Abstract
Late gadolinium-enhanced cardiac magnetic resonance (LGE-CMR) is the reference standard for myocardial scar but is not always clinically feasible. We investigate whether scar presence can be predicted per AHA myocardial segment from contrast-free, multi-view echocardiography, to support triage of LGE-CMR prescription. A multi-view network fuses the available apical views and jointly predicts all 17 segments, evaluated with patient-grouped cross-validation. Segment-level prediction is feasible at moderate discrimination (Macro AUROC ≈ 0.68), while learned view attention does not improve over parameter-free mean pooling. The dominant limitation is a modality gap: a teacher with access to free-text CMR reports reaches ≈ 0.79 AUROC and is strongest precisely where echo is weakest, yet neither output- nor representation-level distillation transfers this advantage, as possibly the missing signal resides in the modality. Finally, while the model ranks segments reliably, its raw probabilities remain poorly calibrated for threshold-based decisions, a limitation resolved by a training-free per-segment recalibration that recovers them without retraining or compromising discrimination.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/MLMI_064.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
Open Review Page: https://openreview.net/forum?id=0N31qwN3Nf
BibTex
@InProceedings{RigFra_Beyond_MICCAISAT2026,
author = { Righetti, Francesca AND Gibello, Riccardo AND Ren, Yijun AND Kawada, Yuka AND Negru, Andra AND Badano, Luigi AND Konukoglu, Ender AND Caiani, Enrico},
title = { { Beyond Discrimination: Recalibrating Echocardiographic Scar Prediction } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17272},
month = {pending},
page = {pending}
}