The enhanced qXR demonstrated strong performance, especially in younger children, and offers a scalable tool for early TB detection in settings with limited pediatric radiology expertise. Further studies incorporating clinically diagnosed TB (unconfirmed TB) and distinguishing severe from non-severe cases are needed to strengthen screening programs.

The qXR QC module is a critical safeguard for AI-based CXR interpretation in LMICs. By reliably identifying poor-quality images before CAD analysis, it reduces diagnostic errors, prevents resource waste, and provides real-time feedback for reacquisition. By reducing repeat scans and minimizing human QC reliance, this module enhances workflow efficiency and enables decentralisation of TB screening operations. This quality assurance layer enhances the integrity of automated TB screening and supports responsible AI deployment in resource-constrained settings.

The qXR silicosis model demonstrated excellent diagnostic accuracy, with near-perfect sensitivity and specificity against radiologist reports, and minimal overlap with radiological signs of TB. Integrating such tools into occupational health programs could enable timely diagnosis, support worker compensation processes, and advance combined silicosis-TB screening strategies in high-risk populations.