Using Connectionist Approach for Finding Ideal Cut-Off Values for Quantitative Laboratory Tests

Abstract

A connectionist method is used in order to produce decision making limits for clinician from example patent data. A clinical patient material with symptoms and signs typical for Nephropathia epidemica (NE) is used as an example material. A single layer neural net was built using Minsky-Papert’s perceptron algorithm. Half of the material was used for learning and half for testing. The performance of the net was measured as a correctness rate for the classification of test material. A clear improvement was detected in the performance after the optimal cut-off points were searched using the connectionist method. We conclude that with connectionist algorithm we can imitate the learning process of a clinician and produce from example cases decision making limits which correlate more or less to those learned by clinician in their work.