Comprehensive Summary
This study investigated whether cancer-testis antigen (CTA) gene expression signatures combined with machine learning (ML) models could predict cervical lymph node metastasis in oral tongue squamous cell carcinoma (OTSCC). A total of 121 patients were analyzed, with gene expression profiled using NanoString technology. Multiple ML approaches, including decision trees, t-distributed stochastic neighbor embedding (t-SNE), and convolutional neural networks (CNNs), were applied to identify predictive CTA gene sets. The resulting CTA signature demonstrated robust discrimination between node-positive and node-negative patients, with CNN models achieving the highest classification accuracy. The predictive performance of CTA-based models outperformed traditional clinical-pathological risk factors.
Outcomes and Implications
Nodal spread is the single most important prognostic factor in OTSCC, yet up to 70% of patients with clinically node-negative disease undergo unnecessary elective neck dissection. By providing a molecular and ML-driven tool for nodal risk prediction, this study suggests a pathway toward more personalized surgical decision-making. At the bedside, integrating CTA gene signatures with ML models could help spare low-risk patients from overtreatment while ensuring high-risk patients receive timely intervention. Broader validation in larger, multi-center cohorts is required, but this approach highlights how AI-enhanced molecular profiling may refine risk stratification and improve outcomes in head and neck oncology.