Leveraging AI Technology in Interpretation of X-Ray Images and Its Integration Into the Public Health System

Abstract: Artificial Intelligence (AI) is increasingly used in chest X-ray (CXR) interpretation to improve the speed and scalability of screening in resource-constrained settings like India. This study systematically reviewed literature published after 2014, following PRISMA guidelines, to evaluate the diagnostic performance, cost, and efficiency of AI-assisted CXR interpretation compared to radiologists. Radiologists demonstrated an average sensitivity of 71.0% (95% CI: 68.0–76.2%) and specificity of 86.2% (95% CI: 84.0–87.8%), while AI showed higher sensitivity (86.8%; 95% CI: 81.5–90.4%) and comparable specificity (87.0%; 95% CI: 81.3–90.1%). Across prevalence levels relevant to India (0.5–5%), AI consistently yielded higher positive and negative predictive values, although PPV remained low at lower prevalence levels due to screening population characteristics. Sensitivity analyses indicated that AI reduced reporting time and costs under most scenarios, with variability depending on assumptions. AI demonstrates strong potential as a triage tool in large-scale screening programs such as tuberculosis under NTEP. However, its effectiveness depends on prevalence and implementation context. AI should be integrated as a complementary tool within existing workflows rather than a standalone replacement. Key-words: AI diagnostics, Computer-aided detection, X-ray interpretation, Digital health