SUN-306 Uncovering Tumor Microenvironment Differences In High-grade Serous Ovarian Cancer By Machine Learning

Abstract Disclosure: S. Khan: None. Background: Ovarian carcinoma (OC) is a rare but highly lethal gynecological malignancy that is often associated with poor prognosis and drug resistance. Among its histotypes, high-grade serous ovarian carcinoma (HGSOC) accounts for approximately 75% of all epithelial OCs. It is characterized by late detection due to unreliable diagnostic markers, a low five-year survival rate (∼47.5%), and frequent recurrence driven by platinum resistance. This resistance is attributed to the heterogeneous cellular architecture of HGSOC, which contributes to varied patient outcomes following standard chemotherapy. HGSOC exhibits significant morphological diversity both within and between tumors, which is further complicated by racial disparities that underscore the need for precise and timely diagnosis. HGSOC is highly heterogeneous and complicates its treatment. Cellular Morphometric Biomarkers (CMBs) provide valuable insights into diagnosis, prognosis, and treatment response. Machine learning-based cellular morphometric biomarkers (CMB-ML) have been identified across multiple tumor types, capturing tissue heterogeneity, predicting tumor microenvironments (TME), and informing clinical outcomes. This study aimed to identify ethnicity-specific CMBs in African American and White HGSOC patients using whole-slide images (WSIs) and assess their association with overall survival (OS).Methods: We analyzed 109 patients from The Cancer Genome Atlas Ovarian Cancer (TCGA-OV) cohort and validated our findings using WSIs from 22 patients in the Loma Linda University (LLU) cohort. Immune checkpoint markers (ICMs) were correlated with CMB scores and confirmed by immunohistochemistry. Results: Three ethnicity-specific CMBs (73, 80, and 215) were identified, validated, and showed significant differences in frequency. Higher frequencies of CMB 73 and 80 correlated with shorter OS in African Americans (p=0.022 and p=0.023, respectively), while a higher frequency of CMB 215 was associated with improved OS in White patients (p=0.051). Molecular analysis of the TCGA-OV cohort revealed lower immune infiltration in African Americans and higher ICM expression in Whites (PDCD1, p=0.033; PDCD1LG2, p=0.014; CD8A, p=0.014). Immunohistochemistry in the LLU-OV cohort confirmed the expression of CD3, CD8, and PDCD1.Conclusion: While no significant differences were observed between ethnic groups, CMB-ML provides a novel approach to understanding health disparities in HGSOC. This technology enables a deeper understanding of how variations in TME composition, such as immune cell infiltration and stromal interactions, could influence tumor behavior and treatment responses, even without overt differences between ethnic groups. Presentation: Sunday, July 13, 2025