Abstract
The essential criteria of efficient CXR triaging system is reliable spatial cues and proper uncertainty calibration. We performed retrospective selective-referral pipeline testing on the 30,194 4-class COVIDQU-Ex CXR image database. The selective-referral pipeline utilized DenseNet-121, Grad-CAM++, and MC Dropout. We examined the entire pipeline using only one test set, unlike other works using different approaches, based on classification, localization of spatial cues, estimation of uncertainty, selective-referral strategy and global results. The system showed 97.81% of classification accuracy on the 4-classification task using 4,530 test images, with the macro F1 of 97.62% and macro-AUC of 0.9979. Concerning localizations for the 502 COVID-19 CXRs, Grad-CAM++ achieved the Pointing Game score of 82.05% with IoU equal to 0.63.MC Dropout produced an ECE value of 0.018, and an Error-AUROC of 0.941. The proposed process was able to refer 90 out of 99 wrongly classified cases (90.9%), based on referral criteria generated from validation. The system managed to retain 96.09% of the overall data with a classification accuracy rate of 99.79%. The above tests performed have been done retrospectively. They haven’t been clinically tested or human-based trials have not been done yet.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/AMAI_049.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=%7EMoinul_Hossain1
BibTex
@InProceedings{HosMoi_Explainable_MICCAISAT2026,
author = { Hossain, Moinul AND Morol, Md Kishor},
title = { { Explainable and Uncertainty-Aware Chest X-ray Classification with Selective Referral } },
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}
}
