Abstract
Left ventricular (LV) dysfunction assessment from echocardiographic videos is challenging in cardio-oncology surveillance, particularly under severe class imbalance where video models may favor sensitivity at the expense of specificity. We propose Deform-MoE, a deformation-aware mixture-of-experts framework that integrates cross-view video representation with myocardial deformation evidence. The Cross-view Echocardiographic Expert (CEE) extracts multi-view video features using a pretrained PanEcho encoder and cross-view attention. The Myocardial Deformation Expert (MDE) introduces myocardial deformation-derived global longitudinal strain (MD-GLS), computed from ECG-aligned LV segmentation, mitral-annular plane exclusion, myocardial point tracking, and contour-length reconstruction. Rather than combining video and deformation features through unconstrained concatenation, Deform-MoE derives a preserved-deformation low-risk prior from myocardial deformation evidence and integrates it with the video expert through expert-level calibrated fusion, enabling deformation-aware risk inference that suppresses false-positive video predictions under severe class imbalance. On validation, Deform-MoE achieved an AUC of 0.884 (95\% CI: 0.805-0.951) and a balanced accuracy of 0.852. These results support Deform-MoE as an effective deformation-aware expert fusion strategy for improving the specificity and robustness of video-based LV dysfunction assessment. This work was carried out by the BIT-ARIMED team.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/EchoRisk2026_019.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=mjSCDkcrEp
BibTex
@InProceedings{JiaRui_DeformationAware_MICCAISAT2026,
author = { Jia, Ruizhe AND Zhou, Jinghan AND Chen, Jiajun AND Wang, Yuanyuan AND Fan, Jingfan AND Yang, Jian},
title = { { Deformation-Aware Expert Fusion for Echocardiographic Left Ventricular Dysfunction Assessment } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17268},
month = {pending},
page = {pending}
}