Neural Smithing: Supervised Learning In Feedforward Artificial Neural Networks EXCLUSIVE Download

Artificial neural networks are considered as “black boxes” for different problems such as pattern recognition and function approximation. The networks are trained or trained under supervision of the learning algorithm by using the given training data set.

The glucose target level, the GIR and the current blood glucose level are displayed on the monitor. The GIR control is performed by the closed-loop controller. HEGC is a simple method but it provides an accurate measurement of insulin resistance and muscle insulin sensitivity [12], [13]. The stochastic nature of glucose kinetics makes control a more complex task. Automatic control of the GIR during HEGC tests has been already proposed in [14] but the proposed model was not trained from a large scale data base. In the present work, we aim at developing an algorithm for real time automatic regulation of glucose levels during HEGC, by taking into account the stochastic nature of glucose kinetics and the nonlinear relationship between input parameters. In addition, the algorithm should be performed with the use of minimum blood samples required for evaluation of IR. In this study, we have used a multilayer perceptron neural network model for the controller.

3. Glucose control algorithm using artificial neural networks

Preliminaries

3.1. Glucose kinetics and glucose infusion rate

Glucose kinetics is a complex process because the nonlinear relationship between input parameters. In the present work we propose a controller based on a dynamic artificial neural network model, that predicts the effect of infusion rate on blood glucose level. If the network model accurately predicts the response, the controller can be shown to be self-consistent in that only one model can predict the response correctly.

3.2. Notation

3.3. Standard glucose clamp test

The standard glucose clamp test is described in detail by DeFronzo et al. [5]. The details are omitted here for the sake of brevity. The glucose infusion rate is calculated according to the equation:

G = G 0 - (G 0 + G c ) 1 + 1 / 2,

where G is the glucose infusion rate, and G c is the target blood glucose concentration.

The equation is usually solved by numerical iteration. The blood samples are collected from the tail vein and are measured using a glucose oxidase-based test kit. The response of the blood glucose level is monitored and recorded on a screen.

The results support the hypothesis that artificial neural networks model are a promising tool for real time automatic control of glucose levels during clamp experiments. The predictive control algorithm was implemented for a glucose pump controller. The controller was validated with use of an in-house developed glucose pump, and also by comparison with a commercial control system. The control system was found to be robust and flexible to changes in glucose pump, blood, and insulin infusion rate. 827ec27edc