Nonmasslike Enhancement at Breast MR Imaging: The Added Value of Mammography and US for Lesion Categorization

Abstract

To determine the value of adding conventional imaging (mammography and ultrasonography [US]) to nonmasslike enhancement (NMLE) analysis with breast magnetic resonance (MR) imaging for predicting malignancy and for building an interpretation model incorporating all imaging modalities. The institutional ethics committees approved the study and granted a waiver of informed consent. In 115 women (mean age, 48.3 years; range, 21-76 years; 56 malignant, 12 high-risk, and 63 benign lesions), 131 NMLE lesions were analyzed. Two independent readers first classified MR images by using descriptive Breast Imaging Reporting and Data System (BI-RADS) criteria (BI-RADS classification with MR images alone [BI-RADS(MR)]) and later repeated this classification, adding information from conventional imaging (BI-RADS classification with combination of MR images and conventional images [BI-RADS(MR+Con)]). Lesion diagnosis was established with surgical histopathologic findings (n = 68), percutaneous biopsy results (n = 25), or 2 years of stability at MR imaging (n = 38). Receiver operating characteristic curves were built to compare BI-RADS(MR) with BI-RADS(MR+Con). A multivariate interpretation model was constructed and validated in a distinct cohort of 44 women. Values for inter- and intraobserver agreement, respectively, were better for BI-RADS(MR+Con) (κ = 0.847 and 0.937) than for BI-RADS(MR) (κ = 0.748 and 0.861). For both readers, the areas under the receiver operating characteristic curve (AUCs) for diagnosis of malignancy were also superior when BI-RADS(MR+Con) (AUC = 0.91 [reader 1] and 0.93 [reader 2]) was compared with BI-RADS(MR) (AUC = 0.84 [reader 1] and 0.87 [reader 2]) (P < .05). An interpretation model combining conventional imaging with MR imaging criteria showed very good discrimination (AUC = 0.89 [training set] and 0.90 [validating set]). Adding conventional imaging to NMLE lesion characterization at breast MR imaging improved the diagnostic performance of radiologists, and the interpretation model used offers good accuracy with the potential to optimize the reproducibility of NMLE analysis at MR imaging.