Arpita Swami MBBS, Presenter: Nothing to Disclose

Sughra Raza MD, Abstract Co-Author: Research grant, Hitachi, Ltd

Rebecca Tauber Sivarajah MD, Abstract Co-Author: Nothing to Disclose

Sona Ajit Chikarmane MD, Abstract Co-Author: Nothing to Disclose

Shiva Gautam PhD, Abstract Co-Author: Nothing to Disclose

Robyn L. Birdwell MD, Abstract Co-Author: Nothing to Disclose

To assess variability in use of the ACR BI-RADS lexicon for MRI in the analysis and reporting of morphologic and kinetic features of breast MRI lesions.

This retrospective HIPAA compliant study consisted of 102 MR detected masses and nonmasslike enhancement (NMLE) in 94 patients selected from the department database, with pathologic correlation (n=92) or two year imaging stability (n=10). Following an initial training session, all cases were independently reviewed by 10 breast imagers (MRI experience 2–11 years) for breast composition, background enhancement, mass enhancement (shape, margins, internal enhancement), NMLE (distribution, internal enhancement), kinetics, and final BI-RADs assessment.

There were 50 masses and 52 NMLEs. Statistical analysis for agreement beyond chance was computed using Cohen’s and Fleiss’ Kappa (ĸ). Fleiss definitions for level of agreement state that agreement is very good for ĸ of 0.81-1.0, good for 0.61-.80, moderate for 0.41-0.6, fair for 0.2-0.4, and poor for 0-0.2. Among the 10 readers, mean kappa agreement in tested categories was as follows: breast composition moderate (ĸ=0.45), background enhancement poor (ĸ=0.19), distinction of mass versus NMLE good (ĸ=0.71), mass shape moderate (ĸ=0.47), mass margins moderate (ĸ=0.51), mass internal enhancement moderate (ĸ=0.49), NMLE distribution fair (ĸ=0.39), NMLE internal enhancement fair (ĸ=0.29), kinetics fair (ĸ=0.38), and BI-RADS fair (ĸ=0.37). The 10 readers then re-reviewed all lesions with a final recommendation of biopsy or no biopsy, resulting again in fair agreement (ĸ=0.31).   

Amongst 10 independent reviewers we found good agreement for distinction of lesions as mass vs NMLE, smooth mass margins and irregular mass shape; and moderate agreement for breast composition, spiculated margins, rim enhancement, non-enhancing septations, and focal area of distribution. For all other BI-RADS lexicon modifiers agreement based on Kappa statistics was poor to fair.

As the ACR BI-RADS MR lexicon matures, studies such as this will help validate specific MR feature descriptors as optimal and identify others which may need modification.

Swami, A, Raza, S, Sivarajah, R, Chikarmane, S, Gautam, S, Birdwell, R, Interobserver Variability in Use of the BI-RADS Lexicon for Breast MRI.  Radiological Society of North America 2009 Scientific Assembly and Annual Meeting, November 29 - December 4, 2009 ,Chicago IL. http://archive.rsna.org/2009/8007494.html