Radiogenomic Analysis of Contrast-Enhanced MRI for Predicting Pathogenic BRCA Variants in Ovarian Cancer

This study evaluated the feasibility of predicting pathogenic BRCA1/2 variants in ovarian cancer using radiogenomic analysis of contrast-enhanced MRI. Twenty six patients were analyzed, including 14 with pathogenic BRCA1/2 variants and 12 without. Contrast enhanced MRI was used to generate two dimensional (2D) and three dimensional (3D) tumor segmentations. Radiomic features extracted with PyRadiomics were reduced using least absolute shrinkage and selection operator (LASSO) regression, yielding two selected features. Linear discriminant analysis (LDA) models were trained on 18 patients and tested on 8, with performance assessed using sensitivity, specificity, accuracy, and the area under the receiver operating characteristic (ROC) curve. The DeLong test compared AUCs between segmentation methods. Automated machine learning (AutoML) was performed with a training set of 20 patients and a test set of 6, with multiple performance metrics calculated. LDA achieved approximately 60% accuracy in both 2D and 3D test datasets. AUCs were 0.80 (3D) and 0.77 (2D), with no significant difference (p=0.85). AutoML identified CatBoost_BAG_L2_FULL as the optimal model, achieving accuracy above 80%. MRI-based radiogenomic analysis demonstrated moderate classification performance for identifying pathogenic BRCA1/2 variants in ovarian cancer. These preliminary findings suggest a potential role for noninvasive prediction, warranting further validation in larger cohorts.