Objective:

To develop and internally validate a combined radiomics-clinical prediction model for axillary pathological complete response (apCR) in clinically node-positive breast cancer patients after neoadjuvant chemotherapy.

Approach:

• Study Design: Single-center retrospective study enrolling 386 cN+ breast cancer patients (training, n = 270; validation, n = 116).
• Data Collection: Pre-NAC DCE-MRI radiomic features were extracted from primary tumors.
• Model Development: LASSO regression selected eight features for Rad-score construction; clinical predictors identified via logistic regression.
• Performance Evaluation: Model performance assessed using AUC, calibration metrics, and decision curve analysis.

Key Findings:

• Overall apCR rate was 43.5% (168/386).
• Combined model achieved a validation AUC of 0.703 (95% CI, 0.610–0.792), outperforming the radiomics-only model (ΔAUC = 0.094, P = 0.004) but not the clinical-only model (ΔAUC = 0.020, P = 0.713).
• Calibration slope of the combined model was 0.811 with an intercept of 0.018, indicating moderate overfitting.
• Risk stratification showed low (18.8%), intermediate (48.0%), and high (58.8%) apCR rates across tertiles.
• After bootstrap bias correction, the optimism-corrected training AUC of the combined model was 0.742.

Interpretation:

The combined model demonstrated moderate discriminatory ability but did not significantly outperform clinical predictors alone.

Limitations:

• Current misclassification rate precludes direct clinical application for surgical de-escalation.
• External multicenter validation is warranted.

Conclusion:

The combined model shows potential but requires further validation before clinical application.