Construction and validation of a pretreatment DCE-MRI-based radiomics prediction model for axillary pathologic response after neoadjuvant therapy in node-positive breast cancer.

Background: Accurate pre-treatment identification of patients likely to achieve axillary pathologic complete response (pCR) after neoadjuvant therapy (NAT) is important for individualized axillary management in initially node-positive breast cancer. Dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) may capture intratumoral and peritumoral heterogeneity associated with treatment sensitivity, but its value for predicting axillary response remains unclear. Therefore, this study aimed to develop and validate a pretreatment DCE-MRI-based radiomics model for predicting axillary pCR after NAT in initially node-positive breast cancer. Methods: In this retrospective study, pretreatment MRI scans were acquired in patients with initially axillary lymph node (ALN)-positive breast cancer who underwent NAT followed by surgery between January 2017 and December 2019. Patients were randomly assigned to primary and validation cohorts at a ratio of 8:2. Radiomics features were extracted from intratumoral and 3-mm peritumoral regions across one precontrast and five postcontrast DCE-MRI phases. The variances of each feature across the six DCE-MRI phases were calculated and used to construct a radiomics signature in the primary cohort. ALN pCR was defined as the absence of micrometastasis and macrometastasis in ALNs on postoperative histopathology. Univariate analyses and multivariate logistic regression analyses were used to identify predictors of ALN pCR. Clinical, radiomics, and combined models were developed based on the independent predictors, and evaluated in the validation cohort. Results: A total of 175 women were included (mean age ± standard deviation, 47.83±10.39 years), of whom 58 achieved axillary pCR. Initial clinical N stage, estrogen receptor (ER) status, human epidermal growth factor receptor 2 (HER2) status, perinodal infiltration, and radiomics score (R-score) were independent predictors of ALN pCR. The combined model showed the best predictive performance, with area under the curves (AUCs) of 0.90 [95% confidence interval (CI): 0.84-0.96] and 0.87 (95% CI: 0.75-0.99) in the primary and validation cohorts, respectively. In the validation cohort, the combined model achieved a sensitivity of 78.6% and a specificity of 85.7%. Conclusions: A pretreatment DCE-MRI-based combined model integrating radiomics and clinicopathologic factors showed promising performance for predicting axillary pCR after NAT in initially ALN-positive breast cancer. This model may support pretreatment risk stratification of axillary response, but external validation is required before clinical application.