Reproducibility and consistency in classification of breast parenchymal patterns.

Abstract

The Wolfe Classification System of evaluating risk of developing breast cancer from analysis of mammographic parenchymal patterns (MPP) has received worldwide attention. In addition to confirming its validity, it is necessary to establish the ability of radiologists to apply the classification of mammograms consistently and reproducibly. This paper reports the results of 12 radiologists associated with the former Breast Cancer Detection Demonstration Projects (BCDDP) in Ann Arbor, Honolulu, Seattle, and Tucson. The participating radiologists had all had some experience with classifying mammograms, then were exposed to more intensive instruction, and finally worked with atlases developed at each institution. The results reported here show the comparisons of the radiologists' readings with Dr. Wolfe in a series of three separate exercises and also three tests of intraobserver consistency. Two of the participating institutions used xeromammograms and two, film mammograms. The study was designed and supervised by an independent data coordinating center with masking of mammograms, without benefit of information of previous readings and no information available except age. The results indicate that this group of radiologists could classify mammograms in the method of Wolfe at an acceptable level compared to other similar exercises.