Artificial intelligence versus radiologist interpretation in predicting treatment response in lung cancer: A systematic review and meta-analysis.

Abstract

e23192 Background: Predicting treatment response in lung cancer is vital for optimizing outcomes. The emergence of artificial intelligence (AI) has shown significant promise in augmenting diagnostics. By leveraging imaging modalities, AI can improve sensitivity, specificity, and diagnostic accuracy, offering a hopeful future for patient care. While radiologists remain central to imaging analysis, AI's comparative performance in predicting treatment response remains underexplored. Methods: This meta-analysis evaluates AI systems versus radiologists in predicting treatment response in lung cancer. Focusing on sensitivity, specificity, accuracy, and diagnostic performance, a systematic review of retrospective studies was conducted. Data sources included PubMed, Embase, Cochrane Library, Google Scholar, Web of Science, and CINAHL/EBSCO. Inclusion criteria centered on diagnostic metrics. Data extraction used spreadsheets; analysis employed RevMan 5.4.1. Results: A total of 11 studies involving 6,615 patients were included. AI outperformed radiologists in sensitivity (Risk Ratio [RR] = 1.15, 95% CI: 1.07–1.24, P = 0.0002) and accuracy (Odds Ratio [OR] = 1.45, 95% CI: 1.33–1.58, P < 0.0001). Specificity was comparable (RR = 1.02, 95% CI: 0.84–1.24, P = 0.85). AI's pooled AUC was 0.91 versus radiologists' 0.75–0.85 range. Significant heterogeneity in sensitivity (I² = 96%), specificity (I² = 98%), and accuracy (I² = 96%) arose from variations in AI models and imaging modalities. Subgroup analyses showed AI superiority in advanced-stage lung cancer and EGFR-targeted therapy cases. Conclusions: AI-based systems outperform radiologists in sensitivity, accuracy, and diagnostic consistency for predicting lung cancer treatment response. Specificity was similar, but AI reduced interobserver variability and accelerated assessments. These findings support integrating AI as a valuable adjunct to radiologists in complex cases. Further prospective studies with standardized protocols are needed to validate AI's clinical utility in oncology.