A study on chronic glomerulonephritis discrimination of clinical urine analysis data using neural network

Chronic glomerulonephritis (CGN), which is caused by an inflammation of the kidney, can be diagnosed via urine, X-ray and biopsy. The objective of this paper was to present an method for the automatic discrimination of clinical test data of CGN patients via a urine analyzer, independently of an inspector’s visual inspection, using both statistical methods and a neural network. 10 parameters on a urine strip pad were utilized in order to diagnose disease during the urine test. The indicators of strip pads explain the meanings of the reacted color levels of urobilinogen, glucose, ketones, bilirubin, protein, nitrite, pH, occult blood, specific gravity and leukocytes. In order to detect CGN using the neural network, 4 parameters which are without statistic significance regarding CGN are removed via chi-square and t-tests. The input layer of the neural network consists of a reduced number of 6 parameters and the output layer is used to determine whether or not the given sample is dispositive for CGN. The learning process involved training with the data from 73 normal and 66 CGN patients. The validation set consisted of 50 normals and 50 patients, and was not used in the training set. The neural network was able to discriminate correctly in 96 percent of the normal patients and 94 percent of the CGN patients. Tests of statistic significance and the application of such a neural network appear to be quite effective with regard to the discrimination of CGN.