A multimodal ultrasound-based model combining tumor radiomics and axillary lymph node morphologic classification for predicting axillary nodal burden in breast cancer.

OBJECTIVE

To develop and validate a multimodal ultrasound-based model integrating tumor radiomics and axillary lymph node (ALN) morphologic classification for preoperative prediction of axillary nodal metastasis burden in breast cancer.

METHODS

This retrospective study included 583 patients with pathologically confirmed breast cancer, randomly divided into training and testing cohorts (7:3). Radiomic features were extracted from primary tumor ultrasound images, and ALN classification and clinicopathological variables were collected. A hierarchical modeling strategy was applied: first-level models predicted ALN metastasis (N0 vs. N+), and second-level models distinguished axillary nodal tumor burden (N1-2 vs. N≥3). Machine learning algorithms were used to construct radiomics models, and combined models were developed by integrating the radiomics score (Rad-score) with independent predictors. Model performance was evaluated using ROC analysis and decision curve analysis.

RESULTS

At the first level, the radiomics model achieved an AUC of 0.79, while the combined model incorporating ALN classification improved performance to 0.90. At the second level, the radiomics model yielded an AUC of 0.74, which increased to 0.78 after integration. Ki-67 was identified as an independent predictor. Subgroup analysis showed that the combined model performed consistently well in predicting ALN metastasis across Ki-67 subgroups, whereas its performance in distinguishing nodal burden was superior in the low Ki-67 subgroup (AUC 0.82 vs. 0.68).

CONCLUSION

The proposed multimodal model enables accurate, noninvasive prediction of ALN metastasis and axillary nodal tumor burden. Integrating tumor radiomics with lymph node morphology may support individualized risk stratification and treatment planning and optimize axillary management.