Evaluation of CAD products in community and facility TB screening: 2025 policy update

Background Since the WHO issued its initial recommendation on CAD use in 2021, the number of commercially available products has expanded substantially, necessitating an updated independent evaluation to inform global policy. Methods We conducted a standardized, head-to-head evaluation of eight commercially available CAD products for adult TB screening using two large, well-characterized, geographically diverse CXR libraries representing community-based and facility-based screening use cases. CAD products were selected through a WHO public call for expressions of interest and evaluated independently on a secure validation platform with offline installation. Diagnostic accuracy was assessed using receiver operating characteristic (ROC) analysis, partial area under the curve (pAUC) in clinically relevant specificity ranges (40–60%), and comparisons with expert radiologist performance and WHO TPP thresholds. Results Diagnostic performance varied substantially across products. In community screening, AUCs ranged from 0.72 to 0.84, and in facility screening from 0.77 to 0.86. Six CAD products—qXR, Insight CXR, CAD4TB, InferRead DR Chest, Genki Edge, and DrAid demonstrated optimal diagnostic accuracy as they had overlapping confidence intervals to the expert radiologists'estimates and met WHO performance considerations for screening use, while two products (Radify Chest and JVIEWER-X) underperformed. At a fixed sensitivity of 90%, specificities ranged from 26–56% in community screening and 30–49% in facility screening. Conclusions These findings demonstrate that CAD performance varies markedly across products, underscoring the need for standardized, product-specific evaluation and careful threshold selection. This work constituted evidence informing the WHO policy update on CAD use for TB screening.