Abstract
Cardiac magnetic resonance imaging (CMR) examinations have variable sequence availability and slice coverage. SegMIL is presented as a segmentation-guided multiple instance learning framework that represents each examination as a bag of slices. A U-Net branch predicts myocardial probability maps that supervise representation learning and guide decoder-feature pooling before patient-level aggregation. Evaluation comprised 934 internal patients with 12 primary diagnoses and two label-harmonized cine datasets. Five-checkpoint patient ensembles achieved macro AUCs of 0.709 (95% CI 0.671–0.746) internally, 0.702 (0.657–0.745) on ACDC, and 0.701 (0.619–0.776) on Sunnybrook. The strongest respective baselines achieved 0.703, 0.658, and 0.604, although paired confidence intervals included zero. On the common 119-case ACDC subset, replacing reference-mask with predicted-mask preprocessing reduced SegMIL AUC by 0.140, compared with reductions of 0.195–0.359 for mask-input baselines. Performance varied markedly by diagnosis, and macro F1 remained low. SegMIL therefore provides a research framework for anatomy-guided patient aggregation, but the results do not establish robustness to arbitrary missing data or clinical readiness.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/AMAI_005.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/profile?id=%7EKeno_Bressem1
BibTex
@InProceedings{BreKen_SegmentationGuided_MICCAISAT2026,
author = { Bressem, Keno AND Dürner, Celine AND Ben Chaaben, Zeineb AND Chami, Alessa AND Hendrich, Eva AND Adams, Lisa C. AND Hadamitzky, Martin},
title = { { Segmentation-Guided Multiple Instance Learning for Cardiac MRI Disease Classification } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17273},
month = {pending},
page = {pending}
}