Predictive Accuracy of a Radiomics Model for Prediction of Axillary Lymph Node Metastasis in Patients with Breast Cancer

Abstract

Background: Axillary lymph node metastasis is an important prognostic factor in breast cancer and directly influences staging, treatment planning, and surgical decision-making. Radiomics may provide a non-invasive method for predicting nodal metastasis using quantitative imaging features.

Objective: To evaluate the predictive accuracy of a radiomics model for axillary lymph node metastasis in patients with histopathologically confirmed breast cancer.

Methods: This analytical cross-sectional study included 70 female patients with newly diagnosed breast cancer who underwent preoperative contrast-enhanced breast imaging followed by surgical axillary assessment. Axillary lymph node status was confirmed using final histopathology as the reference standard. Radiomic features were extracted from the primary breast tumor after image preprocessing and segmentation. Patients were divided into a training cohort of 50 and an internal testing cohort of 20. Clinical, radiomics-only, and combined clinical-radiomics models were developed. Model performance was assessed using area under the receiver operating characteristic curve, sensitivity, specificity, positive predictive value, negative predictive value, accuracy, calibration analysis, and decision-curve analysis.

Results: Axillary lymph node metastasis was present in 27 patients (38.6%), while 43 patients (61.4%) were node-negative. Patients with nodal metastasis had significantly larger tumors, higher frequency of tumor size >3 cm, grade III disease, lymphovascular invasion, and Ki-67 ≥20%. The radiomics score was significantly higher in node-positive patients than in node-negative patients (0.78 ± 0.69 vs. -0.42 ± 0.61; p<0.001). In the training cohort, the clinical, radiomics-only, and combined models achieved AUC values of 0.80, 0.88, and 0.93, respectively. In the internal testing cohort, the corresponding AUC values were 0.76, 0.82, and 0.89. The combined model showed the best diagnostic performance, with 87.5% sensitivity, 83.3% specificity, 85.0% accuracy, 77.8% positive predictive value, and 90.9% negative predictive value in the testing cohort. On multivariable logistic regression, lymphovascular invasion, tumor size >3 cm, and radiomics score were independent predictors of axillary lymph node metastasis.

Conclusion: The combined clinical-radiomics model showed good predictive accuracy for axillary lymph node metastasis in patients with breast cancer and performed better than clinical or radiomics-only models. This non-invasive approach may help improve preoperative nodal risk stratification and support individualized surgical planning.