Prediction of Residual Axillary Lymph Node Metastasis Following Neoadjuvant Therapy for Breast Cancer: Habitat Radiomics Analysis Based on Breast MRI.

Rationale

AND

Objectives

To develop and validate a habitat radiomics model based on pretreatment breast magnetic resonance imaging (MRI) in predicting residual axillary lymph node metastasis (RALNM) after neoadjuvant therapy (NAT) in patients with clinically node-positive (cN+) breast cancer.

Materials And Methods

This retrospective study included patients with cN+ breast cancer who underwent NAT at two centers between March 2018 and December 2022. Tumor region on pretreatment dynamic contrast-enhanced (DCE) MRI was segmented into distinct habitats using K-means clustering, and radiomics features were extracted from each habitat to build a habitat radiomics model. The conventional radiomics model was developed for comparison. Furthermore, a combined model integrating clinicopathological features with habitat radiomics signature was built. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC).

Results

Among 686 included women (mean age ± standard deviation, 49.38 ± 10.37 years), RALNM was present in 351 patients (51.17%) after NAT. The habitat radiomics model achieved AUCs of 0.805 (95% confidence interval [CI]: 0.739-0.872) in the internal validation set and 0.802 (95% CI: 0.734-0.869) in the external test set, significantly outperforming the whole-tumor radiomics model in all cohorts (all p < 0.05). Integration of habitat radiomics with independent clinicopathological predictors yielded a combined model with further improved AUCs of 0.871 (95% CI: 0.816-0.926) and 0.857 (95% CI: 0.800-0.915) in the validation and external test sets, respectively.

Conclusion

Habitat radiomics analysis of pretreatment breast DCE-MRI demonstrates promising performance for predicting RALNM after NAT. The combined model incorporating habitat radiomics and clinicopathological factors further improves predictive accuracy.