Abstract
BACKGROUND: The accurate preoperative assessment of axillary lymph node (ALN) status is critical for therapeutic decision-making in primary breast cancer (BC), yet current methods are either invasive or lack precision. The objective of this study was to investigate the performance of machine learning models based on dynamic contrast-enhanced magnetic resonance imaging (MRI), in conjunction with clinicopathologic data, in predicting different American Joint Committee on Cancer (AJCC) lymph node (N) stages in patients with BC. METHODS: The data of 605 BC patients were retrospectively analyzed and separated into training and test sets. Following dimensionality reduction and feature selection, a predictive model was established via machine learning techniques. Clinicopathologic features were assessed through both univariable and multivariable logistic regressions (LRs) to select variables for constructing clinical models. The optimal radiomics and clinical models were identified via receiver operating characteristic (ROC) curve analysis and integrated into a combined model. The clinical utility of this combined model was evaluated via decision curve analysis (DCA), which confirmed its superior diagnostic accuracy in detecting axillary lymph node metastasis (ALNM). RESULTS: The combined model yielded area under the curve (AUC) values of 0.890 and 0.854 in the training and test sets, respectively. Additionally, in differentiating the N1 group from the N2-3 group, the combined model showed strong performance, with AUC values of 0.973 and 0.835 in the training and test sets, respectively. Moreover, the model effectively classified the N0, N1, and N2-3 groups, achieving a micro-AUC of 0.861 and a macro-AUC of 0.812. CONCLUSIONS: The integration of radiomics features with clinicopathologic characteristics provides a robust predictive tool for ALNM, potentially offering a noninvasive and effective approach for clinical decision-making.