Abstract
Objective: To enhance the diagnostic accuracy of new nodules on the surgical side after breast cancer surgery using machine learning techniques and to explore the role of multifeature fusion. Methods: Data from 137 breast cancer postoperative patients with new nodules from January 2016 to April 2024 were analyzed. Clinical, ultrasound, immunohistochemistry, and surgical features were combined. Multiple machine learning models, including support vector machine (SVM), random forest, gradient boosting, AdaBoost, and XGBoost, were trained and tested. Model performance was evaluated using stratified ten-fold cross-validation. Ablation experiments assessed the impact of different feature combinations on diagnostic performance. Results: The SVM model performed best, with an AUC of 0.8664, an accuracy of 0.8099, a sensitivity of 0.565, and a specificity of 0.9267. Ablation experiments indicated that multifeature fusion significantly improved diagnostic performance, especially when combining clinical, ultrasound, immunohistochemistry, and surgical features. Gradient boosting and random forest models showed slightly inferior performance, while AdaBoost had balanced but lower effectiveness. Conclusion: Machine learning, particularly the multifeature fusion SVM model, shows significant potential in diagnosing new nodules after breast cancer surgery. It can assist doctors in developing more effective treatment plans, improving patient outcomes. Future studies should expand sample sizes, include multicenter data, and explore advanced algorithms to further enhance diagnostic performance.