Neural Network Modeling Approaches for Patient Specific Glycemic Forecasting

Abstract

Prediction of glucose in patients with diabetes has been a major thrust of research in hopes to develop an artificial pancreas capable of automated and closed-loop glycemic control. Although it is less known, prediction of glucose in the critical care setting has also been the subject of considerable research endeavors. Successful prediction of glucose in both these patient populations requires the analysis of multiple factors and variables many of which are “patient specific” and vary from patient to patient. Thus, a well suited modeling technique for prediction of physiological glucose levels to needs to be adaptive and incorporate the effect of numerous dependent factors or variables to accurately forecast future glucose concentrations. Neural network models are a particularly well suited approach as they have the ability to learn and quantify the effect of various input factors/variables on a desired/predicted output. In this chapter, the application of neural network modeling to the prediction of glucose in diabetic and critical care patients that exhibit a lack of glucose control will be discussed, in addition to, the advantages of neural network modeling for patient specific glycemic forecasting with respect to other modeling techniques.