Objective This study aims to develop and validate a model based on clinical and radiomic features to investigate its value in distinguishing between benign and malignant breast nodules. Methods The study included 139 patients with breast diseases, divided into a training set (n=111) and a validation set (n=28) at an 8:2 ratio. All patients’ dynamic contrast-enhanced MRI (DCE-MRI), diffusion-weighted imaging (DWI), T1-weighted imaging (T1WI), T2-weighted imaging (T2WI), and ultrasound (US) images were uploaded to the 3D Slicer software. Using a double-blind method, regions of interest (ROIs) were manually delineated on T1WI, T2WI, DWI, the first phase of DCE, and US images. Radiomic models were constructed using radiomic features. A comprehensive model was built by combining clinical and radiomic features through multivariate logistic regression and visualized as a nomogram. The area under the curve (AUC), accuracy, specificity, and sensitivity of five different radiomic models were compared to evaluate their discriminatory performance. A combined model was created using the T2WI radiomic model and clinical features, and the predictive performance of the clinical model, radiomic model, and combined model were compared and validated. Results For the T1WI radiomic model, the AUC values for the training and test sets were 0.885 and 0.778, respectively. For the T2WI radiomic model, the AUC values were 0.950 and 0.871. For the DCE radiomic model, the AUC values were 0.854 and 0.749. For the DWI radiomic model, the AUC values were 0.878 and 0.763. For the US radiomic model, the AUC values were 0.878 and 0.737. The combined model using T2WI and clinical features achieved AUC values of 0.975 and 0.942 for the training and test sets, respectively. Conclusion The model combining T2WI and clinical features demonstrated higher value in non-invasively distinguishing between benign and malignant breast nodules.
Liu et al. (Wed,) studied this question.