MRI subregion Radiomics for Predicting Recurrence Risk in ER+/HER2- Breast Cancer

Motivation: The application of 21-gene assays in clinical practice is jeopardized by the cost and availability. Goal(s): To identify MRI subregion radiomics signatures associated with the Oncotype Dx 21-gene recurrence score (RS). Approach: 154 patients were enrolled. Radiomics features were extracted from intratumoral subregions on T2WI and the last phase of post-enhancement (CL) images, which were combined with clininal-imaging features to construct models for distinguishing high (RS ≥ 26) from low (RS < 26) recurrence risk. Results: The model that combined T2 and CL subregional radiomics features with clinical-imaging characteristics had the highest efficacy, with an AUC of 0.81 for the validation cohort. Impact: This study first identified radiomics signatures using intratumoral subregional features to predict RS accurately and cost-effectively in ER+/HER2- breast cancer.