Abstract
PURPOSE: We investigated a predictive framework that integrates MRI-derived radiomic characteristics with clinical indicators to assess how breast tumors respond to neoadjuvant chemotherapy. METHODS: A retrospective review was conducted on 301 patients with pathologically confirmed breast cancer. From their baseline MRI scans, 1,196 radiomic features were extracted. Feature reduction was carried out through ANOVA followed by LASSO regression to select the most relevant variables. Eight machine learning algorithms, including Random Forest and XGBoost, were used to develop predictive models incorporating both radiomic and clinical data. Patients were randomly divided into a training set (n = 240) and a validation set (n = 61). Model performance was assessed using the area under the ROC curve (AUC), sensitivity, specificity, and accuracy. RESULTS: In performance evaluation, the Random Forest approach yielded area under the curve values of 0.82 for training and 0.75 for validation, reflecting consistent predictive strength. A nomogram constructed using the selected features achieved an AUC of 0.75 in the validation cohort, with a sensitivity of 0.64 and a specificity of 0.88. CONCLUSION: The integration of imaging biomarkers and clinical profiles enables reliable prediction of tumor response post-NAC, supporting more informed and tailored treatment strategies.