Abstract 2449: Using digital pathology to differentiate survival phenotypes in high-grade serous ovarian cancer

Abstract Background: High-grade serous ovarian cancer (HGSOC) has a median survival < 5 years. Yet approximately 10% of patients become long-term survivors (LTS), living > 10 years. Predicting extremes of survival may reveal insights into HGSOC biology. Machine learning (ML) can predict survival outcome in other malignancies from hematoxylin and eosin (H&E) whole slide images (WSIs) though no published models address overall survival (OS) in HGSOC. This project uses ML to predict OS in HGSOC H&E WSIs, comparing tumor and stromal regions. Methods: Short-term survivors (STS) were defined as OS < 2 years and LTS as OS > 7 years. The model was trained on 138 H&E WSIs (56 LTS, 82 STS) at 20x magnification with 5-fold cross-validation repeated 3x and validated on The Cancer Genome Atlas-Ovary (TCGA) dataset (34 WSIs; 6 LTS, 28 STS). After converting WSIs to tiles, features were extracted using the UNI foundation model. A linear discriminant analysis predicted tile association with survival, a pretrained tumor segmentation model separated tumor and stroma, and a logistic regression classifier predicted survival based on the ratio of STS to LTS. Performance was compared across regions: whole slide, tumor alone, stroma alone, a 500-µm band at the tumor-stroma interface, and a combination of tumor-based and stroma-based predictions. Conclusions: Models using both stroma and tumor outperformed compartmentalized models. This trend was consistent in the TCGA dataset though performance dropped significantly, proving the feasibility of using ML on H&E WSIs to predict survival in HGSOC. Our results identify a need to optimize generalizability for this relatively data-scarce malignancy and suggest that successful prediction of OS should acknowledge tumor-stromal interactions. Citation Format: Kelsey C. Goon, Leonhard Donle, Ernst Lengyel. Using digital pathology to differentiate survival phenotypes in high-grade serous ovarian cancer [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 2449.