This retrospective validation study evaluated a deep learning (DL)-based AI device (qXR) for detecting tuberculosis (TB) signs across 10,000 chest X-rays (CXRs) collected from various sites in India between April 2020 and March 2021. The AI system achieved an AUC of 0.928, with sensitivity of 0.841 and specificity of 0.806 on digital DICOM images. When tested on smartphone-captured analog X-rays, the AUC differences were minimal (2.55% for Samsung S8, 5.10% for iPhone 8, and 1.91% for iPhone XS), demonstrating the system’s robust performance in both digital and analog formats. These findings highlight the clinical utility of AI for TB detection in resource-limited settings, where printed films may be the only available diagnostic tool.
