Abstract
Serial left ventricular ejection fraction (LVEF) monitoring is essential for managing cardiotoxicity in breast cancer patients, yet current automated methods often fail to detect subtle declines. This occurs because the narrow distribution of LVEF in the dataset often leads to appearance-based regressors to concentrate predictions around the dominant range. To improve sensitivity in the clinically important low EF range, we propose a multi-signal fusion framework. We first enhance the discriminative power of appearance backbone through transfer learning and targeted augmentation. We combine appearance features with complementary geometry (biplane Simpson) and motion signals to reduce correlated prediction errors. Additionally, we incorporate view-aware routing and a longitudinal patient anchor to handle single-view exams and extreme EF values. Validated on the EchoRisk-MICCAI 2026 dataset, our method yields an MAE of 4.03, an RMSE of 5.25 and a Pearson r of 0.69 which helps early dysfunction detection.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/EchoRisk2026_009.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=2j1rbazLNO
BibTex
@InProceedings{ShiMin_MultiPhysEF_MICCAISAT2026,
author = { Shi, Minghai AND Zhang, Xiaoxian AND Ren, Yihui AND Li, Lei},
title = { { MultiPhys-EF: Decorrelated Multi-Signal Fusion for LVEF Estimation in Cardiotoxicity Surveillance } },
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}
}