The clinical value of radiomics models based on multi-parameter MRI features in evaluating the different expression status of HER2 in breast cancer

Accurate preoperative non-invasive assessment of HER2 expression in breast cancer is crucial for personalized treatment and prognostic stratification. To evaluate the effectiveness of radiomics models based on multi-parametric magnetic resonance imaging (MRI) in distinguishing HER2 expression status in invasive breast cancer. We conducted a retrospective analysis of baseline MRI scans and clinical data from 400 patients with breast cancer between January 2018 and December 2019. Two-dimensional regions of interest were manually segmented on the maximum tumor images obtained from turbo inversion recovery magnitude (TIRM), dynamic contrast-enhanced magnetic resonance imaging phase 2 (DCE2), dynamic contrast-enhanced magnetic resonance imaging phase 4 (DCE4), diffusion-weighted imaging (DWI), and apparent diffusion coefficient (ADC) sequences using ITK-SNAP software. Features were extracted and screened for dimensionality reduction. Logistic regression models were developed to predict HER2 expression status. In distinguishing HER2-overexpression from non-HER2-overexpression, the DCE2 model outperformed other single-parameter models, with areas under the curve (AUCs) of 0.91 (training) and 0.88 (test). Combination models with DCE features showed significantly improved performance (P ≤ 0.001). The multiparameter model achieved the highest AUCs of 0.93 (training) and 0.91 (test). In distinguishing HER2-low from HER2-zero, the TIRM model performed best among single-parameter models, with AUCs of 0.80 (training) and 0.72 (test). The multiparameter model further enhanced prediction, yielding an AUC of 0.83 (test). Radiomics models based on multi-parametric MRI features demonstrated strong clinical utility in assessing HER2 expression status in invasive breast cancer, particularly in identifying HER2-overexpression and HER2-low expression subtypes.