Ovarian cancer remains one of the most lethal gynecological malignancies, with borderline ovarian tumors (BOTs) representing lesions of low malignant potential requiring distinct surgical management. Accurate preoperative differentiation between BOTs and epithelial ovarian cancer (OC) is essential for optimizing treatment planning and patient counseling. We developed machine learning models to distinguish BOTs from OC using routine clinical data from 404 patients (199 BOTs, 205 °C) comprising 49 demographic, clinical, and biochemical parameters. Data were obtained from a public repository (n = 349) and a single institutional cohort (n = 55). We evaluated five algorithms: Random Forest, Support Vector Machine, Neural Network, Logistic Regression, and Decision Tree, using stratified train–test splits and stratified 5-fold cross-validation repeated five times to ensure robust performance estimation. The Random Forest model achieved the highest performance with an area under the receiver operating characteristic curve (AUC-ROC) of 0.95 (95% CI: 0.92–0.98), accuracy of 0.91 (95% CI: 0.87–0.94), sensitivity of 0.88 (95% CI: 0.83–0.92), and specificity of 0.94 (95% CI: 0.90–0.97) at the optimal threshold determined by Youden’s index. Feature importance analysis identified HE4 (weight = 0.224), CA125 (weight = 0.089), and neutrophil count (weight = 0.072) as the most discriminative predictors. Performance was comparable across both data sources, with no significant domain shift detected. Machine learning analysis of readily available laboratory parameters demonstrates potential for preoperative differentiation of BOTs from OC. A web-based prototype tool has been developed to facilitate future validation studies.
SamadiAfshar et al. (2026) studied this question.