A multiparametric classification system for lesions detected by breast magnetic resonance imaging

Abstract

BackgroundTo validate a new categorisation scheme for suspicious breast lesions according to the well-defined Breast Imaging Reporting and Data System (BI-RADS) magnetic resonance imaging (MRI) lexicon descriptors, apparent diffusion coefficients (ADC), T2-weighted signal intensity (SI), and kinetic curve assessment categories. MethodsThe MRI descriptors and ADC were analysed in 697 lesions detected in 499 subjects. The descriptors were classified into Minor, Intermediate, and Major findings, and were divided into the BI-RADS subcategories 3, 4A, 4B, 4C, and 5 according to the number of descriptors. Positive predictive values (PPV) were calculated for each descriptor. The descriptors were then fitted into a multinomial logistic regression model to determine the odds ratio for a malignant diagnosis. The PPV were measured for the new categories and compared with the assigned PPV of the BI-RADS descriptors. ResultsThe PPV for MRI descriptors ranged from 17.9%–100%. Of the 697 lesions assessed, 19 (2.7 %) were categorized as BI-RADS 3, 27 (3.9 %) as 4A, 53 (7.6 %) as 4B, 174 (25.0 %) as 4C, and 424 (60.8 %) as 5. None of the subjects in BI-RADS category 3 had a malignant diagnosis. The PPV for malignancy increased progressively with increasing BI-RADS category (4A, 11.1 %; 4B, 28.3 %; 4C, 64.4 %; 5, 94.8 %). All descriptor groups were significant in the logistic regression model. ConclusionsThis study shows that using BI-RADS MRI descriptors together with ADC and T2-weighted SI in a multiparametric classification system can yield an applicable categorisation of lesions with PPV values within the recommended ranges for BI-RADS categories.