Abstract
AI-based disease severity estimation in ulcerative colitis (UC) could substantially improve the detection of treatment effects in clinical trials. Their real-world deployment, however, is hampered by the lack of uncertainty models that can defer difficult cases to experts without eroding the treatment-effect sensitivity gains provided by AI scoring. Motivated by the weak labels and distributional shifts common in endoscopies from large, multi-site trials, we identify Quantile Regression (QR) as a principled strategy for selective regression. QR is distribution-free, yields interpretable uncertainty intervals, and accommodates non-Gaussian errors. We compare four single-model uncertainty approaches and ensembling on four large multicenter UC clinical-trial datasets. To our knowledge, this is the first study evaluating selective regression for trial-level decision making in UC trials. We evaluate selective behavior using , RC curves, and trial-level treatment-effect size (Hedges’ ) for dose selection and go/no-go decisions. Despite being a single model, QR consistently matches or competes with ensembles while reducing relative to an uncertainty-agnostic baseline. Crucially, in a deployment-realistic dual-score deferral setting where uncertain AI scores are replaced by less treatment-sensitive expert scores, QR is the only method that increases Hedges’ over the baselines, showing that uncertainty-aware deferral preserves or improves AI treatment-effect sensitivity. These gains persist when up to 25% of cases are deferred, supporting QR as a reliable uncertainty-aware approach for disease-severity estimation in large-scale UC trials.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/UNSURE2026_053.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=~Krishna_Chaitanya1
BibTex
@InProceedings{SanSar_Quantile_MICCAISAT2026,
author = { Sangalli, Sara AND Mobadersany, Pooya AND Yamamoto, Shinobu AND Parmar, Chaitanya AND Skomrock, Nicholas AND Mansi, Tommaso AND Cula, Gabriela Oana AND Standish, Kristopher A. AND Damasceno, Pablo F. AND Chaitanya, Krishna},
title = { { Quantile regression enables reliable, uncertainty-aware endoscopic scoring in ulcerative colitis clinical trials } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17260},
month = {pending},
page = {pending}
}