Purpose: To determine interobserver agreement and diagnostic accuracy using a lexicon for standardized interpretation of molecular breast imaging (MBI) studies by breast radiologists.
Methods: An MBI lexicon was developed, including descriptors of lesion type, background uptake, and associated findings by a consensus of experts. In an institutional review board-exempted protocol, six breast imaging radiologist observers without prior MBI experience attended a 2-h MBI interpretation training session, including definitions of lexicon terminology, case examples, and ten unknown cases with expert feedback. Following training, each radiologist observer interpreted an independent set of MBI images of 50 breasts, including 20 (40%) with malignancies with a median invasive tumor size of 1.7 cm (range 1.0 to 6.3 cm). The findings were described using the lexicon and each breast was given a final assessment of 1 to 5, paralleling BI-RADS assessment categories. Sensitivity, specificity, positive and negative predictive values were determined with core or surgical pathology results or 1-year imaging follow-up as the reference standard. Interobserver agreement for lesion-type classification, lesion and background uptake intensity, and final assessments were determined using Cohen's kappa.
Results: For the six observers, median sensitivity was 1.0 (range 0.90-1.0), specificity 0.88 (range 0.83-0.97), and AUC 0.94 (range 0.93-0.98). Fair interobserver agreement was shown for background uptake (κ = 0.31). Agreement was substantial for lesion type (κ = 0.79) and non-mass distribution (κ = 0.63), and near-perfect for final assessment (κ = 0.84).
Conclusion: Dedicated breast imaging radiologists, newly trained to interpret MBI with the proposed lexicon, achieved high agreement and diagnostic accuracy.