Prognostic significance of intercellular bridges quantified by machine learning in oral squamous cell carcinoma: a retrospective study.

OBJECTIVE: To quantify the intercellular bridges in oral squamous cell carcinoma (OSCC) using artificial intelligence (AI) and to evaluate their independent prognostic value for recurrence-free survival (RFS). STUDY DESIGN: Whole-slide images from 103 primary OSCC patients were analyzed. Swin-WNet automatically segmented intercellular bridges, cytoplasm, and nuclei at the tumor surface (ROI-S), center, and invasive front to calculate area ratios. ROC, Kaplan-Meier, and multivariate Cox regression analyses identified prognostic factors. RESULTS: Cytoplasm-based ratios exhibited slightly higher predictive accuracies (AUC: 0.77-0.79) than nucleus-based ratios. A low cytoplasm-based ROI-S ratio predicted significantly higher recurrence rates (P < .001). Multivariate analysis identified the cytoplasm-based ROI-S ratio (HR: 2.98, P = .019), depth of invasion (DOI; HR: 2.93, P = .011), and Yamamoto-Kohama classification (HR: 2.54, P = .014) as independent RFS predictors. CONCLUSIONS: AI-driven quantification of the cytoplasm-based intercellular bridge ratio at the tumor surface has the potential to serve as a novel, objective, and independent predictor of OSCC recurrence. Integrating this cytological metric with macroscopic parameters (DOI, invasion patterns) may contribute to a comprehensive, multispatial prognostic system. Further prospective studies are required to validate these findings.