A Fuzzy Model for the Risk of Urinary Tract Infections Prediction Using Microscopic and Chemical Urine Analysis

Abstract

Urinary Tract Infections (UTIs) are the most common type of bacterial infection and the physical, chemical and microscopic analysis of urine, followed by culture is a commonly conducted procedure in microbiological laboratories. Even though physical, chemical and microscopic analysis are more time efficient, culture is still considered the gold standard in diagnosing UTI, although its results and the results of antibiogram are obtained within a period of days. The aim of this research is to create a fuzzy model to predict the risk of urinary tract infections using microscopic and chemical urine analysis. A dataset containing 595 samples was used to create a fuzzy model using a Mamdani-type inference system with 5 inputs (nitrite, leukocyte esterase, bacteria, white blood cells and red blood cells) and 1 output (prediction of infection). The accuracy of the model achieved was 86%, after the contaminated samples were excluded. The recall value for predicting the absence of UTI samples was the highest, 89%, while the recall value for predicting the presence of UTI was 71%. The lowest recall achieved was for samples that could present both, the absence and the presence of UTI, with a value of 29.6%. The model was successful and revealed adequate accuracy, however future experiments should use more parameters.