Abstract
BACKGROUND: The objective of this study was to establish a new model for predicting the prognosis of endometrial carcinoma (EC) using tumor-infiltrating lymphocytes (TILs) based on artificial intelligence (AI). METHODS: Patients with EC who were treated between 1989 and 2022 were included in this study. For each patient, one hematoxylin and eosin-stained slide containing the most invasive frontline of the tumor was selected and digitized. The area within a 500 μm width span, extending 250 μm toward the stroma and tumor from the manually annotated invasive frontline, was automatically annotated. The average number of lymphocytes per area (μm(2)) in the annotated area was calculated using AI. Patients were classified into the High-TIL and Low-TIL groups, and survival analysis was conducted. Four mismatch repair (MMR)-related proteins were evaluated using immunohistochemical staining. RESULTS: A total of 659 patients were included: 346 (52.5%) in the High-TIL group and 313 (47.5%) in the Low-TIL group. MMR deficiency was observed more frequently in the High-TIL group than in the Low-TIL group (p < 0.01). Progression-free survival (PFS) and overall survival (OS) were better in the High-TIL group than in the Low-TIL group (both p < 0.01). Multivariate analysis revealed that TIL status was a prognostic factor for PFS (hazard ratio [HR] (95% confidence interval [CI]) 0.61 (0.43-0.87); p < 0.01) and OS (HR (95% CI) 0.54 (0.33-0.86); p = 0.01). CONCLUSION: TILs evaluated using AI could accurately and significantly predict the prognosis of EC. Further studies are needed to establish new methods for evaluating TILs in ECs.