Towards modeling human body responsiveness to glucose intake and insulin injection based on artificial neural networks

Abstract

Diabetes is one of the most widespread diseases around the world, especially in the western world where non-healthy and fast foods are widely used. Many types of research have been conducted for developing methods for predicting, diagnosing and treating diabetes. One of the tools used for this purpose is mathematical modelling, which is used for developing models of blood glucose and insulin intake. In this paper, a model to determine the proper insulin dose for diabetic inpatients was implemented using Artificial Neural Network (ANN). The model is developed by taking into consideration ten different parameters (Patient's Gender, Patient's Age, Body Mass Index for Patient, Disease History, Total Daily Insulin Doses, Diabetes Type, Smoking Factor, Genetic Factor, Creatinine Clearance and Accumulative Glucose), in addition to real-time blood glucose readings. The model is developed based on a dataset from 159 inpatients from three different hospitals. It was found that the model with the best performance was based on one hidden layer with six neurons and seven inputs. The significant inputs were glucose readouts, glucose difference, normal range, accumulative glucose, history of the disease, total insulin dose and the patient's gender. The MSE of the best model was 5.413 and the correlation was 0.9315 with negligible training time.