Evaluation of breast lesions based on modified BI-RADS using high-resolution readout-segmented diffusion-weighted echo-planar imaging and T2/T1-weighted image

Abstract

PurposeTo evaluate the diagnostic performance of a non-contrast magnetic resonance imaging (MRI) protocol combining high-resolution diffusion-weighted images (HR-DWI) using readout-segmented echo planar imaging, T1-weighted imaging (T1WI), and T2-weighted imaging (T2WI), using our modified Breast Imaging-Reporting and Data System (modified BI-RADS). MethodsTwo experienced radiologists, blinded to the final pathological diagnosis, categorized a total of 108 breast lesions (61 malignant and 47 benign) acquired with the above protocol using the modified BI-RADS with a diagnostic decision tree. The decision tree included subcategories of category 4, as in mammography (categories 2, 3, 4A, 4B, 4C, and 5). These results were compared with the pathological diagnoses. ResultsThe area under the ROC curve (AUC) was 0.89 (95% confidence interval [CI]: 0.83–0.95) for reader 1, and 0.89 (95% CI: 0.82–0.96) for reader 2. When categories 4C and above were classified as malignant, the sensitivity, specificity, and accuracy were 73.8%, 93.6%, and 82.4%, for reader 1; and 82.0%, 89.4%, and 85.2% for reader 2, respectively. ConclusionOur results suggest that using HR-DWI, T1WI/T2WI analyzed with a modified BI-RADS and a decision tree showed promising diagnostic performance in breast lesions, and is worthy of further study.