Two distinct types of linear distribution in nonmass enhancement at Breast MR Imaging: Difference in positive predictive value between linear and branching patterns

Purpose: To compare positive predictive values (PPVs) of linearly distributed nonmass enhancement (NME) with linear and branching patterns and to identify imaging characteristics of NME that would enable classification as Breast Imaging Reporting and Data System category 3 lesions. Materials and Methods: The institutional review board approved this retrospective study and waived the requirement to obtain informed consent. Reports of breast magnetic resonance (MR) examinations (n = 9453) that described NME were reviewed from examinations performed at the study institution from January 2008 to December 2011. NME with linear distribution was allocated to one of two subtypes: linear pattern (arrayed in a line) or branching pattern (with branches). The χ² test, Fisher exact test, or Student t test was performed for univariate analyses. Factors that showed a significant association with outcome at univariate analyses were assessed with multivariate analyses by using a logistic regression model. Interobserver agreement of the two subtypes between initial interpretation and the interpretation by two additional radiologists who were blinded to any clinical or pathologic information was evaluated with κ analysis. Results: Within the 156 linearly distributed NME lesions, the PPV of the branching pattern (71 of 95 lesions [75%]; 95% confidence interval [CI]: 66%, 84%) was significantly higher than that of the linear pattern (five of 61 lesions [8%]; 95% CI: 1%, 15%) (P < .0001). The PPV of linear pattern lesions smaller than 1 cm was 0% (0 of 30 lesions; 95% CI: 0%, 0%). At multivariate analysis, branching pattern and NME lesion size of 1 cm or greater were significant predictors of malignancy (P < .0001 [odds ratio: 21.6; 95% CI: 7.5, 62.2] and P = .015 [odds ratio: 5.8; 95% CI: 1.4, 24.0], respectively). Substantial interobserver agreement was obtained for differentiating the two subtypes, with κ values of 0.64 (95% CI: 0.51, 0.76), 0.70 (95% CI: 0.59, 0.82), and 0.64 (95% CI: 0.51, 0.76) between the initial interpreter and reviewer 1, the initial interpreter and reviewer 2, and reviewer 1 and reviewer 2, respectively. Conclusion: The branching pattern was a significantly stronger predictor of malignancy than was the linear pattern. NME lesions with a linear pattern that are smaller than 1 cm can be managed with follow-up.