Deep learning based detection of active pulmonary tuberculosis from chest radiographs in a retrospective prospective multicenter study
Abstract
Active tuberculosis remains a major global health problem, and distinguishing active from non-active disease on chest radiographs could accelerate triage where radiology resources are limited. We developed ShuffleNetV2 using 4,613 radiographs from two centers and evaluated it in retrospective external (1,318 patients) and prospective (560 patients) cohorts. The model achieved AUCs of 0.917 and 0.896, respectively, and performance was stable across prespecified demographic and scanner subgroups. In the prospective cohort, artificial-intelligence assistance increased radiologist AUCs to 0.897–0.937 across experience levels. These findings support ShuffleNetV2 as a rapid decision-support tool for prioritizing confirmatory clinical and microbiological assessment; positive predictions should not be used alone to initiate treatment.
Related
- Early sputum bacterial load dynamics predict treatment response in multidrug-resistant/rifampicin-resistant tuberculosis: A multicenter prospective cohort study
- Neuropsychiatric effects of cycloserine with bedaquiline, delamanid, and conventional therapy in drug-resistant tuberculosis: a multicenter retrospective cohort study from China
- Venturi-effect ultrasonography for early detection of AVF stenosis: a prospective multicenter study
- Risk stratification by circulating tumor DNA-based pretreatment genomic profile and post-operative molecular residual disease detection in resectable colorectal oligometastases: PRECISION, a prospective multicenter study
- Additional tuberculosis case detection through repeat household contact investigation at 6 and 12 months: a retrospective cohort study in Kasama, a peri-urban district in Zambia
BibTeX
@article{tuersun2026deep,
title = {{Deep learning based detection of active pulmonary tuberculosis from chest radiographs in a retrospective prospective multicenter study}},
author = {Abudouresuli Tuersun and Saimaitikari Abudoubari and Jingxu Xu and Juan Ma and Pahatijiang Nijiati and Xierenayi Warisi and Xingzhi Chen and Yiming Li and Xiaoguang Zou and Chencui Huang and Mayidili Nijiati},
journal = {iScience},
year = {2026},
month = oct,
doi = {10.1016/j.isci.2026.117960},
url = {https://www.cell.com/iscience/fulltext/S2589-0042(26)03343-2},
note = {CC BY-NC-ND 4.0},
}
RIS
TY - JOUR
TI - Deep learning based detection of active pulmonary tuberculosis from chest radiographs in a retrospective prospective multicenter study
AU - Abudouresuli Tuersun
AU - Saimaitikari Abudoubari
AU - Jingxu Xu
AU - Juan Ma
AU - Pahatijiang Nijiati
AU - Xierenayi Warisi
AU - Xingzhi Chen
AU - Yiming Li
AU - Xiaoguang Zou
AU - Chencui Huang
AU - Mayidili Nijiati
PY - 2026
DA - 2026/10/20
JO - iScience
DO - 10.1016/j.isci.2026.117960
UR - https://www.cell.com/iscience/fulltext/S2589-0042(26)03343-2
AB - Active tuberculosis remains a major global health problem, and distinguishing active from non-active disease on chest radiographs could accelerate triage where radiology resources are limited. We developed ShuffleNetV2 using 4,613 radiographs from two centers and evaluated it in retrospective external (1,318 patients) and prospective (560 patients) cohorts. The model achieved AUCs of 0.917 and 0.896, respectively, and performance was stable across prespecified demographic and scanner subgroups. In the prospective cohort, artificial-intelligence assistance increased radiologist AUCs to 0.897–0.937 across experience levels. These findings support ShuffleNetV2 as a rapid decision-support tool for prioritizing confirmatory clinical and microbiological assessment; positive predictions should not be used alone to initiate treatment.
SN - 2589-0042
ER -