Development an Artificial Intelligence Model to Identify BRCA Mutations in Prostate Cancer Through prostate MRI images

Motivation: Prostate cancer is the leading male malignancy in South Korea, with BRCA1/2 mutations critical for PARP inhibitor use in personalized therapy. Goal(s): Build a model to predict BRCA mutations using MRI, aiming to reduce NGS costs and improve personalized treatment. Approach: In a retrospective study of 204 patients, BRCA mutations were identified with NGS. T2-weighted MRIs were segmented, and 1,422 radiomic features extracted. Feature selection included ICC, LASSO, and classifiers (KNN, RF, GNB) were applied. Results: The KNN model achieved AUROCs of 1.00 (train) and 0.80 (test) with TNR of 0.84. RF and GNB had TNRs of 0.86 and 1.00 for negatives. Impact: MRI-based radiomics models offer a non-invasive, cost-effective alternative to NGS, potentially reducing the need for NGS testing. These models could aid in drug selection and prognosis prediction, enhancing personalized treatment strategies.