Multiparametric MRI-based radiomics nomogram for predicting the hormone receptor status of HER2-positive breast cancer

Abstract

To investigate the value of multiparametric magnetic resonance imaging (MRI)-based radiomics nomograms for predicting the hormone receptor (HR) status of HER2-positive breast cancer. Patients with HER2-positive invasive breast cancer were divided randomly into training (68 patients) and validation (30 patients) sets. All were classified as either HR-positive (HR+) or negative (HR-) at histopathology. Two radiologists outlined the three-dimensional (3D) volumetric regions of interest (VOI) on the MRI images. Features (n=1,096) were extracted from the T2-weighted imaging (WI), apparent diffusion coefficient (ADC), and dynamic contrast-enhanced (DCE) images separately. Dimensionality was reduced using feature screening. Binary radiomics prediction models were established using a logistic regression classifier and were validated in the validation set. To construct a nomogram, independent predictors were identified using multivariate logistic regression analysis. The predictive efficacy of the model was assessed using the area under the receiver operating characteristic curve (AUC). Ten radiomics features were obtained after feature dimensionality reduction based on the merged T2WI, ADC, and DCE images. The diagnostic efficacy of the radiomics signature using the three sequences was better than that of any single sequence (training set AUC: 0.797; validation set AUC: 0.75). Using multivariate logistic regression analysis, the independent predictors for identifying HR status were combined radiomics signature and peritumoural oedema. Nomograms constructed by combining the radiomics signature and peritumoural oedema showed good discrimination in both the training and validation sets (AUC: 0.815 and 0. 805, respectively). A multiparametric MRI-based nomogram incorporating the radiomics signature and peritumoural oedema can assess the HR status of HER2-positive breast cancer. The resulting model can improve diagnostic accuracy, improving patient outcomes.