Radiomics-based MRI model for differentiating ovarian cystadenoma and cystadenocarcinoma.

Background: Accurate differentiation between benign and malignant ovarian tumors is crucial to ensure timely intervention for high-risk patients and minimizing overtreatment in others. Thus, the objective was to develop a radiomics-based machine learning model to differentiate between non-cancerous and cancerous lesions in ovaries. Methods: This retrospective study included 271 patients with ovarian cystadenocarcinoma and 266 patients with ovarian cystadenoma who underwent T2-weighted magnetic resonance imaging (MRI). Lesions were manually segmented, and 2286 radiomics features were extracted. Following dimensionality reduction, features were used to train machine learning models with different combinations of data scaling (min-max, Z-score, and quantile transformations) and classifiers (logistic regression and partial least squares discriminant analysis). In addition, a model combining radiomics features and cancer biomarker data (CA-125 and HE-4) was developed. Models were compared using the DeLong's test and McNemar's tests. Results: A total of 37 features were used for model development. The quantile logistic regression model exhibited the highest performance, achieving an AUC of 0.92, a sensitivity of 0.87, and a specificity of 0.83, which was significantly higher compared to all models except for Z-score logistic regression model. The inclusion of serum biomarkers (CA-125 and HE-4) significantly improved the performance of the model. Conclusion: T2-weighted MRI radiomics-based models accurately differentiated between benign and malignant epithelial ovarian tumors.