieves 72.4% F1 and 84.9% accuracy on segment-level MI localization, reportedly beating motion-only, vision-only, and fusion baselines.