Prediction of response to targeted therapy in HER2-positive breast cancer using MR radiomics

Motivation: There is a need for non-invasive approaches to predict HER2-targeted therapy response in HER2-positive breast cancer patients. However, studies in this area remain limited. Goal(s): This study assessed the ability of response prediction using radiomic features extracted from dynamic contrast enhancement (DCE) MRI. Approach: Radiomic features were extracted from DCE images with and without subtraction from the baseline images to construct machine-learning classification models. Model performance was evaluated with ROC curves. Results: Predictive models with combined radiomic features from raw and subtracted images demonstrated better performance with an AUC of 0.883 compared to models with features from either raw or subtracted images. Impact: This study demonstrated that combining raw and subtracted dynamic contrast enhancement images could enhance response prediction to targeted therapy in HER2-positive breast cancer. Our findings can facilitate personalized treatment for breast cancer patients.