Abstract
BACKGROUND: India accounts for 27% of global TB incidence and bears the highest TB burden worldwide. This study evaluates the performance of an AI-assisted computer-aided detection (AI-CAD) solution in a community-based, active case-finding TB screening programme conducted in Tamil Nadu, India. It also provides a comparative analysis of AI-assisted screening and conventional screening methods. METHODS: Community-based TB screening was carried out using mobile diagnostic units equipped with digital X-ray machines. The performance of the AI-CAD solution was evaluated by calculating area under the receiver operating characteristic (AUROC) curve, sensitivity, and specificity. Additionally, data from five districts that used conventional screening methods were analysed for comparative analysis against AI-assisted screening. RESULTS: AI-CAD exceeded the World Health Organization (WHO)-recommended minimum target product profile (TPP) with a sensitivity of 0.93 (95% confidence interval [CI]: 0.88, 0.97) and a specificity of 0.83 (95% CI: 0.82, 0.83). AI interpretation was significantly associated with positive TB diagnosis (odds ratio: 58.95, P < 0.0001). AI-assisted screening led to a 2.09-fold increase in TB diagnoses (P < 0.05) and a 2.86-fold higher sputum positivity rate (P < 0.05) compared with the conventional screening approach. CONCLUSION: The AI-CAD met and exceeded the WHO's minimal TPP for TB detection. The higher sputum-positive yield reinforces AI-CAD's potential in large-scale TB screening programmes.