Computer-assisted diagnosis by a model-free system of direct data analysis

The basis of the method of data analysis presented is, in the case of any diagnostic test, the automatic compilation of separate frequency distributions for each diagnostic classification. The distinction of different test results for different diseases (the correlation for which the tests are used) can thus be quantitatively monitored. This offers opportunities for more specific control of the accuracy of the data base. Measurements of relative frequencies obtained from the frequency distributions of individuals with and without a given disease can serve as a quantitative handle for the selection of the combination of tests, and for adjustments of individual parameters, which will maximize the discrimination. The usual cutoffs are not used. A data-processing system can serve for the direct incorporation of patient chart data (including test results), and for the automation of the analysis described, with pattern recognition or cluster-seeking techniques. The ability of this system of analysis to minimize some of the problems associated with methods utilizing mathematical models is discussed.