Development and validation of pathomics signature for predicting prognosis of advanced high-grade serous ovarian carcinoma patients after platinum-based chemotherapy.

In patients with advanced high-grade serous ovarian carcinoma (HGSOC), the standard treatment typically involves platinum-based chemotherapy after debulking surgery. However, the prognosis of patients following platinum-based chemotherapy varies, and known prognostic factors do not fully explain this variability. Therefore, developing an accurate and validated prognostic tool is essential. This study retrospectively analyzed hematoxylin and eosin (H&E) staining whole-slide images (WSIs) from 392 advanced HGSOC patients. A pathomics signature (PS) was developed and validated using pathomics features derived from the best model in an integrated machine learning framework. Discrimination performance was quantified using time-dependent 5-year receiver operating characteristic (ROC) curves. Calibration curves were also generated to compare predicted survival probabilities with actual outcomes. Genomic analysis was performed to elucidate the molecular mechanisms underlying the model's predictions. PS demonstrated significant risk stratification for overall survival and successfully identified patients likely to benefit from platinum-based chemotherapy. The predictive power of PS was validated in an external validation cohort. Furthermore, genomic analysis suggested that PS is influenced by the cytoskeleton, protein digestion and absorption pathways, and is correlated with immune cell infiltration, which may help predict the prognosis of advanced HGSOC patients after platinum-based chemotherapy. PS based on pathomics features is a potential prognostic marker for advanced HGSOC patients after platinum-based chemotherapy. This marker demonstrates robust prognostic ability and broad applicability. Further prospective studies are needed to assess the practical application of this model in precision therapy.