Radiomics and AI for pre-operative characterization of Breast Cancer using Dynamic Contrast Enhanced MRI

Forum Details

Date: Tuesday, September 9, 2025 11AM Eastern Time

Speakers: Prof. Giorgio De Nunzio and Dr. Luana Conte from the University of Salento, Italy

Breast cancer remains a major health challenge, with accurate preoperative characterization crucial for treatment. Radiomics and artificial intelligence (AI) offer promising solutions.

We first used radiomics and machine learning (ML) to distinguish in situ from invasive breast cancer on dynamic contrast-enhanced MRI (DCE-MRI) in 71 patients. Feature selection and data augmentation improved performance, with Random Forest and Support Vector Machine performing best; shape and texture descriptors were key predictors.

Next, we addressed preoperative detection of triple-negative breast cancer (TNBC) in multiparametric MRI using federated learning (FL) on the multi-institutional MAMA-MIA dataset (>1,200 patients). MRI intensity standardization reduced inter-scanner variability, enhancing model generalizability, while FL enabled privacy-preserving collaboration.

These studies show that radiomics and AI, combined with standardized imaging and federated architectures, can refine preoperative breast cancer diagnosis, supporting precision oncology and multicenter collaborations.