Tumor Stroma Ratio-Based Radiomics Model for Predicting Platinum Resistance and Prognosis in Epithelial Ovarian Cancer

Background The tumor-stroma ratio (TSR) has emerged as a promising prognostic biomarker in epithelial ovarian cancer (EOC); however, its preoperative assessment remains challenging. Objective To develop a non-invasive CT-based radiomics machine learning model for preoperative TSR prediction and to evaluate its association with platinum resistance and survival outcomes in EOC. Methods This retrospective study included 172 patients with pathologically confirmed EOC. TSR was histologically classified as stroma-rich (≥ 50%) or stroma-poor (< 50%). A total of 718 radiological features—including 4 conventional imaging features and 714 quantitative radiomic descriptors—were extracted from contrast-enhanced CT images, along with 26 clinical variables. Feature selection was performed using least absolute shrinkage and selection operator (LASSO) regression, and a linear support vector machine (SVM) classifier was constructed. The dataset was randomly divided into a training cohort (70%) and an independent validation cohort (30%). Model performance was evaluated using five-fold cross-validation in the training cohort and tested on the independent validation cohort. Associations between the predicted-TSR and clinical outcomes were analyzed using multivariable logistic and Cox regression models to assess the clinical value of the model. Results Stroma-rich tumors were significantly associated with advanced FIGO stage, poor differentiation, ascites, lymph node metastasis, and platinum resistance (all p  < 0.05). The SVM model achieved a mean cross-validated AUC of 0.83 ± 0.08 in the training cohort and an AUC of 0.83 in the independent validation cohort. The predicted TSR score was independently associated with platinum resistance (odds ratio [OR] = 1.29, 95% confidence interval [CI]: 1.03–1.61, p  = 0.026), shorter progression-free survival (hazard ratio [HR] = 1.55, 95% CI: 1.08–2.22, p  = 0.020), and overall survival (HR = 1.50, 95% CI: 1.02–2.21, p  = 0.041). Conclusions The proposed CT-based radiomics model enables reliable, non-invasive estimation of TSR and provides a biologically interpretable imaging biomarker for risk stratification in EOC. Radiomics-derived TSR may help identify patients at increased risk of platinum resistance and poor prognosis, supporting individualized treatment planning. Prospective multicenter validation is warranted to confirm its clinical applicability.