Breast MRI in the evaluation of locally recurrent or new breast cancer in the postoperative patient: correlation of morphology and enhancement features with the BI-RADS category

Abstract

While breast magnetic resonance imaging (MRI) is a highly sensitive test for detecting breast carcinoma, its specificity is lower, and several methods have been described on how to optimize specificity. To compare the specificity and sensitivity of the BI-RADS category with the Fischer score in breast MRI for diagnosing cancer in women previously treated for breast cancer. Women referred for evaluation of possible local recurrence or new breast cancer underwent breast MRI examination. Morphologic and kinetic enhancement characteristics were evaluated. BI-RADS category and Fischer score were assigned for each enhancing lesion and compared using a chi-square test. Sensitivity, specificity,and positive predictive values for 27 morphologic and enhancement characteristics were calculated. Pathologic diagnosis was obtained in all patients with enhancing lesions who had ultrasound or mammographic correlation. In those without correlate, 6-, 12-, and 24-month follow-up breast MRIs were obtained. Interobserver kappa correlation was determined for each variable studied. 34 benign and 32 malignant lesions were identified in 26 of 30 patients. BIRADS category yielded a specificity of 77.1% and a sensitivity of 81.8%. Fischer score had a lower specificity and sensitivity (62.9% and 72.7%, respectively) (P<0.0001). Of the 27 variables studied, >100% enhancement was more sensitive than BI-RADS for malignant lesions. Specificity was highest for rim enhancement (97.1%), but sensitivity was low (24.2%). Interobserver kappa correlation was good for all 27 characteristics(k=0.84), and highest for BI-RADS assessment (k=0.91). BI-RADS category in breast MRI had the highest combination of specificity and sensitivity, and the highest interobserver correlation. Fischer score and other morphologic and enhancement features lack sensitivity or specificity and do not have high positive predictive values when analyzed as single independent variables.