Mammography, US, and MRI for Preoperative Prediction of Extensive Intraductal Component of Invasive Breast Cancer: Interobserver Variability and Performances

Background Interobserver variability and performances of imaging studies for predicting an extensive intraductal component (EIC) of invasive breast cancer have not been well established. Materials and Methods Two independent readers retrospectively reviewed every preoperative mammography, ultrasonography (US), and magnetic resonance imaging (MRI) studies of 145 breast cancers in 144 patients with surgically confirmed EIC status and recorded the EIC presence for each study, using our own descriptors referred to in prior articles. Agreement and performance of each study for the prediction of an EIC were assessed. The reference standard was surgical pathologic findings. Results Of 145 breast cancers, an EIC was present in 49 cancers (33.8%) in 49 patients. Overall agreement was perfect on mammography (κ = 0.944), and substantial in US (κ = 0.691) or in MRI (κ = 0.627), and moderate to perfect agreement was found on most descriptors (κ = 0.443-0.81), except some US descriptors (κ = 0.23-0.396). The sensitivity of each study showed no significant differences in both readers (0.73-0.82). For the specificity, mammography was better than US in 2 readers (0.69/0.5; P = .001; 0.72/0.6; P = .007, respectively), and MRI better than US in 1 reader (0.79/0.5; P = .039). Performances between the readers showed no significant differences in each study. Conclusion According to our data, mammography, US, and MRI are valid and reproducible methods for the preoperative prediction of an EIC of invasive breast cancer. However, US shows low agreement on some descriptors and lower performance than mammography or MRI.