Utilization of MRI-based radiomics nomogram for predicting HER2-zero, -low, and -overexpression breast cancer

This study sought to retrospectively investigate the clinical utility of a radiomics nomogram based on MRI for stratifying HER2 expression status in breast tumors into three categories.

This study recruited females from two centers who had a pathological confirmation of invasive breast cancer (BC) between January 2024 and March 2025. Based on the T2-weighted image (T2WI) and dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI), radiomic features were extracted. Feature selection was subsequently performed by analysis of variance (ANOVA), Spearman's correlation analysis, the Wilcoxon test, and the least absolute shrinkage and selection operator (LASSO) method. Two classification tasks were determined: Task 1, differentiating HER2 overexpression (positive) from HER2-low/zero (negative) tumors and Task 2, distinguishing HER2-low from HER2-zero tumors. To pinpoint clinicopathological markers associated with the HER2 status, univariate and multivariate logistic regression analyses were carried out. Moreover, a nomogram, integrated by key MRI radiomics features, was crafted. Model efficacy was gauged using the receiver operating characteristic (ROC) curve, sensitivity, specificity, and accuracy. Clinica applicability was examined via decision curve analysis (DCA).

The integrated model incorporating T2WI and DCE-MRI features achieved better results than models using only one modality. In Task 1, the nomogram demonstrated good calibration and discrimination capabilities, with an area under the curve (AUC) of 0.82 and 0.80 in training and validation sets. For Task 2, the nomogram achieved an AUC of 0.88 (training) and 0.80 (validation).

The radiomics nomogram provides a non-invasive choice for predicting HER2 expression in BC, which may assist in personalized treatment planning.