Prediction of breast cancer recurrence based on automatically extracted quantitative MR features

Motivation: There has been growing interest in predicting breast cancer outcomes using MR imaging features related to cancer or background parenchymal enhancement. Goal(s): To evaluate the potential for recurrence prediction by extracting features from both breasts using a fully automated quantitative method. Approach: Quantitative imaging features were extracted by applying automatically segmented masks to the subtraction images.Using these features, a recurrence prediction model was trained and evaluated. Results: Incorporating features from the contralateral FGT mask slightly improved prediction accuracy compared to using cancer mask features alone. Impact: We suggested a cancer recurrence prediction model using breast MRIs from over 1,600 subjects, employing an automated feature extraction process to investigate its feasibility. Additionally, incorporating features from both the ipsilateral and contralateral sides demonstrated an improvement in predictions.