Abstract
Cardiotoxicity from anthracyclines and HER2-targeted therapies frequently causes myocardial injury before left ventricular ejection fraction (LVEF) declines. Task 2 of the EchoRisk Challenge addresses binary classification of subclinical dysfunction from paired apical four-chamber and two-chamber cine loops, where approximately 90% of positive exams have preserved LVEF. We propose a motion-aware multi-stream network coupling an $R(2+1)D$-LSTM video stream with an explicit end-diastole/end-systole ($\mathrm{ED}/\mathrm{ES}$) deformation stream. The network uses grayscale, frame-difference motion, and LV cavity mask channels and is trained with a single dysfunction classification head. Scanner models strongly confound the dataset, with scanner identity alone yielding an AUC of 0.762. We therefore introduce a within-scanner evaluation protocol ($\mathrm{AUC}_{\mathrm{ws}}$) that restricts concordance comparisons to examinations acquired on the same scanner model. On the official validation split, the single-head model achieved an AUC of 0.806 with a 95% bootstrap confidence interval of $[0.698, 0.898]$ and an $\mathrm{AUC}_{\mathrm{ws}}$ of 0.807, compared with an $\mathrm{AUC}_{\mathrm{ws}}$ of 0.700 for the appearance ensemble.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/EchoRisk2026_013.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=l91rrugId9
BibTex
@InProceedings{AgaKhu_MotionAware_MICCAISAT2026,
author = { Agarwal, Khush AND Noh, Jiyoo AND Chan, Jonathan H.},
title = { { Motion-Aware Multi-Stream Learning for Subclinical LV Dysfunction Detection from Echocardiography Videos } },
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}
}