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

Analysis and Design Process for Predicting and Controlling Blood Glucose in Type 1 Diabetic Patients

Engineering smart software that can monitor, predict, and control blood glucose is critical to improving patients' quality of treatments with type 1 Diabetic Mellitus (T1DM). However, ensuring a reasonable glycemic level in diabetic patients is quite challenging, as many methods do not adequately capture the complexities involved in glycemic control. This problem introduces a new level of complexity and uncertainty to the patient's psychological state, thereby making this problem nonlinear and unobservable. In this paper, we formulated a mathematical model using carbohydrate counting, insulin requirements, and the Harris-Benedict energy equations to establish the framework for predicting and controlling blood glucose level regulation in T1DM. We implemented the framework and evaluated its performance using root mean square error (RMSE) and mean absolute error (MAE) on a case study. Our framework had less error rate in terms of RMSE and MAE, which indicates a better fit with reasonable accuracy.

The challenge of predicting blood glucose concentration changes in patients with type I diabetes

Patients with Type I Diabetes (T1D) must take insulin injections to prevent the serious long term effects of hyperglycemia. They must also be careful not to inject too much insulin because this could induce (potentially fatal) hypoglycemia. Patients therefore follow a "regimen" that determines how much insulin to inject at each time, based on various measurements. We can produce an effective regimen if we can accurately predict a patient's future blood glucose (BG) values from his/her current features. This study explores the challenges of predicting future BG by applying a number of machine learning algorithms, as well as various data preprocessing variations (corresponding to 312 [learner, preprocessed-dataset] combinations), to a new T1D dataset that contains 29,601 entries from 47 different patients. Our most accurate predictor, a weighted ensemble of two Gaussian Process Regression models, achieved a (cross-validation) errL1 loss of 2.7 mmol/L (48.65 mg/dl). This result was unexpectedly poor given that one can obtain an errL1 of 2.9 mmol/L (52.43 mg/dl) using the naive approach of simply predicting the patient's average BG. These results suggest that the diabetes diary data that is typically collected may be insufficient to produce accurate BG prediction models; additional data may be necessary to build accurate BG prediction models over hours.

Stacked LSTM based deep recurrent neural network with kalman smoothing for blood glucose prediction

BACKGROUND

Blood glucose (BG) management is crucial for type-1 diabetes patients resulting in the necessity of reliable artificial pancreas or insulin infusion systems. In recent years, deep learning techniques have been utilized for a more accurate BG level prediction system. However, continuous glucose monitoring (CGM) readings are susceptible to sensor errors. As a result, inaccurate CGM readings would affect BG prediction and make it unreliable, even if the most optimal machine learning model is used.

METHODS

In this work, we propose a novel approach to predicting blood glucose level with a stacked Long short-term memory (LSTM) based deep recurrent neural network (RNN) model considering sensor fault. We use the Kalman smoothing technique for the correction of the inaccurate CGM readings due to sensor error.

RESULTS

For the OhioT1DM (2018) dataset, containing eight weeks' data from six different patients, we achieve an average RMSE of 6.45 and 17.24 mg/dl for 30 min and 60 min of prediction horizon (PH), respectively.

CONCLUSIONS

To the best of our knowledge, this is the leading average prediction accuracy for the ohioT1DM dataset. Different physiological information, e.g., Kalman smoothed CGM data, carbohydrates from the meal, bolus insulin, and cumulative step counts in a fixed time interval, are crafted to represent meaningful features used as input to the model. The goal of our approach is to lower the difference between the predicted CGM values and the fingerstick blood glucose readings-the ground truth. Our results indicate that the proposed approach is feasible for more reliable BG forecasting that might improve the performance of the artificial pancreas and insulin infusion system for T1D diabetes management.

CarbMetSim: A discrete-event simulator for carbohydrate metabolism in humans

This paper describes CarbMetSim, a discrete-event simulator that tracks the blood glucose level of a person in response to a timed sequence of diet and exercise activities. CarbMetSim implements broader aspects of carbohydrate metabolism in human beings with the objective of capturing the average impact of various diet/exercise activities on the blood glucose level. Key organs (stomach, intestine, portal vein, liver, kidney, muscles, adipose tissue, brain and heart) are implemented to the extent necessary to capture their impact on the production and consumption of glucose. Key metabolic pathways (glucose oxidation, glycolysis and gluconeogenesis) are accounted for in the operation of different organs. The impact of insulin and insulin resistance on the operation of various organs and pathways is captured in accordance with published research. CarbMetSim provides broad flexibility to configure the insulin production ability, the average flux along various metabolic pathways and the impact of insulin resistance on different aspects of carbohydrate metabolism. The simulator does not yet have a detailed implementation of protein and lipid metabolism. This paper contains a preliminary validation of the simulator's behavior. Significant additional validation is required before the simulator can be considered ready for use by people with Diabetes.

