Chernetsov

The research was a part of the project aimed at developing an automated optimal insulin dose forecasting system, which, when combined with an insulin pump and continuous glucose monitoring system, could serve as an artificial pancreas. Paper investigates the effectiveness of neural network algorithms for predicting the blood glucose values in type I diabetes patients. The aim is to choose the optimal neural network and its training algorithm for use in predicting optimal doses of insulin. This research was performed using a software system MatLab.

Optimal doses of artificial insulin depend on a variety of factors. Adjustment of insulin dosages is a sophisticated task and can be far too complex for most patients. In order to simplify the solution to the problem the Continuous Glucose Monitoring System (CGMS) and continuous subcutaneous insulin infusion systems (insulin pumps) were developed.Blood glucose level (BG) control systems based on CGMS and insulin pumps are being actively developed. Algorithmically, these systems consist of two subsystems: the BG forecasting subsystem and the optimal dose determination subsystem. The problem of blood glucose forecasting in type I diabetes patients is being researched here. Artificial neural network (ANN)-based and maximum similarity extrapolation model based on maximum likeness set (EMMLS)-based approaches have been considered. The approach effectiveness has been compared. It is shown here, that neural networks provide more accurate results for short-term forecasting, while EMMLS is more accurate on long-term forecasting. Prospects of combined EMMLS and ANN model are also shown.

Optimal doses of artificial insulin (further – insulin) depend on a variety of factors. Adjustment of insulin dosages is a sophisticated task and can be far too complex for most patients. In order to simplify the solution to the problem the Continuous Glucose Monitoring System (CGMS) and continuous subcutaneous insulin infusion systems (insulin pumps) were developed.Blood glucose level (BG) control systems based on CGMS and insulin pumps are being actively developed. Algorithmically, these systems consist of two subsystems: the BG forecasting subsystem and the optimal dose determination subsystem. In the long run this research aims to synthesize the former of the two subsystemsThe problem of BG prediction of patients with insulin-dependent diabetes is considered in this paper. An approach to solving this problem by using adaptive neuro-fuzzy inference system ANFIS is proposed.  The results of research of efficiency of the method are presented. It was shown that ANFIS provides us with high quality of prediction when used on relatively short periods of time.

Method for dynamical systems identification using self-organizing neural networks is considered. Using method VQTAM was constructed self-organizing neural model for blood glucose level of 1-st type diabetic patients. Set of experimental researches for constructed model is performed y is estimated its accuracy

English language publications covering optimal insulin doses forecasting methods for insulin-dependant diabetes patients are presented in this review. Mathematical models of insulin dynamics in human body, neural network-based and composite algorithms and optimal insulin dose forecasting systems are considered.

The paper discusses a class of problems with two-level tree information structure graph with leaves’ computational complexity a priori unknown. Using LSF component of a well-known GRID-system Globus for distributed heterogeneous computing system load balancing while parallelizing problems of the specified class is described.