Multiparametric MRI Radiomics for Noninvasive Prediction of HER2-Zero, -Low, and -Positive Breast Cancers

Motivation: HER2 status plays a key role in treatment decisions and prognosis of breast cancer (BC), but biopsy method is invasive and limited by tissue heterogeneity. Goal(s): Investigating the feasibility of multiparametric MRI radiomics for differentiating HER2-zero, HER2-low, and HER2-positive. Approach: 314 patients were finally used. Two tasks were carried out including differentiating HER2-low/positive from HER2-zero (Task1), and distinguishing HER2-low from HER2-positive (Task2). Results: Combining radiomic features obtained from T2WI and DCE showed good performance for predicting HER2 status (AUC=0.861 and 0.822 for Task1 and Task2, respectively). The AUC range was 0.725 to 0.794 on internal and external validation cohorts for the two tasks. Impact: Multiparametric MRI radiomics has the potential to provide a noninvasive tool for identifying HER2-zero, HER2-low, and HER2-positive in breast cancer patients, thus enabling guide the selection of HER2-targed antibody-drug conjugates.