Combining radiomic features with background parenchymal enhancement from DCE-MRI data for predicting treatment response in breast cancer

Our objective is to predict pathological complete response (pCR) outcome to neoadjuvent chemotherapy in breast cancer patients. We combine radiomic features with background parenchymal enhancement -- computed from standard-of DCE-MRI data from the ISPY-2 trial -- and model treatment outcome via multivariable logistic regression. During training and testing, we demonstrate that models including BPE alongside radiomic and clinical covariates yielded the highest AUC values among all tested regression models, thus improving the prediction of pCR outcome.