Development of a Deep Learning Model for Dynamic Forecasting of Blood Glucose Level for Type 2 Diabetes Mellitus: Secondary Analysis of a Randomized Controlled Trial

BACKGROUND

Type 2 diabetes mellitus (T2DM) is a major public health burden. Self-management of diabetes including maintaining a healthy lifestyle is essential for glycemic control and to prevent diabetes complications. Mobile-based health data can play an important role in the forecasting of blood glucose levels for lifestyle management and control of T2DM.

OBJECTIVE

The objective of this work was to dynamically forecast daily glucose levels in patients with T2DM based on their daily mobile health lifestyle data including diet, physical activity, weight, and glucose level from the day before.

METHODS

We used data from 10 T2DM patients who were overweight or obese in a behavioral lifestyle intervention using mobile tools for daily monitoring of diet, physical activity, weight, and blood glucose over 6 months. We developed a deep learning model based on long short-term memory-based recurrent neural networks to forecast the next-day glucose levels in individual patients. The neural network used several layers of computational nodes to model how mobile health data (food intake including consumed calories, fat, and carbohydrates; exercise; and weight) were progressing from one day to another from noisy data.

RESULTS

The model was validated based on a data set of 10 patients who had been monitored daily for over 6 months. The proposed deep learning model demonstrated considerable accuracy in predicting the next day glucose level based on Clark Error Grid and ±10% range of the actual values.

CONCLUSIONS

Using machine learning methodologies may leverage mobile health lifestyle data to develop effective individualized prediction plans for T2DM management. However, predicting future glucose levels is challenging as glucose level is determined by multiple factors. Future study with more rigorous study design is warranted to better predict future glucose levels for T2DM management.

The Progress of Glucose Monitoring-A Review of Invasive to Minimally and Non-Invasive Techniques, Devices and Sensors

Current glucose monitoring methods for the ever-increasing number of diabetic people around the world are invasive, painful, time-consuming, and a constant burden for the household budget. The non-invasive glucose monitoring technology overcomes these limitations, for which this topic is significantly being researched and represents an exciting and highly sought after market for many companies. This review aims to offer an up-to-date report on the leading technologies for non-invasive (NI) and minimally-invasive (MI) glucose monitoring sensors, devices currently available in the market, regulatory framework for accuracy assessment, new approaches currently under study by representative groups and developers, and algorithm types for signal enhancement and value prediction. The review also discusses the future trend of glucose detection by analyzing the usage of the different bands in the electromagnetic spectrum. The review concludes that the adoption and use of new technologies for glucose detection is unavoidable and closer to become a reality.

The Research of Improved Grey GM (1, 1) Model to Predict the Postprandial Glucose in Type 2 Diabetes

Diabetes may result in some complications and increase the risk of many serious health problems. The purpose of clinical treatment is to carefully manage the blood glucose concentration. If the blood glucose concentration is predicted, treatments can be taken in advance to reduce the harm to patients. For this purpose, an improved grey GM (1, 1) model is applied to predict blood glucose with a small amount of data, and especially in terms of improved smoothness it can get higher prediction accuracy. The original data of blood glucose of type 2 diabetes is acquired by CGMS. Then the prediction model is established. Finally, 50 cases of blood glucose from the Henan Province People's Hospital are predicted in 5, 10, 15, 20, 25, and 30 minutes, respectively, in advance to verify the prediction model. The prediction result of blood glucose is evaluated by the EGA, MSE, and MAE. Particularly, the prediction results of postprandial blood glucose are presented and analyzed. The result shows that the improved grey GM (1, 1) model has excellent performance in postprandial blood glucose prediction.

