A multimodal imaging-based decision tree for preoperative axillary risk stratification in breast cancer: a...

To evaluate the clinical utility of a preoperative multimodal imaging-based decision tree for axillary risk stratification in breast cancer and to explore its associations, in a retrospective manner with axillary management patterns and short-term postoperative outcomes. This retrospective single-center study included 204 women with primary breast cancer who underwent surgery between January 2023 and December 2024. Preoperative breast MRI and axillary ultrasound performed within 2 weeks before surgery were reviewed in a blinded manner. Postoperative paraffin pathology served as the reference standard for axillary lymph node metastasis (ALNM). Imaging variables associated with ALNM were identified by multivariable logistic regression, and a CART decision tree was constructed for preoperative risk stratification. Model discrimination was evaluated in the overall cohort and in randomly split training and validation sets. Retrospective analyses were further performed to examine the associations of model-defined risk strata with actual axillary surgical management, postoperative morbidity, and 1-year disease-free survival (DFS). Five imaging variables were associated with ALNM: lesion margin morphology, RI, lymph node hilum integrity, lymph node microcalcification, and Adler blood flow grade. The decision tree showed moderate discriminative ability, with an AUC of 0.829 in the overall cohort and AUCs of 0.835 and 0.803 in the training and validation sets, respectively. Model-defined risk strata showed a stepwise increase in the observed rate of ALNM and were retrospectively associated with different patterns of axillary surgery and postoperative morbidity. Exploratory survival analyses based on 1-year follow-up did not support definitive conclusions regarding benefit from postoperative adjuvant therapy. In conclusion, a decision tree derived from routine preoperative breast MRI and axillary ultrasound features may serve as an interpretable tool for preoperative axillary risk stratification in breast cancer. Its potential relevance to axillary management should be interpreted as exploratory and hypothesis-generating, and prospective multicenter validation is required before clinical decision support can be considered.