Evaluation of an ultra-portable X-ray system with automated interpretation for tuberculosis active case finding in carceral settings: a diagnostic test accuracy study

TL;DR

This study tested a portable X-ray system with automated analysis for detecting tuberculosis in prisons, showing high accuracy, especially among symptomatic individuals.

Contribution

The study evaluates a novel ultra-portable X-ray system with AI for TB screening in carceral settings, demonstrating its diagnostic accuracy.

Findings

The system achieved an AUC of 0.89 for TB detection, with higher accuracy in symptomatic individuals (AUC 0.93).

TB prevalence was 4.1% among 3399 screened individuals in a Brazilian prison.

The system enabled rapid screening with high diagnostic accuracy in a high-risk population.

Abstract

The World Health Organization recommends systematic active case finding for tuberculosis (TB) among high-risk populations, including incarcerated individuals; however, many prisons lack screening capacity. In this study, we aimed to evaluate the diagnostic performance of an ultra-portable digital chest radiography system paired with LunitTB, an automated interpretation algorithm, to detect TB disease.

We conducted a diagnostic test accuracy study using data collected during a prospective active TB case finding effort TB in a Brazilian prison from February 2023 through May 2024. Eligible individuals included adults (≥18 years) without a history of TB in the past two years. A Fujifilm Digital Radiography (FDR) Xair paired with LunitTB algorithm (version v3.1.5.1) system was used to screen consented individuals for TB disease, irrespective of their TB symptoms. Area under curves (AUC) and 95% confidence intervals (CI) were estimated to determine the accuracy of FDR Xair and LunitTB interpretation when compared to a rigorous microbiologic reference standard.

We screened a total of 3409 individuals for TB disease as part of our active TB case finding study, and 3399 (99.7%) met our eligibility criteria for the diagnostic test accuracy study. TB prevalence was 4.1% (139/3399, 95%CI: 3.5–4.8%). The AUC for FDR Xair and LunitTB interpretation was 0.89 (95%CI: 0.86–0.93). The accuracy of FDR Xair and LunitTB interpretation among those with any TB symptoms was significantly higher (AUC = 0.93, 95%CI: 0.90–0.97) compared to those without TB symptoms (AUC = 0.87, 95%CI: 0.81–0.92) (DeLong p = 0.033).

The FDR Xair and LunitTB interpretation enabled us to screen persons deprived of liberty rapidly, with a high diagnostic accuracy particularly among those who reported any TB symptoms.

The online version contains supplementary material available at 10.1186/s12879-025-11835-0.

Figures