Diagnostic accuracy of artificial intelligence in detecting pulmonary nodules on computed tomography: A systematic review and meta-analysis

Abstract

Objectives: Pulmonary nodules are frequent findings on chest computed tomography (CT) and are crucial for early lung cancer detection. Artificial intelligence (AI), particularly deep learning (DL), has emerged as a powerful tool for automated nodule detection, but diagnostic performance varies across studies. The objective of this study is to evaluate the diagnostic accuracy of AI-based imaging systems in detecting pulmonary nodules on CT and compare DL with traditional computer-aided detection (CAD) and radiologists. Material and Methods: A systematic review and meta-analysis were conducted in accordance with PRISMADTA guidelines. PubMed, Embase, Scopus, and Web of Science were searched up to March 2025. Eligible studies assessed AI for pulmonary nodule detection on CT and reported sensitivity, specificity, or sufficient data for 2×2 contingency tables. Pooled estimates of sensitivity, specificity, and diagnostic odds ratio (DOR) were calculated using a bivariate random-effects model. Subgroup analyses examined DL versus CAD, detection of nodules <10 mm, and AI performance compared with radiologists. Study quality was evaluated using QUADAS-2. Results: Twenty-six studies comprising 31,742 CT scans were included. Pooled sensitivity and specificity of AI were 0.90 (95% CI 0.87–0.93) and 0.87 (95% CI 0.84–0.90), respectively, with a DOR of 60.2 and an area under the sROC curve of 0.94. DL outperformed CAD (sensitivity 0.92 vs 0.83; specificity 0.88 vs 0.84). Sensitivity for nodules <10 mm was lower (DL 0.82; CAD 0.68). AI matched radiologists, and AI-assisted radiologists showed improved sensitivity (0.93). Conclusion: AI demonstrates high diagnostic accuracy in pulmonary nodule detection, outperforming CAD and matching radiologists, with the greatest potential as a second reader in lung cancer screening.