Interobserver Variation in the Interpretation of Breast Imaging

Abstract

Purpose: To analyze interobserver agreement in the interpretation of palpable noncalcified breast masses by means of mammography, ultrasonography, and a combination of both methods.

Material and Methods: Mammograms and ultrasonograms of 100 benign breast masses and 100 malignant ones in 200 patients were retrospectively analyzed by 4 radiologists with experience in both imaging modalities. The images were analyzed independently and without knowledge of the final diagnosis. The analysis was performed in 3 sessions and used a 5-point rating scale for probability of malignancy. The interobserver variation was analyzed by means of observed agreement, kappa, and weighted kappa statistics based on the 5-point rating scale and a 3-level scale of the collapsed 5-point scale. The chi-square statistic was used for testing the equality of the kappa values.

Results: The overall kappa value on the 3-level scale was 0.48 (range 0.37-0.61) for ultrasonography, 0.58 (range 0.52-0.66) for mammography, and 0.71 (range 0.63-0.79) for the combined reading. The kappa values were statistically different for ultrasonography but did not differ significantly for the mammographic and combined readings. The combined reading showed higher kappa values than mammography alone, and the improvement was statistically significant for 3 of the 6 pairs of comparison.

Conclusion: Radiologists differ substantially in their interpretations of mammograms and breast ultrasonograms. Agreement was highest in the combined reading, intermediate in mammography, and lowest in ultrasonography.