S886 Artificial Intelligence-Based Radiomic Prediction of Neoadjuvant Chemoradiotherapy Treatment Response in Esophageal Squamous Cell Carcinoma: A Systematic Review and Meta-Analysis of Diagnostic Test Accuracy

Introduction: Accurate prediction of pathological response to neoadjuvant chemoradiotherapy (nCRT) in esophageal squamous cell carcinoma (ESCC) remains a major clinical challenge. While current predictors are inadequate, artificial intelligence (AI)-driven radiomics offers a transformative approach. This meta-analysis evaluates the performance of AI-based radiomics for predicting nCRT response, which directly addresses the need for personalized therapeutic strategies. Methods: We conducted a PRISMA-compliant systematic review, searching PubMed, Embase, Scopus, and Cochrane through March 2025. Original studies developing and testing AI-radiomic models (machine/deep learning) to predict pathological response in ESCC post-nCRT were included. Methodological quality was assessed via QUADAS-2. Bivariate random-effects models pooled sensitivity, specificity, and area under the curve (AUC) (95% confidence interval [CI]), with heterogeneity quantified using I² statistics. PROSPERO ID: (CRD420251024295). Results: Ten studies and 1,519 patients were included in our analysis. Radiomics achieved the following diagnostic test measures: (sensitivity: 0.804; 95% CI 0.75-0.85; I² = 66.5%; P < 0.001), (specificity: 0.816; 95% CI 0.78-0.85; I² = 51.0%; P < 0.001), and (AUC: 0.85; 95% CI 0.81-0.88; I² = 86.7%; P < 0.0001). Heterogeneity reflected variations in radiomic feature extraction but was balanced by consistent endpoint assessment (Mandard Tumor Regression Grade 1-2). Eighty percent of the included studies showed low bias risk. Conclusion: AI-based radiomics demonstrates high discriminatory power (AUC 0.85) for stratifying ESCC patients by nCRT response likelihood. This supports its clinical utility in guiding treatment intensification in predicted poor responders, enabling personalized de-escalation (e.g., organ preservation) in predicted responders, and potentially reducing unnecessary surgery. Prospective multicenter validation and standardized radiomic pipelines are imperative for clinical adoption.