Assessment of type II diabetes mellitus using irregularly sampled measurements with missing data

Diabetes mellitus is one of the leading diseases in the developed world. In order to better regulate blood glucose in a diabetic patient, improved modelling of insulin-glucose dynamics is a key factor in the treatment of diabetes mellitus. In the current work, the insulin-glucose dynamics in type II diabetes mellitus can be modelled by using a stochastic nonlinear state-space model. Estimating the parameters of such a model is difficult as only a few blood glucose and insulin measurements per day are available in a non-clinical setting. Therefore, developing a predictive model of the blood glucose of a person with type II diabetes mellitus is important when the glucose and insulin concentrations are only available at irregular intervals. To overcome these difficulties, we resort to online sequential Monte Carlo (SMC) estimation of states and parameters of the state-space model for type II diabetic patients under various levels of randomly missing clinical data. Our results show that this method is efficient in monitoring and estimating the dynamics of the peripheral glucose, insulin and incretins concentration when 10, 25 and 50% of the simulated clinical data were randomly removed.

A clinical decision support system for femoral peripheral arterial disease treatment

One of the major challenges of providing reliable healthcare services is to diagnose and treat diseases in an accurate and timely manner. Recently, many researchers have successfully used artificial neural networks as a diagnostic assessment tool. In this study, the validation of such an assessment tool has been developed for treatment of the femoral peripheral arterial disease using a radial basis function neural network (RBFNN). A data set for training the RBFNN has been prepared by analyzing records of patients who had been treated by the thoracic and cardiovascular surgery clinic of a university hospital. The data set includes 186 patient records having 16 characteristic features associated with a binary treatment decision, namely, being a medical or a surgical one. K-means clustering algorithm has been used to determine the parameters of radial basis functions and the number of hidden nodes of the RBFNN is determined experimentally. For performance evaluation, the proposed RBFNN was compared to three different multilayer perceptron models having Pareto optimal hidden layer combinations using various performance indicators. Results of comparison indicate that the RBFNN can be used as an effective assessment tool for femoral peripheral arterial disease treatment.

Evaluation of a model for glycemic prediction in critically ill surgical patients

We evaluated a neural network model for prediction of glucose in critically ill trauma and post-operative cardiothoracic surgical patients. A prospective, feasibility trial evaluating a continuous glucose-monitoring device was performed. After institutional review board approval, clinical data from all consenting surgical intensive care unit patients were converted to an electronic format using novel software. This data was utilized to develop and train a neural network model for real-time prediction of serum glucose concentration implementing a prediction horizon of 75 minutes. Glycemic data from 19 patients were used to “train” the neural network model. Subsequent real-time simulated testing was performed in 5 patients to whom the neural network model was naive. Performance of the model was evaluated by calculating the mean absolute difference percent (MAD%), Clarke Error Grid Analysis, and calculation of the percent of hypoglycemic (≤70 mg/dL), normoglycemic (>70 and <150 mg/dL), and hyperglycemic (≥150 mg/dL) values accurately predicted by the model; 9,405 data points were analyzed. The models successfully predicted trends in glucose in the 5 test patients. Clark Error Grid Analysis indicated that 100.0% of predictions were clinically acceptable with 87.3% and 12.7% of predicted values falling within regions A and B of the error grid respectively. Overall model error (MAD%) was 9.0% with respect to actual continuous glucose modeling data. Our model successfully predicted 96.7% and 53.6% of the normo- and hyperglycemic values respectively. No hypoglycemic events occurred in these patients. Use of neural network models for real-time prediction of glucose in the surgical intensive care unit setting offers healthcare providers potentially useful information which could facilitate optimization of glycemic control, patient safety, and improved care. Similar models can be implemented across a wider scale of biomedical variables to offer real-time optimization, training, and adaptation that increase predictive accuracy and performance of therapies.

Analysis of intravenous glucose tolerance test data using parametric and nonparametric modeling: application to a population at risk for diabetes

BACKGROUND

Modeling studies of the insulin-glucose relationship have mainly utilized parametric models, most notably the minimal model (MM) of glucose disappearance. This article presents results from the comparative analysis of the parametric MM and a nonparametric Laguerre based Volterra Model (LVM) applied to the analysis of insulin modified (IM) intravenous glucose tolerance test (IVGTT) data from a clinical study of gestational diabetes mellitus (GDM).

