A Novel Model for Standalone Evaluation of Breast Diffusion-Weighted Imaging: Prediction of BI-RADS Categories

INTRODUCTION: To evaluate whether Breast Imaging Reporting and Data System (BI-RADS) categories can be determined using diffusion-weighted imaging (DWI) alone through a new evaluation model, and to compare DWI-based BI-RADS categorization with multiparametric magnetic resonance imaging (MP-MRI) assessments and available histopathological findings.
METHODS: Anonymised DWI datasets of 400 consecutive patients were retrospectively reviewed, and 354 patients were involved in the final analysis using a newly developed classification model comprising lesion morphology and diffusion restriction patterns. For each patient, the highest BI-RADS category based on DWI was recorded. DWI-based BI-RADS categories were compared with the categories reported on MP-MRI and, when available, with histopathological results. Normalized apparent diffusion coefficient (nADC) values were calculated, and their diagnostic performance in distinguishing benign from malignant lesions was assessed.
RESULTS: Using MP-MRI as the reference standard, the DWI-based model achieved a sensitivity of 64%, specificity of 98%, positive predictive value (PPV) of 90%, negative predictive value (NPV) of 89%, and an overall accuracy of 89.5%. When histopathology was used as the reference, sensitivity, specificity, PPV, NPV, and accuracy were 87%, 68%, 70%, 86%, and 77%, respectively. Malignant lesions demonstrated significantly lower nADC values than benign lesions. An nADC threshold of 0.784 yielded a sensitivity of 68% and specificity of 73% for benign–malignant discrimination.
DISCUSSION AND CONCLUSION: A DWI-based standalone evaluation model incorporating diffusion patterns and morphological features demonstrates acceptable diagnostic performance for BI-RADS categorization and may provide practical guidance when DWI must be interpreted independently.

Keywords: Breast, diffusion-weighted imaging, BI-RADS, magnetic resonance imaging