Association of AI-based quantitative B7-H3 evaluation in the tumor microenvironment with molecular subtypes in high-grade serous ovarian carcinoma.

5599 Background: High-grade serous ovarian carcinoma (HGSC) exhibits heterogeneous tumor microenvironments (TME) that correlate with clinical outcomes. We previously reported four molecular histopathological subtypes of HGSC based on gene expression profiles reflecting TME characteristics: Immune Reactive (IR), Mesenchymal (MT), Solid/Proliferative (SP), and Papillary/Glandular (PG). The IR subtype, characterized by tumor-infiltrating lymphocytes, demonstrates favorable prognosis (PMID:26993207, 30853361). We also identified lower B7-H3 expression in IR compared to non-IR subtypes (PMID:34799346). Here, we developed an AI-based B7-H3 IHC quantification method across TME compartments and investigated its association with molecular subtypes. Methods: Forty-five HGSC patients who underwent primary debulking surgery were enrolled. Molecular histopathological subtyping categorized cases into IR (n=16), MT (n=20), and Other (SP/PG, n=9). B7-H3 immunohistochemistry-stained whole slide images were analyzed using DeepPathFinder, an AI-based pathological image analysis tool that performs automated tissue segmentation and IHC marker quantification. B7-H3 expression was quantified across three tumor-associated compartments: tumor, peritumor (within 1000 μm of the tumor border), and necrosis. An integrated B7-H3 score was calculated as the mean density across these compartments. Survival analyses were performed using the Kaplan-Meier method and Cox proportional hazards regression. Results: The integrated B7-H3 score showed a stepwise increase across subtypes: IR (0.148) < MT (0.219) < Other (0.252) (Kruskal-Wallis p=0.091), and was significantly lower in IR compared to non-IR subtypes (0.148 vs 0.229, p=0.034). B7-H3 expression was consistently lower in IR across all compartments: tumor (IR: 0.171, MT: 0.205, Other: 0.257), peritumor (IR: 0.100, MT: 0.154, Other: 0.164), and necrosis (IR: 0.171, MT: 0.298, Other: 0.336; p=0.025). Survival analysis revealed significant prognostic differences among subtypes. Compared to IR, MT showed significantly worse outcomes (PFS: HR=8.35, 95% CI 3.00–23.24, p<0.0001; OS: HR=15.35, 95% CI 3.50–67.31, p=0.0003). The Other subtype also demonstrated inferior OS (HR=6.38, 95% CI 1.17–34.95, p=0.033). Both MT and Other (SP/PG) subtypes, which exhibited higher B7-H3 expression, were associated with worse prognosis. Conclusions: AI-based quantitative B7-H3 evaluation revealed subtype-specific expression patterns in the HGSC tumor microenvironment. B7-H3 expression was lowest in IR and higher in non-IR subtypes across all tumor-associated compartments. These findings suggest that patients with higher B7-H3-expressing subtypes may benefit from B7-H3-targeted therapy.