An expert system architecture and its application to the evaluation of scintigraphic image sequences

Abstract

A general framework for expert systems in the area of image understanding is described. It represents declarative knowledge in an associative network, allows the attachment of arbitrary procedural knowledge, and is used by a problem-independent control algorithm. This framework is used to implement a complete system for the diagnostic evaluation of scintigraphic images of the heart. The input to the system is an image sequence which is first preprocessed and segmented to obtain the left ventricle and its sectors. The contours of the ventricle and the sectors are the basis for knowledge-based analysis of the images. Analysis proceeds from the contour segments via objects and their motions to diagnostic descriptions of the image sequence. Experiments show that the system successfully completes analysis, and in particular does not make wrong diagnostic suggestions.<>