Reproducibility of mammographic classifications.

Wolfe's mammographic classification and a percentage classification are statistically evaluated for inter- and intraobserver bias and agreement by seven mammographers with a set of 200 xeromammograms. The results demonstrate significant bias and disagreement with both methods, raising questions about the clinical limitations of these or other mammographic classifications. However, about 90% of the percentage classifications of pairs of readers are within adjacent categories. This suggests that (1) more experience with precisely defined classifications and protocols, (2) the development and application of readily available instructional materials, and (3) studies to identify and evaluate sources of variation in such classifications may eventually lead to acceptable levels of reproducibility.