Neural-network models for the blood glucose metabolism of a diabetic

Abstract

We study the application of neural networks to modeling the blood glucose metabolism of a diabetic. In particular we consider recurrent neural networks and time series convolution neural networks which we compare to linear models and to nonlinear compartment models. We include a linear error model to take into account the uncertainty in the system and for handling missing blood glucose observations. Our results indicate that best performance can be achieved by the combination of the recurrent neural network and the linear error model.

Artificial Neural Network Algorithm for Online Glucose Prediction from Continuous Glucose Monitoring

“…Volker Tresp et al [6] also presented a hybrid CM-NNM system. The CM simulates the delay in carbohydrate intake absorption, the physical exercise and the administered insulin dynamics.…”

“…All of the predictors referred to the use of self-CBGM [4][5][6] and required auxiliary details like insulin and intake data, hypoglycemic events and exercise among others. The availability of CGM opens a new scenario: glucose measurement is continuous information and any method suitable for time series analysis could be explored in the attempt to find a solution for the prediction problem [9].…”

Glucose Prediction in Type 1 and Type 2 Diabetic Patients Using Data Driven Techniques

“…During the last decade, the application of machine learning methods in predictive modelling of glucose concentration in patients with diabetes has gained much attention. Simple feed forward neural networks [Kok, 2004;Zitar & Al-Labali, 2005;Quchani & Tahami, 2007;Baghdadi & Nasrabadi, 2007] as well as more sophisticated types such as recurrent [Tresp et al, 1999;Mougiakakou et al, 2006] and wavelet neural networks [Zainuddin et al, 2009] have been utilised up to now for the prediction of the glucose concentration in diabetic patients. The results obtained in these works show that reasonably accurate glucose predictions can be made; however, a direct comparison between them is not feasible since they refer to different prediction horizons.…”

“…The only difference between the above mentioned works concerns the feature selection technique and the neural network that is employed. A number of prediction models specific to type 1 diabetes, including non-linear compartmental models, time series convolution neural networks and RNNs, were compared in [Tresp et al, 1999]. The combination of the RNNs with a linear error model gave the best results deriving a root mean squared error (RMSE) of 51 mg/dl.…”

“…and glucose levels is nonlinear, dynamic, interactive and patient-specific [Tresp et al, 1999], necessitates the application of non-linear regression models such as artificial neural networks, support vector regression and Gaussian processes. Different types of neural networks have been considered in modelling the blood glucose metabolism, such as multilayer perceptron (MLP) [Kok, 2004;Zitar & Al-Labali, 2005;Quchani & Tahami, 2007], radial basis function (RBF) [Baghdadi & Nasrabadi, 2007], wavelet [Zainuddin et al, 2009], time series convolution [Tresp et al, 1999] and recurrent neural networks (RNN) [Tresp et al, 1999;Mougiakakou et al, 2006]. Additionally, Gaussian processes have derived prominent results regarding glucose prediction [Valletta et al, 2009].…”