MRI-Based Radiomics Model for Classifying Axillary Lymph Node Burden and Disease-Free Survival in Patients With Early-Stage Breast Cancer.
Authors
Affiliations (5)
Affiliations (5)
- Department of Radiology, The First Affiliated Hospital of Wenzhou Medical University, Wenzhou, China.
- Key Laboratory of Intelligent Medical Imaging of Wenzhou, The First Affiliated Hospital of Wenzhou Medical University, Wenzhou, China.
- Department of Radiology, The Second Affiliated Hospital of Wenzhou Medical University, Wenzhou, China.
- Department of Radiological Sciences, University of California, Irvine, California, USA.
- Department of Medical Imaging and Radiological Sciences, Kaohsiung Medical University, Kaohsiung, Taiwan.
Abstract
Axillary lymph node (ALN) burden is a key prognostic determinant in breast cancer and plays an important role in diagnosis and treatment planning. The noninvasive assessment of ALN burden might improve patient stratification and guide individualized treatment. To explore the potential of MRI-based radiomics in preoperative classification of ALN burden in early-stage breast cancer and to assess survival differences between patients with high- and low-ALN burden. Retrospective. Pathologically confirmed breast cancer patients (n = 343): training (n = 170), testing (n = 73) and internal validation (n = 50) from center 1; center 2 (n = 50) for external validation. 3T, dynamic contrast-enhanced (DCE) sequence. Four different machine learning classifiers were used to develop clinical, radiomics, and combined models for preoperative ALN burden assessment (66 high-burden cases). DCE-MRI radiomics features were extracted, and the optimal model was used to determine the Radscore. A clinical model was derived from clinicopathological variables, and integrated with the Radscore to form a combined model. Kaplan-Meier and Cox regression analyses were performed to compare disease-free survival (DFS) between high- and low-burden groups. Intraclass Correlation Coefficient (ICC), LASSO, logistic regression, Mann-Whitney U tests, Chi-squared tests, DeLong's test, Area Under the Curve (AUC), Decision Curve Analysis (DCA), calibration curves and Kaplan-Meier analysis, with p < 0.05 as significant. The Random Forest-based combined model yielded AUCs of 0.881 (95% CI, 0.811-0.941) in the training set, 0.826 (0.716-0.917) in the testing set, 0.912 (0.811-0.985) in the internal validation set, and 0.881 (0.737-0.985) in the external validation set. When using the cut-off value determined from the training set, the overall accuracy was 0.759, 0.795, 0.840, and 0.860, respectively. Kaplan-Meier analysis revealed significant DFS differences between the model-classified high- and low-burden groups (p = 0.022, HR = 2.9). MRI-based radiomics models show promise for noninvasive evaluation of ALN burden and prognostic stratification of survival outcomes in breast cancer patients. Stage 2.