Multimodal Radiomics Based on Magnetic Resonance Imaging and Mammography for Axillary Lymph Node Metastasis Prediction in Clinically Node-Negative Breast Cancer Patients

This study aimed to develop a noninvasive tool to predict axillary lymph node metastasis (ALNM) using a multimodality radiomics based on mammography (MMG) and magnetic resonance imaging (MRI) and its combination with the conventional clinical model. Datasets from 203 clinically node-negative breast cancer patients were collected and randomly allocated into train/validation (n = 160), and test (n = 43) datasets. Radiomic features derived from both MMG and MRI were computed. Subsequently, the important features were selected by intraclass correlation coefficient (ICC), univariate analysis, and recursive feature elimination (RFE). Multivariate logistic regression was employed to construct 7 models, which were clinical, MMG, MRI, MMG+MRI, MMG+Clinical,MRI+Clinical, and MMG+MRI+Clinical models. The area under the receiver operating characteristic curve (AUC) was used to evaluate the model performance. The MMG+MRI model and its combination with the clinical model achieved significantly superior performance in ALNM prediction compared to the clinical and single-modality models alone, yielding AUC values of 0.875 1 0.052 and 0.874 1 0.050 in the validation, and 0.779 and 0.764 in test dataset, respectively. The MMG+MRI and MMG+MRI+Clinical models demonstrated high performance in ALNM prediction, and hold promise to be an effective noninvasive tool to predict ALNM in clinically node-negative breast cancer patients.