Abstract
Multi-label artificial intelligence (AI) systems in radiology and cardiology can flag many findings per patient, each of which may require clinician review. As alerts accumulate, they create patient-level review workload. To manage this workload, thresholds are commonly retuned to reduce alert volume. However, tuning on alert volume alone can drift away from a pre-specified reference setting and may increase missed high-priority findings. To address this, we propose Baseline-Anchored Rank-Weighted Guarded Projection (BARW-GP), a post-hoc protocol that updates thresholds without retraining. BARW-GP starts from the pre-specified reference setting and ranks candidate thresholds on a heldout threshold split. It then adopts the highest-ranked candidate that passes screening on a disjoint risk-calibration split for capped falsepositive workload and missed high-priority findings; otherwise, the reference is retained. We retrospectively evaluated BARW-GP across chest Xray (CXR) and electrocardiogram (ECG) settings, including NIH ChestXray14 external transfer and a PTB-XL ECG stress test. On NIH external transfer, it reduced missed high-priority findings by 568–1,563 cases while limiting Macro-F1 loss to 0.0360–0.0681, versus 0.1274–0.2536 for workload-only thresholding. On PTB-XL, it reduced missed high-priority findings in four of five ECG models and retained the reference for the fifth, avoiding the severe overall performance degradation of workloadonly thresholding. BARW-GP provides an auditable, evidence-anchored layer for workload-aware multi-label clinical AI alerting.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/AMAI_034.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=~Jiawen_Li10
BibTex
@InProceedings{CheYi_BARWGP_MICCAISAT2026,
author = { Chen, Yi AND Liu, Xingping AND To, Minh-Son AND Psaltis, Peter J. AND Li, Jiawen},
title = { { BARW-GP: Workload-Aware Threshold Selection for Multi-Label Medical AI Alerting Across Chest X-ray and ECG } },
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}
}