Abstract
BACKGROUND: Invasive ductal carcinoma (IDC) is the most common histological subtype of breast cancer, and axillary lymph node metastasis (ALNM) is a pivotal factor in clinical staging, prognostic assessment, and treatment planning. This study aims to develop and validate a deep learning (DL) model based on dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) for the prediction of ALNM in IDC patients. METHODS: This multicenter study conducted a retrospective analysis of DCE-MRI images from 520 patients diagnosed with IDC of the breast. The training and internal validation sets consisted of 411 patients from The First Hospital of Qinhuangdao, while the external testing set included 109 patients from the Maternal and Child Health Hospital of Qinhuangdao. Radiomics and DL features were extracted separately from the DCE-MRI images. We evaluated five models (Clinical, Radiomics, Radiomics-Clinical, DL, DL-Clinical) using radiomics features, DL features, and clinical features. Finally, the predictive performance of the models was evaluated using the receiver operating characteristic (ROC) curve and the area under the curve (AUC). RESULTS: The AUCs for the Clinical model and Radiomics model, which are machine learning models, and the DL-model, were 0.807, 0.840, and 0.865, respectively. The combined models incorporating clinical features, namely the Radiomics-Clinical and DL-Clinical models, achieved AUCs of 0.824 and 0.935, respectively. Among the five models, the DL-Clinical model demonstrated a significant advantage in predicting ALNM. Additionally, this model exhibited robust performance in both internal validation and external testing sets, with AUCs of 0.946 and 0.951, respectively. CONCLUSIONS: The DCE-MRI-based DL-Clinical model provides a non-invasive adjunct tool for preoperative identification of ALNM in patients with breast IDC, thereby enhancing the efficacy of personalized treatment strategies and improving patient quality of life.