Abstract
Tuberculosis (TB) remains a major public health problem in resource-constrained settings, where delayed chest X-ray interpretation may slow triage and referral for confirmatory testing. This study evaluated whether external TB-specific pretraining could improve sensitivity and reduce missed TB cases when applied to Nigerian chest X-rays. A Nigerian four-class chest X-ray dataset containing COVID-19, normal, pneumonia, and TB images was reframed as a binary triage task: TB-positive versus non-TB. DenseNet-121 was trained using two pathways: Nigerian-only training and external TB/normal pretraining followed by Nigerian fine-tuning. Both models were evaluated on the same Nigerian test set of 600 images. External pretraining reduced false-negative TB predictions from 4 to 1, lowering the false-negative rate from 2.67% to 0.67%, while maintaining high specificity (0.99). Threshold adjustment further achieved complete sensitivity at a decision threshold of 0.18 with specificity above 95%. These findings suggest that TB-specific pretraining combined with local fine-tuning can improve the safety of AI-assisted TB triage in Nigerian and similar resource-constrained settings.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/MIRASOL_029.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/forum?id=jPLJxbnaYr
BibTex
@InProceedings{JonWis_SensitivityOptimized_MICCAISAT2026,
author = { Jonathan, Wisdom AND Iorumbur, Aondona Moses AND Raymond, Confidence},
title = { { Sensitivity-Optimized Tuberculosis Triage Using Nigerian Chest X-Rays } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17265},
month = {pending},
page = {pending}
}