Abstract
Mitral regurgitation (MR) is a common valvular heart disease requiring accurate echocardiographic assessment, yet grading from color Doppler echocardiography remains challenging due to patient variability, imaging artifacts, and the need to interpret complex spatio-temporal flow patterns. In this work, we systematically evaluate recent spatio-temporal foundation models for MR severity grading using apical four-chamber echocardiographic videos. To this end, we compare contrastive, reconstructive, and latent-predictive pre-training paradigms across echocardiography-specific and natural video pre-training domains, including EchoCLIP, EchoFM, EchoJEPA, and V-JEPA variants, alongside a fully supervised 3D CNN baseline. We consider both a four-class grading task and a binary clinically significant MR triage task. Models are evaluated using frozen encoders with linear and attentive probing to assess global and localized representation quality. Across a retrospectively curated dataset of 413 videos, attentive probing consistently improves performance over linear evaluation, highlighting the importance of localized spatio-temporal aggregation. Latent-predictive JEPA-based models achieve the strongest performance among foundation models, approaching or surpassing the fully supervised baseline, while natural video pre-training remains competitive with domain-specific approaches.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/ASMUS_040.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: https://papers.miccai.org/miccai-2026-sat/supp/ASMUS_040_supp.pdf
Link to Open Review
Open Review Page: https://openreview.net/forum?id=3IWTGp5QZ0
BibTex
@InProceedings{FalJon_Assessing_MICCAISAT2026,
author = { Fallmann, Jonas AND Schobesberger, Martin AND Tretiak, Volodymyr AND Damian, Ilinca AND Roschger, Cornelia AND Kellermair, Jörg AND Steinwender, Clemens AND Zierer, Andreas AND Ferscha, Alois AND Kobler, Erich},
title = { { Assessing Spatio-temporal Foundation Models for Clinical Mitral Regurgitation Grading } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17276},
month = {pending},
page = {pending}
}