Improving Prediction Accuracy of Residual Axillary Lymph Node Metastases in Node-Positive Triple-Negative Breast Cancer: A Radiomics Analysis of Ultrasound-Guided Clip Locations Using the SHAP Method

Rationale and Objectives

To construct a radiomics nomogram derived from multiparametric ultrasound (US) imaging using the SHapley Additive exPlanations (SHAP) method for the accurate identification of residual axillary lymph node metastases post-neoadjuvant chemotherapy (NAC) among patients with triple-negative breast cancer (TNBC).

Methods

A total of 405 consecutive patients with pathologically confirmed TNBC between 2016 and 2023 were recruited in the study and were divided into training (n = 284) and validation cohorts (n = 121). Radiomics features capturing detailed tumor characteristics were extracted from pre-NAC gray-scale US images at the locations of US-guided clip placement. The least absolute shrinkage and selection operator and the maximum relevance minimum redundancy algorithm were employed to identify key features and formulate the radiomics signature (RS). A nomogram based on US radiomics was then constructed using multivariable logistic regression analysis. The predictive efficacy of this model was evaluated through receiver operating characteristic curve analysis, calibration assessment, and decision curve analysis. SHAP summary plots were used to visualize the distribution of SHAP values across all features.

Results

The nomogram integrates clinical and US characteristics with RS, yielded optimal AUC of 0.922 (95% CI, 0.890–0.954) in the training cohort, 0.904 (95% CI, 0.853–0.955) in the validation cohort. The calibration and decision curves confirmed favorable calibration and clinical value of the nomogram. SHAP provided further insight into the contributions of each feature to the model’s outcomes.

Conclusion

The combined multiparametric US based radiomics nomogram plays a potential role in predicting residual axillary lymph node metastases after NAC in TNBCs.