Elicitation and Representation of Expert Knowledge for Computer Aided Diagnosis in Mammography

Abstract

Summary Objectives: To study how professional radiologists describe, interpret and make decisions about micro-calcifications in mammograms. The purpose was to develop a model of the radiologists’ decision making for use in CADMIUM II, a computerized aid for mammogram interpretation that combines symbolic reasoning with image processing. Methods: In Study 1, eleven radiologists were asked to ‘think out loud’ as they interpreted 20 sets of calcifications. Participants used 159 terms to describe calcifications. We used these data to design a scheme with 50 descriptors. In Study 2, ten radiologists used the scheme to describe 40 sets of calcifications. We assessed the capacity of the terms to discriminate between benign and malignant calcifications, testing them against radiologists’ assessments of malignancy and follow-up data. Results: The descriptors that were found to be the most discriminating in Study 2 were included in CADIMUM II’s knowledge base. They were represented as arguments for either a benign or a malignant diagnosis. These arguments are the central component of the decision support provided by the system. Other components are: image processing algorithms for the detection and measurement of calcifications and a set of rules that use the measures to decide which of the arguments apply to a given set of calcifications. Conclusions: Preliminary evaluations of the CADMIUM II prototype reinforce the value of representing explicitly decision making processes in computer aided mammography and of deriving these processes from image processing measurements. Decision support is presented here at a level of description that is both relevant and meaningful to the user.