Answers are grounded in this paper and cite their sources.
Purpose This study explores the predictive value of intramural and peritumoral radiomics features based on dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) for lymphovascular invasion (LVI) in breast cancer. It compares the predictive efficacy of different peritumoral regions and feature fusion methods, and constructs a combined prediction model integrating clinical characteristics. Materials and Methods Retrospectively collected images from 757 breast cancer patients with pathologically confirmed diagnoses who underwent breast DCE-MRI examinations at two medical centers. Data from Center 1 served as the training set (n = 529), while data from Center 2 formed the external validation set (n = 228). Tumor lesions were semi-automatically delineated using ITK-SNAP software, with peritumoral regions automatically expanded. Regions of interest (ROIs) were selected on the maximum intensity phase of DCE-MRI. Radiomics features were extracted using PyRadiomics across four modalities: intratumoral, peritumoral (1mm, 3mm, 5mm, 7mm, 10mm, 15mm, 20mm), Tumor+Peri7mm, and ImageFusion7 (a fusion of tumor and optimal 7mm peritumoral regions). Data dimensionality reduction and feature selection were performed using Spearman correlation analysis, Least Absolute Shrinkage and Selection Operator (LASSO) regression, and Maximum Relevance-Minimum Redundancy (mRMR) methods. Five classifiers—support vector machine (SVM), extreme trees (ET), extreme gradient boosting (XGBoost), light gradient boosting machine (LightGBM), and multilayer perceptron (MLP)—were evaluated under different feature models to identify the optimal classifier for constructing the radiomics model. Clinical Features: Single- and multi-factor logistic regression screened clinical risk factors to construct a clinical model. Comprehensive evaluations were conducted on the clinical model, the optimal radiomic model, and a combined model incorporating clinical features with optimal radiomic features. Receiver operating characteristic (ROC) curves analyzed predictive performance, with area under the curve (AUC) calculated. A fusion model was established by integrating the radiomic score (rad-score) from the highest-performing predictive model and visualized via a nomogram. The clinical utility of the predictive models was assessed using calibration curves and decision curve analysis (DCA). Results Clinical characteristics: The AUC values of the clinical model based on histological grade, sentinel lymph node status, and TIC were 0.738 (95% CI: 0.6949–0.7802) and 0.749 (95% CI: 0.6830–0.8141) in the training and validation sets, respectively. As a key indicator reflecting lymph node metastasis in primary tumors, sentinel lymph node status is widely recognized for its role in breast cancer staging and prognostic assessment. For instance, in breast cancer, the area under the ROC curve for a non-sentinel lymph node metastasis risk prediction model in sentinel lymph node-positive patients reached 0.7613, demonstrating its clinical utility. Radiomics: Among the various peritumoral models, the 7 mm peritumoral model demonstrated the best performance. Results indicate that the radiomics model based on the ImageFusion7 (intra-tumor and peritumoral image fusion region) demonstrated optimal diagnostic performance, with the XGBoost classifier showing particularly outstanding results: an AUC of 0.861 (95% CI: 0.8302–0.8914) in the training set and an AUC of 0.709 (95% CI: 0.6407–0.7768). When constructing a clinical-imaging fusion model, the ROC curve based on clinical features and Radscore values achieved AUC values of 0.892 (95% CI: 0.8629–0.9165) and 0.778 (95% CI: 0.7140–0.8375) in the training and validation sets, respectively, surpassing all single models and the intra-tumoral/peritumoral model. The DCA curve demonstrated that the combined model exhibited the highest clinical efficacy and closely followed the diagonal line in the calibration curve, indicating superior model fit and generalization capability. Conclusions An image-based model utilizing DCE-MRI effectively predicts lymphovascular invasion (LVI) status in breast cancer, with the ImageFusion7 model—which integrates images from both the tumor interior and a 7mm peritumoral region—demonstrating optimal predictive performance. Combining this model with clinically independent risk factors further enhances its accuracy. The constructed fusion model exhibits strong discriminatory power and clinical applicability, providing a valuable tool for non-invasive preoperative assessment of LVI risk.
No takes yet. Share an insight, caveat, or question.
Li et al. (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: