Machine Learning Quantification of Intraepithelial Tumor-Infiltrating Lymphocytes as a Significant Prognostic Factor in High-Grade Serous Ovarian Carcinomas

Abstract

The prognostic and predictive role of tumor-infiltrating lymphocytes (TILs) has been demonstrated in various neoplasms. The few publications that have addressed this topic in high-grade serous ovarian carcinoma (HGSOC) have approached TIL quantification from a semiquantitative standpoint. Clinical correlation studies, therefore, need to be conducted based on more accurate TIL quantification. We created a machine learning system based on H&E-stained sections using 76 molecularly and clinically well-characterized advanced HGSOC. This system enabled immune cell classification. These immune parameters were subsequently correlated with overall survival (OS) and progression-free survival (PFI). An intense colonization of the tumor cords by TILs was associated with a better prognosis. Moreover, the multivariate analysis showed that the intraephitelial (ie) TILs concentration was an independent and favorable prognostic factor both for OS (p = 0.02) and PFI (p = 0.001). A synergistic effect between complete surgical cytoreduction and high levels of ieTILs was evidenced, both in terms of OS (p = 0.0005) and PFI (p = 0.0008). We consider that digital analysis with machine learning provided a more accurate TIL quantification in HGSOC. It has been demonstrated that ieTILs quantification in H&E-stained slides is an independent prognostic parameter. It is possible that intraepithelial TIL quantification could help identify candidate patients for immunotherapy.

Keywords

• ovarian cancer
• tumor-infiltrating lymphocytes
• digital quantification
• algorithms
• machine learning