Prediction of HER2 expression in breast cancer patients based on multi-parametric MRI intratumoral and peritumoral radiomics features combined with clinical and imaging indicators

Abstract

Objective: To preoperatively evaluate the HER2 status in breast cancer using multiparametric MRI intratumoral and peritumoral radiomics features combined with clinical and imaging characteristics.

Methods: This retrospective study included 252 patients with pathologically confirmed breast cancer (mean age, 50.1 ± 10.1 years) who underwent breast MRI at our hospital. Among them, 202 patients (70 HER2-positive and 132 HER2-negative) were randomly divided into a training set (n = 141) and testing set (n = 61) in a 7:3 ratio from July 2020 to December 2021. The external validation set consisted of 50 breast cancer cases (20 HER2-positive and 30 HER2-negative) from September 2024 to March 2025. Radiomics features extracted from intratumoral and peritumoral regions of the tumor on axial dynamic contrast-enhanced MRI (DCE-MRI), apparent diffusion coefficient (ADC), and T2-weighted fat-suppressed (T2FS) sequences were subjected to dimensionality reduction and model construction using Pearson correlation coefficients, recursive feature elimination, and logistic regression. Univariate and multivariate logistic regression was used to identify the independent risk factors in clinical, pathological and conventional MRI data for constructing the clinical imaging model. The combined model was built from radiomics and clinical imaging features. The area under the receiver operating characteristic curves (AUCs) were used to evaluate the predictive performance of the models.

Results: There were significant statistical differences between the HER2-positive and HER2-negative groups in terms of PR expression (p=0.041), spiculation sign (p<0.001), and uneven margins (p=0.005). The AUC of radiomics models based on DCE, T2FS, and ADC sequences were 0.742, 0.748, 0.791 respectively in the training set, and 0.776, 0.708, 0.713 respectively in the testing set. The AUC of the combined clinical-radiomics model in the training set, testing set and external validation set was 0.923, 0.915 and 0.837, respectively, which was higher than the intratumoral and peritumoral radiomics model based on DCE+T2FS+ADC sequences (0.854,0.748 and 0.770) and clinical imaging model (0.820,0.789 and 0.709).

Conclusions: The combined model based on DCE+T2FS+ADC intratumoral and peritumoral radiomics integrating with clinical imaging features can better predict the HER2 expression status of breast cancer.

Keywords: HER2; breast cancer; intratumoral and peritumoral; multi-parameter MRI; predictive model; radiomics.

Figure 1

Flowchart of the recruitment pathway for patients.

Figure 2

Diagram shows the intratumoral and 3 mm peritumoral image regions within the maximum cross-section on DCE (A, B), T2FS (C, D), and ADC (E, F) sequences of the tumor delineated with 3D Slicer 4.11 software.

Figure 3

Convergence plot of feature coefficients selected by LASSO algorithm based on T1-weighted DCE-MRI for optimal radiomics features.

Figure 4

ROC curves of the models build from radiomic features of DCE, T2FS, and ADC in the training set (A) and testing set (B) for predicting the HER2 status of breast cancer.

Figure 5

ROC curves of the R1, R2, and R3 models in the training set (A), testing set (B) and external validation set (C)) for predicting the HER2 status of breast cancer.

Figure 6

Nomogram of the six variables in the combined prediction model.

Figure 7

DCA curves of testing set for each model.