METHODS

An IM IVGTT study was performed 8 to 10 weeks postpartum in 125 women who were diagnosed with GDM during their pregnancy [population at risk of developing diabetes (PRD)] and in 39 control women with normal pregnancies (control subjects). The measured plasma glucose and insulin from the IM IVGTT in each group were analyzed via a population analysis approach to estimate the insulin sensitivity parameter of the parametric MM. In the nonparametric LVM analysis, the glucose and insulin data were used to calculate the first-order kernel, from which a diagnostic scalar index representing the integrated effect of insulin on glucose was derived.

RESULTS

Both the parametric MM and nonparametric LVM describe the glucose concentration data in each group with good fidelity, with an improved measured versus predicted r² value for the LVM of 0.99 versus 0.97 for the MM analysis in the PRD. However, application of the respective diagnostic indices of the two methods does result in a different classification of 20% of the individuals in the PRD.

CONCLUSIONS

It was found that the data based nonparametric LVM revealed additional insights about the manner in which infused insulin affects blood glucose concentration.

Regulation of breast cancer tumorigenesis and metastasis by miRNAs

miRNAs are a family of 17- to 23-nucleotide noncoding small RNAs that primarily function as gene expression fine regulators. A number of studies have shown that miRNAs play an important role in breast tumorigenesis, metastasis, proliferation and differentiation of breast cancer stem cells. This short review summarizes the progression of miRNA-mediated breast tumorigenesis and metastasis through various signaling pathways associated with drug resistance.

An early warning system for hypoglycemic/hyperglycemic events based on fusion of adaptive prediction models

INTRODUCTION

Early warning of future hypoglycemic and hyperglycemic events can improve the safety of type 1 diabetes mellitus (T1DM) patients. The aim of this study is to design and evaluate a hypoglycemia/hyperglycemia early warning system (EWS) for T1DM patients under sensor-augmented pump (SAP) therapy.

METHODS

The EWS is based on the combination of data-driven online adaptive prediction models and a warning algorithm. Three modeling approaches have been investigated: (i) autoregressive (ARX) models, (ii) auto-regressive with an output correction module (cARX) models, and (iii) recurrent neural network (RNN) models. The warning algorithm performs postprocessing of the models' outputs and issues alerts if upcoming hypoglycemic/hyperglycemic events are detected. Fusion of the cARX and RNN models, due to their complementary prediction performances, resulted in the hybrid autoregressive with an output correction module/recurrent neural network (cARN)-based EWS.

RESULTS

The EWS was evaluated on 23 T1DM patients under SAP therapy. The ARX-based system achieved hypoglycemic (hyperglycemic) event prediction with median values of accuracy of 100.0% (100.0%), detection time of 10.0 (8.0) min, and daily false alarms of 0.7 (0.5). The respective values for the cARX-based system were 100.0% (100.0%), 17.5 (14.8) min, and 1.5 (1.3) and, for the RNN-based system, were 100.0% (92.0%), 8.4 (7.0) min, and 0.1 (0.2). The hybrid cARN-based EWS presented outperforming results with 100.0% (100.0%) prediction accuracy, detection 16.7 (14.7) min in advance, and 0.8 (0.8) daily false alarms.

CONCLUSION

Combined use of cARX and RNN models for the development of an EWS outperformed the single use of each model, achieving accurate and prompt event prediction with few false alarms, thus providing increased safety and comfort.

Multivariate prediction of subcutaneous glucose concentration in type 1 diabetes patients based on support vector regression

Data-driven techniques have recently drawn significant interest in the predictive modeling of subcutaneous (s.c.) glucose concentration in type 1 diabetes. In this study, the s.c. glucose prediction is treated as a multivariate regression problem, which is addressed using support vector regression (SVR). The proposed method is based on variables concerning: (i) the s.c. glucose profile, (ii) the plasma insulin concentration, (iii) the appearance of meal-derived glucose in the systemic circulation, and (iv) the energy expenditure during physical activities. Six cases corresponding to different combinations of the aforementioned variables are used to investigate the influence of the input on the daily glucose prediction. The proposed method is evaluated using a dataset of 27 patients in free-living conditions. 10-fold cross validation is applied to each dataset individually to both optimize and test the SVR model. In the case where all the input variables are considered, the average prediction errors are 5.21, 6.03, 7.14 and 7.62 mg/dl for 15, 30, 60 and 120 min prediction horizons, respectively. The results clearly indicate that the availability of multivariable data and their effective combination can significantly increase the accuracy of both short-term and long-term predictions.

