Development and Validation of Machine Learning Models in Predicting Prognosis of Breast Cancer Patients with Lymph Nodes Metastasis Following Neoadjuvant Chemotherapy.

Background: Lymph node (LN) status is a critical prognostic factor for breast cancer patients undergoing neoadjuvant chemotherapy (NAC). This study aims to develop and validate machine learning models to predict LN response in breast cancer patients with LN metastases.

Methods: Breast cancer patients who received NAC in our hospital were retrospectively analyzed. Clinicopathological data, and MRI imaging were collected. Patients were randomly divided into a training set and a testing set in 7:3 ratio. Radiomic features were extracted from pre-treatment imaging. Random forests and logistic regression were employed alongside Clinical, Clinical-Radiomics and Clinical-Deep-learning-radiomics (Clinical-DLR) in training set. Model performance was evaluated using metrics including sensitivity, specificity, and area under the receiver operating characteristic curve (AUC), accuracy and F1-score. Finally, patients were divided into high-risk and low-risk groups according to the model with the best performance.

Results: Overall, 447 patients were enrolled. In the Clinical, Clinical-Radiomics, and Clinical-DLR logistic regression models, the AUC values in the testing set were 0.738, 0.798, and 0.911, respectively. For the random forest models, the AUC values in the testing set were 0.754, 0.801, and 0.921, respectively. Based on the predictions from the Clinical-DLR model, patients can be stratified into high-risk and low-risk groups. The survival outcomes for high-risk patients were significantly worse compared to those of low-risk patients.

Conclusion: The deep learning radiomics offers a promising approach to predict LN status and survival outcome in breast cancer patients undergoing NAC. This could facilitate personalized treatment strategies and improve clinical decision-making.