Machine Learning Model for Predicting Axillary Lymph Node Metastasis in Clinically Node Positive Breast Cancer Based on Peritumoral Ultrasound Radiomics and SHAP Feature Analysis

Recently, we had the honor of reading the article titled “Machine Learning Model for Predicting Axillary Lymph Node Metastasis in Clinically Node Positive Breast Cancer Based on Peritumoral Ultrasound Radiomics and SHAP Feature Analysis.”1 The clinical data, ultrasound data, and postoperative pathological results of 321 patients with breast cancer were retrospectively collected (224 cases in the training group and 97 cases in the validation group). Through correlation analysis, single factor analysis, and Lasso regression analysis, independent risk factors related to axillary lymph node metastasis of breast cancer were identified from conventional ultrasound and immunohistochemical indicators, and a clinical feature model was constructed. In addition, the ultrasonic image features of 1–5 mm in and around the tumor were extracted, and the radiomics feature formula was established. In addition, the diagnostic effects of six machine learning models (logistic regression, decision tree, support vector machine, extreme gradient enhancement, random forest, and k-nearest neighbor) were compared by combining clinical features and ultrasonic radiological features. A joint prediction model based on the optimal machine learning algorithm is constructed. The AUC of the clinical feature model in the training group and the validation group was 0.779 and 0.777, respectively. Radiomics model analysis showed that the model containing the tumor + peritumoral 3 mm region had the best diagnostic effect, and the AUC of the training group and the verification group were 0.847 and 0.844, respectively. The AUC of the joint prediction model based on XGBoost algorithm reached 0.917 and 0.905 in the training group and the verification group, respectively. The combined model has a significant effect on predicting axillary lymph node metastasis in breast cancer. We sincerely appreciate the contributions made by the authors. However, there are issues in this study that require further exploration. First of all, the clinical factors included in this study lack statistical analysis of carcinoembryonic antigen (CEA), carbohydrate antigen 125 (CA125), and carbohydrate antigen 153 (CA153) detection indicators. The high expression of CA15-3 can help to judge the degree of metastasis and recurrence of breast cancer patients. The level of CA125 is helpful to the detection rate of breast cancer and CA125 also has an upward trend in the process of recurrence and metastasis of breast cancer. CEA is a broad-spectrum tumor marker and has an auxiliary reference effect on a variety of tumors. These indicators can reflect the degree of differentiation and malignancy of breast cancer tumors to a certain extent. And these indicators have a certain correlation with tumor metastasis.2-4 Second, in terms of statistical analysis, if the single-factor analysis of P <.05 indicators and then multifactor analysis will be more statistically significant. Finally, the discussion section lacks an explanation for the radiomics features of the Rad score. Finally, we express our gratitude once again for the authors' contributions to this study. We hope our insights prove valuable for their further research, and we look forward to hearing their opinions.