Abstract
Most breast lesion classifiers prioritize accuracy but overlook whether their confidence can be trusted. For benign-versus-malignant(BvM) classification, an overconfident wrong diagnosis is more dangerous than honest abstention, as well-calibrated confidence allows benign
cases to be safely cleared without a clinician, easing workload in low-resource settings. We evaluate calibration safety in multimodal fusion using the CDD-CESM dataset of 326 patients with paired low-energy(LE) and dual-energy subtracted (DES) mammography and structured radiologist text. Clinical text closes a real discrimination gap, as image-only models reach AUC 0.732–0.752 while fusion raises this to 0.837–0.899 (p < 0.0001). Calibration effects are mixed, as ECE improves for LE under fusion from 0.203 to 0.127 but degrades for DES from 0.047 to 0.112. At a single fixed threshold, both fusion models appear less safe than image-only baselines. Sweeping all thresholds reverses this for LE, as LE+Text reaches zero missed malignancies at only 85% deferral, fewer than either image-only model (90–92%), while DES+Text remains the least safe configuration under both evaluations, requiring 94.9% deferral for the same guarantee. The implementation is available at: https://github.com/joyinola/BI-RADS-descriptor-aware-mammography-AI-for-safe-referral-triage-in-LMIC-screening
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/MIRASOL_030.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: https://papers.miccai.org/miccai-2026-sat/supp/MIRASOL_030_supp.pdf
Link to Open Review
Open Review Page: https://openreview.net/forum?id=UjgGSA2Suq
BibTex
@InProceedings{AdeSim_Discrimination_MICCAISAT2026,
author = { Adetoro, Simbiat Damilola AND Adeyemo, Peter Ifeoluwa AND Bazay, Fatima Ez-Zahraa AND Raymond, Confidence AND Iorumbur, Aondona Moses},
title = { { Discrimination Is Not Enough: Calibration-Aware Multimodal Classification of Benign and Malignant Lesions in Contrast-Enhanced Spectral Mammography } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17264},
month = {pending},
page = {pending}
}