Neural networks-based prediction of insulin resistance by means the homeostatic model assessment without the insulin concentration test

Abstract

Tissues insulin sensitivity has been estimated using the homeostasis model assessment. The insulin resistance is thus calculated from the plasma insulin and glucose concentrations. However, the insulin testing is an expensive test. Here, a computational approach based on neural networks for predicting the insulin resistance index through the homeostasis model assessment without considering the insulin testing results is proposed. A dataset of the prevalence study of metabolic syndrome (MS) is used to develop our approach. A total of 1919 subjects is used. The dataset if randomly split into a training set, a testing set, and a validating set for prediction approach performance assessment. Two of the neural networks commonly used in medical application are selected as functional predictors. The indexes obtained using the predictors are compared with the homeostasis model assessment-based index reported on the used dataset. From the comparison, one of neural networks-based approaches is considered the best predictor.