Issues in automating cardiac SPECT diagnosis.

Abstract

Discusses the computational complexity involved in knowledge discovery when working with images. Data mining and general knowledge discovery techniques appear to be useful for classification of SPECT cardiac images. Creation and mining of the database has illustrated the importance of precision in data input, which is not always present in the narrative, or even in graphics-coded descriptions of images provided by physicians. Using data mining techniques on the raw images themselves may actually improve diagnosis and help clean the database. At the moment, these are time-consuming tasks, but methods are available to improve time requirements. As in all diagnostic systems, addition of more representative cases in each classification should improve performafice. Such techniques may also have application in the quality control of diagnostic laboratories.