Artificial neural networks for closed loop control of in silico and ad hoc type 1 diabetes

The closed loop control of blood glucose levels might help to reduce many short- and long-term complications of type 1 diabetes. Continuous glucose monitoring and insulin pump systems have facilitated the development of the artificial pancreas. In this paper, artificial neural networks are used for both the identification of patient dynamics and the glycaemic regulation. A subcutaneous glucose measuring system together with a Lispro insulin subcutaneous pump were used to gather clinical data for each patient undergoing treatment, and a corresponding in silico and ad hoc neural network model was derived for each patient to represent their particular glucose-insulin relationship. Based on this nonlinear neural network model, an ad hoc neural network controller was designed to close the feedback loop for glycaemic regulation of the in silico patient. Both the neural network model and the controller were tested for each patient under simulation, and the results obtained show a good performance during food intake and variable exercise conditions.

Real-time adaptive models for the personalized prediction of glycemic profile in type 1 diabetes patients

BACKGROUND

Prediction of glycemic profile is an important task for both early recognition of hypoglycemia and enhancement of the control algorithms for optimization of insulin infusion rate. Adaptive models for glucose prediction and recognition of hypoglycemia based on statistical and artificial intelligence techniques are presented.

METHODS

We compared an autoregressive (AR) model using only glucose information, an AR model with external insulin input (ARX), and an artificial neural network (ANN) using both glucose and insulin information. Online adaptive models were used to account for the intra- and inter-subject variability of the population with diabetes. The evaluation of the predictive ability included prediction horizons (PHs) of 30 min and 45 min.

RESULTS

The AR model presented root mean square error (RMSE) values of 14.0-21.6 mg/dL and correlation coefficients (CCs) of 0.92-0.95 for PH=30 min and 23.2-35.9 mg/dL and 0.79-0.87, respectively, for PH=45 min. The respective values for the ARX models were slightly better (PH=30 min, 13.3-18.8 mg/dL and 0.94-0.96; PH=45 min, 22.8-29.4 mg/dL and 0.83-0.88). For the ANN, the RMSE values ranged from 2.8 to 6.3 mg/dL, and the CC was 0.99 for all cases and PHs. The sensitivity of hypoglycemia prediction was 78% for AR, 81% for ARX, and 96% for ANN for PH=30 min and 65%, 67%, and 95%, respectively, for PH=45 min. The corresponding specificities were 96%, 96%, and 99% for PH=30 min and 93%, 93%, and 99% for PH=45 min.

CONCLUSIONS

The ANN appears to be more appropriate for the prediction of glucose profile based on glucose and insulin data.

Neural network-based real-time prediction of glucose in patients with insulin-dependent diabetes

BACKGROUND

Continuous glucose monitoring (CGM) technologies report measurements of interstitial glucose concentration every 5 min. CGM technologies have the potential to be utilized for prediction of prospective glucose concentrations with subsequent optimization of glycemic control. This article outlines a feed-forward neural network model (NNM) utilized for real-time prediction of glucose.

METHODS

A feed-forward NNM was designed for real-time prediction of glucose in patients with diabetes implementing a prediction horizon of 75 min. Inputs to the NNM included CGM values, insulin dosages, metered glucose values, nutritional intake, lifestyle, and emotional factors. Performance of the NNM was assessed in 10 patients not included in the model training set.

RESULTS

The NNM had a root mean squared error of 43.9 mg/dL and a mean absolute difference percentage of 22.1. The NNM routinely overestimates hypoglycemic extremes, which can be attributed to the limited number of hypoglycemic reactions in the model training set. The model predicts 88.6% of normal glucose concentrations (> 70 and < 180 mg/dL), 72.6% of hyperglycemia (≥ 180 mg/dL), and 2.1% of hypoglycemia (≤ 70 mg/dL). Clarke Error Grid Analysis of model predictions indicated that 92.3% of predictions could be regarded as clinically acceptable and not leading to adverse therapeutic direction. Of these predicted values, 62.3% and 30.0% were located within Zones A and B, respectively, of the error grid.

CONCLUSIONS

Real-time prediction of glucose via the proposed NNM may provide a means of intelligent therapeutic guidance and direction.

