Comparative assessment of glucose prediction models for patients with type 1 diabetes mellitus applying sensors for glucose and physical activity monitoring

Abstract

The present work presents the comparative assessment of four glucose prediction models for patients with type 1 diabetes mellitus (T1DM) using data from sensors monitoring blood glucose concentration. The four models are based on a feedforward neural network (FNN), a self-organizing map (SOM), a neuro-fuzzy network with wavelets as activation functions (WFNN), and a linear regression model (LRM), respectively. For the development and evaluation of the models, data from 10 patients with T1DM for a 6-day observation period have been used. The models' predictive performance is evaluated considering a 30-, 60- and 120-min prediction horizon, using both mathematical and clinical criteria. Furthermore, the addition of input data from sensors monitoring physical activity is considered and its effect on the models' predictive performance is investigated. The continuous glucose-error grid analysis indicates that the models' predictive performance benefits mainly in the hypoglycemic range when additional information related to physical activity is fed into the models. The obtained results demonstrate the superiority of SOM over FNN, WFNN, and LRM with SOM leading to better predictive performance in terms of both mathematical and clinical evaluation criteria.

GluNet: A Deep Learning Framework for Accurate Glucose Forecasting

“…There are a few works to date using DL for glucose forecasting. For instance, in [10], [16] traditional dense neural networks were used but without exploiting the advantages of deep layers. In [17], two latent layer neural networks were adopted for hypo/hyperglycemia prediction.…”

Comparative assessment of statistical and machine learning techniques towards estimating the risk of developing type 2 diabetes and cardiovascular complications

“…ANNs have been used in a great number of diagnostic decision support systems for medical applications, and they have demonstrated good predictive power (Buller, Buller, Innocent, & Pawlak, 1996;Verma & Zakos, 2001;Zarkogianni, Vazeou, Mougiakakou, Prountzou, & Nikita, 2011;Zarkogianni et al, 2015b). Ensembles of ANNs (EANN) can improve both the generalization abilities and the performance of an individual ANN, by compensating with each other the errors produced by each ANN (Sharkey, 1996).…”

Continuous Glucose Monitoring Sensors: Past, Present and Future Algorithmic Challenges

“…It is worth noting that more sophisticated prediction algorithms, also exploiting other signals like the amount of insulin injected or physical activity, can be employed, e.g., those of Zhao et al [9,12], Zecchin et al [13,41], Turksoy et al [10], Zarkogianni et al [11], and Georga et al [42,43]. …”

“…From a clinical point of view, it has been widely demonstrated that the additional information provided by CGM sensors, when used in conjunction with SMBG data, improves the quality of glucose control [7,8]. From an academic point of view, the availability of CGM data stimulated, over the last 15 years, the development of several CGM-based applications, e.g., algorithms for the prediction of future glucose concentration to generate preventive hypo/hyperglycemic alerts [9,10,11,12,13], for the real-time modulation of the basal insulin administration [14,15,16], and for the detection of faults with glucose sensor–insulin pumps system [17,18,19,20,21]. Even more interesting is that CGM sensors enabled the realization of the artificial pancreas (AP), i.e., a device designed mainly for Type 1 diabetes (T1D), which is aimed at maintaining the BG concentration within the safety range by automatically injecting insulin via an insulin pump controlled by a closed-loop control algorithm [22,23,24,25].…”