Assessment of Liver Function Using Hybrid Neuro-Fuzzy Model of Blood Albumin

Abstract

This paper presents an assessment of liver function using novel neuro-fuzzy model of blood albumin level (BA). The developed model that is used to predict the BA consists of four inputs: Asparate Aminotransferace (AST), Alkaline Phosphate (ALP), Total Bilirubin (T. Bil.) and Total Protein (T. Prot.), which are measured in any routine liver function test. The proposed BA model was trained using 211 measured data and a root-mean square error (RMSE) of 0.29 for 100 epochs was achieved. The performance of the developed BA model was validated using 57 testing data sets and RMSE of 0.34 for 100 epochs was achieved. The correlation coefficient (CC) between the predicted and measured values of blood albumin is statistically significant (CC=0.83), which ensures the efficiency and accuracy of developed fuzzy model for predicting BA. The main clinical benefit of this model is that it improves the assessment capabilities of liver diseases and can be used as an integral part of any medical expert system denoted for assessment and diagnosis of liver disorders.