Blood Glucose Prediction Using Artificial Neural Networks Trained with the AIDA Diabetes Simulator: A Proof-of-Concept Pilot Study

Diabetes mellitus is a major, and increasing, global problem. However, it has been shown that, through good management of blood glucose levels (BGLs), the associated and costly complications can be reduced significantly. In this pilot study, Elman recurrent artificial neural networks (ANNs) were used to make BGL predictions based on a history of BGLs, meal intake, and insulin injections. Twenty-eight datasets (from a single case scenario) were compiled from the freeware mathematical diabetes simulator, AIDA. It was found that the most accurate predictions were made during the nocturnal period of the 24 hour daily cycle. The accuracy of the nocturnal predictions was measured as the root mean square error over five test days (RMSE5 day) not used during ANN training. For BGL predictions of up to 1 hour aRMSE5 dayof (±SD)0.15±0.04 mmol/L was observed. For BGL predictions up to 10 hours, aRMSE5  dayof (±SD)0.14±0.16 mmol/L was observed. Future research will investigate a wider range of AIDA case scenarios, real-patient data, and data relating to other factors influencing BGLs. ANN paradigms based on real-time recurrent learning will also be explored to accommodate dynamic physiology in diabetes.

PSECMAC intelligent insulin schedule for diabetic blood glucose management under nonmeal announcement

Therapeutically, the closed-loop blood glucose-insulin regulation paradigm via a controllable insulin pump offers a potential solution to the management of diabetes. However, the development of such a closed-loop regulatory system to date has been hampered by two main issues: 1) the limited knowledge on the complex human physiological process of glucose-insulin metabolism that prevents a precise modeling of the biological blood glucose control loop; and 2) the vast metabolic biodiversity of the diabetic population due to varying exogneous and endogenous disturbances such as food intake, exercise, stress, and hormonal factors, etc. In addition, current attempts of closed-loop glucose regulatory techniques generally require some form of prior meal announcement and this constitutes a severe limitation to the applicability of such systems. In this paper, we present a novel intelligent insulin schedule based on the pseudo self-evolving cerebellar model articulation controller (PSECMAC) associative learning memory model that emulates the healthy human insulin response to food ingestion. The proposed PSECMAC intelligent insulin schedule requires no prior meal announcement and delivers the necessary insulin dosage based only on the observed blood glucose fluctuations. Using a simulated healthy subject, the proposed PSECMAC insulin schedule is demonstrated to be able to accurately capture the complex human glucose-insulin dynamics and robustly addresses the intraperson metabolic variability. Subsequently, the PSECMAC intelligent insulin schedule is employed on a group of type-1 diabetic patients to regulate their impaired blood glucose levels. Preliminary simulation results are highly encouraging. The work reported in this paper represents a major paradigm shift in the management of diabetes where patient compliance is poor and the need for prior meal announcement under current treatment regimes poses a significant challenge to an active lifestyle.

Artificial neural network algorithm for online glucose prediction from continuous glucose monitoring

BACKGROUND AND AIMS

Continuous glucose monitoring (CGM) devices could be useful for real-time management of diabetes therapy. In particular, CGM information could be used in real time to predict future glucose levels in order to prevent hypo-/hyperglycemic events. This article proposes a new online method for predicting future glucose concentration levels from CGM data.

METHODS

The predictor is implemented with an artificial neural network model (NNM). The inputs of the NNM are the values provided by the CGM sensor during the preceding 20 min, while the output is the prediction of glucose concentration at the chosen prediction horizon (PH) time. The method performance is assessed using datasets from two different CGM systems (nine subjects using the Medtronic [Northridge, CA] Guardian and six subjects using the Abbott [Abbott Park, IL] Navigator. Three different PHs are used: 15, 30, and 45 min. The NNM accuracy has been estimated by using the root mean square error (RMSE) and prediction delay.

RESULTS

The RMSE is around 10, 18, and 27 mg/dL for 15, 30, and 45 min of PH, respectively. The prediction delay is around 4, 9, and 14 min for upward trends and 5, 15, and 26 min for downward trends, respectively. A comparison with a previously published technique, based on an autoregressive model (ARM), has been performed. The comparison shows that the proposed NNM is more accurate than the ARM, with no significant deterioration in the prediction delay.

CONCLUSIONS

The proposed NNM is a reliable solution for the online prediction of future glucose concentrations from CGM data.

