Value of dynamic enhanced magnetic resonance image-based model in predicting low expression of HER-2 in breast cancer

Abstract

Abstract Objective This study aimed to evaluate the feasibility of evaluating early low expression of HER-2 in patients with breast cancer by applying Dynamic Contrast-Enhanced Magnetic Resonance Imaging (DCE-MRI) based imaging features, which could potentially optimize treatment for patients. Method Clinical and pathology data of 294 female patients with invasive ductal carcinoma confirmed by puncture or surgical pathology were collected. Regions of interest (ROI) were mapped. Features were then extracted from the original Magnetic Resonance Imaging (MRI) image data. Relevant features were screened out by Mann-Whitney U test. Cross-validated LASSO regression was used for feature selection. Inner and outer 10-fold cross-validation (CV) models were used. The inner 10-fold CV was used to select the best model during the Linear SVC modeling in training set, and an outer 10-fold CV was used to validate the efficiency in validation set. Model performance was evaluated by using receiver operator curve (ROC) analysis. The average accuracy, sensitivity, and specificity were calculated. Results After model selection using the inner 10-fold CV in Linear SVC modeling and validation using the outer CV, the average accuracy, sensitivity, and specificity of the validation set were 79.6%, 73.7%, and 85.6%, respectively. The average area under curve (AUC) of ROC analysis was 0.87. The diagnostic efficiency of the replacement dataset after 1000 permutation tests was compared with the original dataset, and the average accuracy, sensitivity, and specificity were all less than 0.05. The differences were all statistically significant. The model established after cross-validation could classify patients as HER2 low expression or HER2 positive. The classification efficiency of the model was higher than the chance level. Conclusion DCE-MRI imaging model can help predict the low expression of HER2 receptor in breast cancer with a high predictive efficiency, which can provide a new method for clinical diagnosis of non-invasive HER2 status.