Multi-sequence MRI-based nomogram for prediction of human epidermal growth factor receptor 2 expression in breast cancer

Abstract

Objective: To develop a nomogram based on multi-sequence MRI (msMRI) radiomics features and imaging characteristics for predicting human epidermal growth factor receptor 2 (HER2) expression in breast cancer (BC). Methods: 206 women diagnosed with invasive BC were retrospectively enrolled and randomly divided into a training set (n = 144) and a validation set (n = 62) at the ratio of 7 : 3. Tumor segmentation and feature extraction were performed on dynamic contrast-enhanced (DCE) MRI, T2-weighted imaging (T2WI), and apparent diffusion coefficient (ADC) map. Radiomics models were constructed using radiomics features and the radiomics score (Rad-score) was calculated. Rad-score and significant imaging characteristics were included in the multivariate analysis to establish the nomogram. The performance was mainly evaluated via the area under the receiver operating characteristic curve (AUC). Results: Edema types on T2WI (OR = 4.480, P = 0.008), enhancement type (OR = 7.550, P = 0.002), and Rad-score (OR = 5.906, P < 0.001) were independent imaging predictors for HER2 expression. Radiomics model based on msMRI (including DCE-MRI, T2WI, and ADC map) had AUCs of 0.936 and 0.880 in the training and validation sets, respectively, exceeding the AUCs of one sequence or dual sequences. With the combination of edema and enhancement types, the nomogram achieved the highest performance in the training set (AUC: 0.940) and validation set (AUC: 0.893). Conclusion: The developed multi-sequence MRI-based nomogram presents a promising tool for predicting HER2 expression, and is expected to improve the diagnosis and treatment of BC.

Keywords

• Breast cancer
• Magnetic resonance imaging
• Radiomics
• Nomogram
• Human epidermal growth factor receptor 2