[Computer-assisted laboratory diagnosis--principles and clinical application].

Abstract

Methods of computer-assisted diagnosis based on laboratory data are divided into so called rule-base system in which the diagnostic knowledge of physicians is programmed and those using multivariate analysis and mathematical models such as fuzzy interference. In this report, the principles and the technique of matrix discrimination method, which belongs to the latter group, are described, and its usefulness in clinical diagnosis evaluated. By the present method, subjects (diseases) are separated from two or more groups of subjects (diseases) to be discriminated on the basis of laboratory data. The method, based on the linear discrimination analysis, is characterized by optimization of selection of subjects to be discriminated and selection of test items and inference by the use of a discrimination curve. When this method was applied to discrimination of healthy individuals and patients with various liver and biliary disorders (11 diseases), all healthy individuals were discriminated from the patients. A mean of 97.0% of patients with hepatic parenchymal disorders were found in the top two, and 86.2% of those with space-occupying diseases in the top three, of the 12 diseases from which the disease of each patient was to be estimated. This diagnostic ability of the matrix discrimination method far exceeded the physicians' expectations.