Radiomics-Based Prediction of Breast Cancer Recurrence, Metastasis, and Clinical Outcomes

Objectives To construct a prognostic framework utilizing preoperative MRI derived radiomics and clinical characteristics in the early prediction of recurrence and metastasis for breast cancer patients. Methods In this retrospective study, breast cancer patients with preoperative MR scans were analyzed. Radiomic features from T2WI, CE-T1WI, and DWI were extracted and refined using ICC analysis and LASSO method. Clinical characteristics were selected via univariate logistic regression. Clinical model, radiomic model, clinical-radiomics score model were constructed using multivariate logistic regression. Model performance was evaluated using AUC, accuracy, sensitivity, and specificity, with AUC comparisons via the DeLong test. Calibration curves and decision curves assessed model fit and clinical benefit, respectively. The log-rank test was used for disease-free survival analysis. Results The study comprised a total of 153 patients, with 109 patients assigned to the training group and 44 patients assigned to the test set. The clinical-radiomics score model demonstrated superior performance compared to the clinical model (AUC = 0.97 vs. 0.74 for the training cohort, p  < 0.001; AUC = 0.87 vs. 0.66 for the test cohort, p  = 0.011). The radiomics model demonstrated superior performance compared to the clinical model, with an AUC of 0.97 versus 0.74 in the training cohort (p < 0.001), and an AUC of 0.86 versus 0.66 in the test cohort (p = 0.046). However, there was no significant advantage observed when combining the clinical and radiomics scores, as the AUC remained at 0.97 for the training cohort (p < 0.504) and 0.87 for the test cohort (p = 0.614) when compared to the radiomics model alone. The log-rank test demonstrated that, according to the clinical-radiomics score model, the groups predicted to be at high risk of recurrence and metastasis exhibited significantly shorter disease-free survival compared to those in the low-risk groups (p < 0.001). Conclusions The prognostic model presented in this research exhibits remarkable accuracy in detecting high-risk recurrence and metastasis among breast cancer patients, thus laying the groundwork for the implementation of individualized treatment protocols in breast cancer care.