Assessment of established prognostic factors and artificial intelligence-based evaluation of tumor-infiltrating lymphocytes in oral tongue squamous cell carcinoma.

BACKGROUND: Traditional risk assessment for tongue cancer relies on clinicopathological parameters. Although tumor-infiltrating lymphocytes (TILs) are promising prognostic markers, their evaluation lacks standardization. This study aimed to validate established prognostic factors and introduce an artificial intelligence (AI)-based TIL assessment method.

METHODS: We analyzed 139 tongue cancer cases from a single institution (2010-2017) to establish prognostic factors and developed an AI model for TIL quantification. Clinicopathological characteristics including AI- and manually assessed TILs were evaluated.

RESULTS: The AI-assessed stromal TIL ratio exerted protective effects across all stages and demonstrated superior discriminative capability compared to manual evaluation (C-index: 0.649 vs. 0.604 for overall survival [OS]), with substantial inter-method agreement (Intraclass Correlation Coefficient = 0.796). In the multivariate analysis, a statistical model incorporating the lymph node ratio, AI-assessed stromal TIL ratio, depth of invasion grade, perineural invasion, lymphovascular invasion, and a close surgical resection margin (<5 mm) showed superior prognostic performance, with excellent discriminative power (OS area under the curve [AUC]: 0.851; recurrence-free survival [AUC]: 0.826). Stage-specific analysis revealed that advanced-stage patients were significantly affected by adverse factors and stromal TIL levels, whereas early stage patients showed trends but no statistically significant associations.

CONCLUSIONS: AI-based stromal TIL assessment outperformed manual TIL assessment as a prognostic marker. This AI approach robustly predicts survival when combined with factors such as the lymph node ratio and a close resection margin status (<5 mm). Our findings may enhance risk stratification, particularly in advanced-stage disease.