A Nonlinear State Space Model for the Blood Glucose Metabolism of a Diabetic (Ein nichtlineares Zustandsraummodell für den Blutglukosemetabolismus eines Diabetikers)

The blood glucose metabolism of a diabetic is a complex nonlinear process closely linked to a number of internal factors which are not easily accessible to measurement. Based on accessible information – such as occasional blood glucose measurements and information about food intake and physical exercise – the system appears highly stochastic and the quantity of main interest, the blood glucose concentration, is very difficult to model and to predict. In this paper we describe a stochastic nonlinear state space model for modeling the blood glucose concentration of a diabetic patient. The model structure is based on physiological prior knowledge and the main nonlinearities are modeled using artificial neural networks. Offline training of the model is performed using a newly developed Monte-Carlo generalized EM (expectation maximization) algorithm. Online prediction is performed using particle filters. Our experimental results show that our approach provides better prediction results than a number of competing approaches.

Nonlinear modeling of the dynamic effects of infused insulin on glucose: comparison of compartmental with Volterra models

This paper presents the results of a computational study that compares simulated compartmental (differential equation) and Volterra models of the dynamic effects of insulin on blood glucose concentration in humans. In the first approach, we employ the widely accepted "minimal model" and an augmented form of it, which incorporates the effect of insulin secretion by the pancreas, in order to represent the actual closed-loop operating conditions of the system, and in the second modeling approach, we employ the general class of Volterra-type models that are estimated from input-output data. We demonstrate both the equivalence between the two approaches analytically and the feasibility of obtaining accurate Volterra models from insulin-glucose data generated from the compartmental models. The results corroborate the proposition that it may be preferable to obtain data-driven (i.e., inductive) models in a more general and realistic operating context, without resorting to the restrictive prior assumptions and simplifications regarding model structure and/or experimental protocols (e.g., glucose tolerance tests) that are necessary for the compartmental models proposed previously. These prior assumptions may lead to results that are improperly constrained or biased by preconceived (and possibly erroneous) notions-a risk that is avoided when we let the data guide the inductive selection of the appropriate model within the general class of Volterra-type models, as our simulation results suggest.

Data mining for blood glucose prediction and knowledge discovery in diabetic patients: the METABO diabetes modeling and management system

METABO is a diabetes monitoring and management system which aims at recording and interpreting patient's context, as well as, at providing decision support to both the patient and the doctor. The METABO system consists of (a) a Patient's Mobile Device (PMD), (b) different types of unobtrusive biosensors, (c) a Central Subsystem (CS) located remotely at the hospital and (d) the Control Panel (CP) from which physicians can follow-up their patients and gain also access to the CS. METABO provides a multi-parametric monitoring system which facilitates the efficient and systematic recording of dietary, physical activity, medication and medical information (continuous and discontinuous glucose measurements). Based on all recorded contextual information, data mining schemes that run in the PMD are responsible to model patients' metabolism, predict hypo/hyper-glycaemic events, and provide the patient with short and long-term alerts. In addition, all past and recently-recorded data are analyzed to extract patterns of behavior, discover new knowledge and provide explanations to the physician through the CP. Advanced tools in the CP allow the physician to prescribe personalized treatment plans and frequently quantify patient's adherence to treatment.

Neural network based glucose - insulin metabolism models for children with Type 1 diabetes

In this paper two models for the simulation of glucose-insulin metabolism of children with Type 1 diabetes are presented. The models are based on the combined use of Compartmental Models (CMs) and artificial Neural Networks (NNs). Data from children with Type 1 diabetes, stored in a database, have been used as input to the models. The data are taken from four children with Type 1 diabetes and contain information about glucose levels taken from continuous glucose monitoring system, insulin intake and food intake, along with corresponding time. The influences of taken insulin on plasma insulin concentration, as well as the effect of food intake on glucose input into the blood from the gut, are estimated from the CMs. The outputs of CMs, along with previous glucose measurements, are fed to a NN, which provides short-term prediction of glucose values. For comparative reasons two different NN architectures have been tested: a Feed-Forward NN (FFNN) trained with the back-propagation algorithm with adaptive learning rate and momentum, and a Recurrent NN (RNN), trained with the Real Time Recurrent Learning (RTRL) algorithm. The results indicate that the best prediction performance can be achieved by the use of RNN.

