Evaluation of a Decision Aid for the Classification of Microcalcifictions

The aim of this work is to provide radiologists with a computerised decision aid. We use the term ‘decision aid’ to indicate a system which makes available small amounts of relevant information on the basis of patient data. In a previous paper we reviewed a variety of approaches being used in systems to assist radiological decision making, approaches based on different techniques for the representation of the relevant information [1]. Some systems use image processing to extract information from images and present it to radiologists. Few such systems are in routine use, perhaps because the information provided is not related to the decisions radiologists are trying to take. Some systems use symbolic representations of radiological expertise to reason about facts supplied by the radiologists, these systems are problematic because it is difficult for radiologists to put information about images into a symbolic form. The approach taken in our work has been to develop a decision aid which combines the two forms of information: image processing is used to provide the descriptions which knowledge-based systems require and symbolic representations to relate image data to radiological decisions.