Abstract
When building a Computer Aided Detection algorithm, there is often a gap between validation and real-world metrics due to its calibration. In this work, we investigate operating-point-aware losses for candidate-level false-positive filtering in mammography. The proposed objective combines binary cross-entropy (BCE), high-sensitivity partial-AUC ranking of weak positives against hard negatives, and a sensitivity-anchored false-positive penalty near the estimated operating boundary. Across three seeds, the selected configuration achieved the highest mean INbreast specificity at the validation-selected 95% sensitivity threshold while preserving global discrimination and markedly improving score quality relative to pure high-sensitivity partial-AUC training. These results support optimizing the samples that determine the intended high-sensitivity operating point.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/Deep_Brea3th_041.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: https://papers.miccai.org/miccai-2026-sat/supp/Deep_Brea3th_041_supp.pdf
Link to Open Review
Open Review Page: https://openreview.net/forum?id=zLRt0IPnZO
BibTex
@InProceedings{HusBur_Towards_MICCAISAT2026,
author = { Hussein, Burhan Rashid AND Tardy, Mickael},
title = { { Towards Reliable Classification: Building Bridge From Validation to Clinics With Improved Partial-AUC Optimization for Mammography False-Positive Detections Filtering } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17256},
month = {pending},
page = {pending}
}