<![CDATA[BACKGROUND Breast imaging reporting and data system (BI-RADS) is a globally recognized method for categorizing breast lesions. However, it is dependent on subjective interpretation, which can result in variability between radiologists. Currently, no scoring system exists to assist in categorizing borderline findings as either benign or malignant. This study aimed to introduce a scoring system that classifies borderline breast lesion findings without altering the standard BI-RADS interpretation and minimizes inter-reader variability. METHODS This single-center retrospective cross-sectional study included 215 women who underwent breast ultrasound (US) and histopathology between January 2021 and December 2022, excluding those with non-neoplastic breast lesions. The index test was US BI-RADS features, and the reference standard was histopathology. Prevalence ratios (PRs) and probability scores were used to assess the risk contribution of individual US features. The diagnostic performance of the BI-RADS interpretation strategy and predictive scoring model was compared with that of the standard BI-RADS classification using SPSS software version 26. RESULTS Margin, orientation, and age had the highest PRs of malignancy, with odds ratios and 95% confidence intervals of 39.86 (13.19–120.47), 17.47 (2.42–125.72), and 9.74 (3.51–27.04), respectively. These three characteristics increased the probability of malignancy by 99.53%. A comparison of diagnostic tests between the modified BI-RADS interpretation strategy and the standard BI-RADS classification revealed improvements in specificity (94.0% versus 84.7%), positive predictive value (85.0% versus 70.5%), accuracy (89.3% versus 84.7%), and area under the curve (0.931 versus 0.873). CONCLUSIONS According to the interpretation strategy, margin, orientation, and age, this US scoring model appear to be promising tools for classifying borderline malignant masses.]]>