A real time simulation model of glucose-insulin metabolism for type 1 diabetes patients

In this paper, a simulation model of glucose-insulin metabolism for Type 1 diabetes patients is presented. The proposed system is based on the combination of Compartmental Models (CMs) and artificial Neural Networks (NNs). This model aims at the development of an accurate system, in order to assist Type 1 diabetes patients to handle their blood glucose profile and recognize dangerous metabolic states. Data from a Type 1 diabetes patient, stored in a database, have been used as input to the hybrid system. The data contain information about measured blood glucose levels, insulin intake, and description of food intake, along with the corresponding time. The data are passed to three separate CMs, which produce estimations about (i) the effect of Short Acting (SA) insulin intake on blood insulin concentration, (ii) the effect of Intermediate Acting (IA) insulin intake on blood insulin concentration, and (iii) the effect of carbohydrate intake on blood glucose absorption from the gut. The outputs of the three CMs are passed to a Recurrent NN (RNN) in order to predict subsequent blood glucose levels. The RNN is trained with the Real Time Recurrent Learning (RTRL) algorithm. The resulted blood glucose predictions are promising for the use of the proposed model for blood glucose level estimation for Type 1 diabetes patients.

Fuzzy-based controller for glucose regulation in type-1 diabetic patients by subcutaneous route

This paper presents an advisory/control algorithm for a type-1 diabetes mellitus (TIDM) patient under an intensive insulin treatment based on a multiple daily injections regimen (MDIR). The advisory/control algorithm incorporates expert knowledge about the treatment of this disease by using Mamdani-type fuzzy logic controllers to regulate the blood glucose level (BGL). The overall control strategy is based on a two-loop feedback strategy to overcome the variability in the glucose-insulin dynamics from patient to patient. An inner-loop provides the amount of both rapid/short and intermediate/long acting insulin (RSAI and ILAI) formulations that are programmed in a three-shots daily basis before meals. The combined preparation is then injected by the patient through a subcutaneous route. Meanwhile, an outer-loop adjusts the maximum amounts of insulin provided to the patient in a time-scale of days. The outer-loop controller aims to work as a supervisor of the inner-loop controller. Extensive closed-loop simulations are illustrated, using a detailed compartmental model of the insulin-glucose dynamics in a TIDM patient with meal intake.

Using the EM Algorithm to Train Neural Networks: Misconceptions and a New Algorithm for Multiclass Classification

The expectation-maximization (EM) algorithm has been of considerable interest in recent years as the basis for various algorithms in application areas of neural networks such as pattern recognition. However, there exists some misconceptions concerning its application to neural networks. In this paper, we clarify these misconceptions and consider how the EM algorithm can be adopted to train multilayer perceptron (MLP) and mixture of experts (ME) networks in applications to multiclass classification. We identify some situations where the application of the EM algorithm to train MLP networks may be of limited value and discuss some ways of handling the difficulties. For ME networks, it is reported in the literature that networks trained by the EM algorithm using iteratively reweighted least squares (IRLS) algorithm in the inner loop of the M-step, often performed poorly in multiclass classification. However, we found that the convergence of the IRLS algorithm is stable and that the log likelihood is monotonic increasing when a learning rate smaller than one is adopted. Also, we propose the use of an expectation-conditional maximization (ECM) algorithm to train ME networks. Its performance is demonstrated to be superior to the IRLS algorithm on some simulated and real data sets.

A neural network approach for insulin regime and dose adjustment in type 1 diabetes

BACKGROUND

A decision support system based on a neural network approach is proposed to advise on insulin regime and dose adjustment for type 1 diabetes patients.

METHOD

The system consists of two feed-forward neural networks, trained with the back-propagation algorithm with momentum and adaptive learning rate. The input to the system consists of patient's glucose levels, insulin intake, and observed hypoglycemia symptoms during a short time period. The output of the first neural network provides the insulin regime, which is applied as input to the second neural network to estimate the appropriate insulin doses for a short time period.

RESULTS

The system's ability in order to recommend on insulin regime is excellent, while its performance in adjusting the insulin dosages for a specific patient is highly dependent on the data set used during the training procedure.

CONCLUSIONS

Despite the limitations of computer-based approaches, this study shows that artificial neural networks can assist diabetes patients in insulin adjustment.