Abstract
Purpose The objective of this research was to investigate the significance of different machine learning models based on dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) combined with clinicopathologic and radiological analysis in predicting axillary lymph node metastasis (ALNM) of primary breast cancer (BC).
Methods The clinical data of 605 patients with BC underwent preoperative DCE-MRI examination from The Cancer Imaging Archive (TCIA) publicly available dataset were retrospectively analyzed and casually seperated into training and test cohort at a ratio of 8:2. After dimensionality reduction and selection, a prediction model was established using machine learning algorithms. Clinicopathologic characteristics were analyzed using univariate and multivariate logistic regression to identify variables for constructing clinical models. Receiver operating characteristic (ROC) curves analysis was used to screen out the best radiomics and clinical models, and a combined model was established. Decision curve analysis (DCA) was used to assess the clinical significance of the combined model.
Results The combined model exhibited superior diagnostic predictive capability in determining the presence or absence of ALNM. The training and test cohorts yielded area under the curve (AUC) values of 0.890 and 0.854, respectively.Additionally, a distinct combined model was developed to distinguish between the N1 group (1-3 ALNM) and the N2-3 group (≥4 ALNM), demonstrating promising efficacy with AUC values of 0.973 and 0.835 in the training and test groups, respectively. Furthermore, the integrated model discriminated between N0, N1, and N2-3, yielding a micro AUC of 0.861 and a macro AUC of 0.812.
Conclusion The integration of radiomics and clinicopathologic characteristics demonstrated outstanding predictive capability for ALNM, potentially offering a non-invasive and effective approach for clinical decision-making.