DCE-MRI Based Machine Learning Predictor for HER2-Positive Breast Cancer: A Feasibility and Validation Multicenter Study

Abstract

Abstract Background Human epidermal growth factor receptor 2 (HER2) status of breast cancer plays a critical role in guiding clinical treatment. We aimed to develop and validate a predictive model for HER2 status using preoperative dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI). Methods A total of 570 patients (282, 121 and 167 patients for training, internal and external test sets, respectively) with pathologically confirmed breast cancer and known HER2 status were recruited. A total of 851 radiomics features for each patient were extracted from preoperative DCE-MRI images. VarianceThreshold, SelectKBest and least absolute shrinkage and selection operator (LASSO) regression were applied to identify the optimal predictive features. Logistic regression was adopted to incorporate the Rad-score and clinical predictors into a nomogram. The performance of the nomogram was evaluated by area under receiver operating characteristic curve (AUC), calibration curve and decision curve. Additionally, gene expression analysis based on the Cancer Image Archive database was conducted to validate the biological interpretability of the model. Results Twenty-three radiomics features were selected to calculate the Rad-score. The Rad-score, along with breast imaging reporting and data system (BI-RADS) parameter, were independent predictors for HER2 status and were incorporated into the predictive model. The combined model achieved AUCs of 0.881, 0.883, and 0.798 in the training, internal and external test sets, respectively. Calibration curves demonstrated well agreement between the model predictions and actual HER2 status. Decision curve analysis further confirmed the clinical utility of the model. Differentially expressed genes between HER2-positive and HER2-negative patients were primarily involved in signaling pathways such as PI3K-AKT, endocrine resistance, and p53. Conclusions The combined model, which incorporated the Rad-score and BI-RADS, representing a potential and efficient alternative tool to evaluate HER2 status in breast cancer.