A Review of Emerging Technologies for the Management of Diabetes Mellitus

Artificial Intelligence and Machine Learning for Improving Glycemic Control in Diabetes: Best Practices, Pitfalls, and Opportunities

OBJECTIVE

Artificial intelligence and machine learning are transforming many fields including medicine. In diabetes, robust biosensing technologies and automated insulin delivery therapies have created a substantial opportunity to improve health. While the number of manuscripts addressing the topic of applying machine learning to diabetes has grown in recent years, there has been a lack of consistency in the methods, metrics, and data used to train and evaluate these algorithms. This manuscript provides consensus guidelines for machine learning practitioners in the field of diabetes, including best practice recommended approaches and warnings about pitfalls to avoid.

METHODS

Algorithmic approaches are reviewed and benefits of different algorithms are discussed including importance of clinical accuracy, explainability, interpretability, and personalization. We review the most common features used in machine learning applications in diabetes glucose control and provide an open-source library of functions for calculating features, as well as a framework for specifying data sets using data sheets. A review of current data sets available for training algorithms is provided as well as an online repository of data sources.

SIGNIFICANCE

These consensus guidelines are designed to improve performance and translatability of new machine learning algorithms developed in the field of diabetes for engineers and data scientists.

Breath-based biosensors and system development for noninvasive detection of diabetes: A review

BACKGROUND AND AIMS

In recent years, noninvasive techniques are becoming conspicuous for diabetes detection. Sweat, tear, saliva, urine and breath-based methods showing prominent results in breath acetone detection which is considered as a biomarker of diabetes. A concrete relationship between breath acetone and BG helps in the development of devices for diabetes detection.

METHODS

The primary source for this study includes scholarly publications that primarily focus on the development of biosensors and systems for diabetes detection using acetone present in breath. Articles were analysed to examine various types of biosensors with their sensing materials to provide acetone detection limits. Recent noninvasive systems and products have been investigated and determine the relationship between breath acetone and BG levels.

RESULTS

Breath-based biosensor technologies are capable for diabetes detection. The acetone biosensor detection ranges from 100 ppb to 100 ppm, and it can applicable from room temperature to 400 °C. In healthy volunteers, acetone level ranges from 0.32 to 2.19 ppm, while patients with diabetes exhibit a wider range of 0.22-21 ppm depending on the biosensor, detection method, and clinical circumstances of patients and lab conditions.

CONCLUSION

This manuscript presents an extensive analysis of breath-based biosensors and their potential for detection of diabetes. Acetone detection methods are promising but unable to provide concrete correlation between breath acetone and blood glucose levels. The present study motivates the continued research and development of biosensors, and electronic devices to provide linear relationship of breath acetone and BG for noninvasive diabetes detection applications.

RFFE - Random Forest Fuzzy Entropy for the classification of Diabetes Mellitus

Diabetes is a category of metabolic disease commonly known as a chronic illness. It causes the body to generate less insulin and raises blood sugar levels, leading to various issues and disrupting the functioning of organs, including the retinal, kidney and nerves. To prevent this, people with chronic illnesses require lifetime access to treatment. As a result, early diabetes detection is essential and might save many lives. Diagnosis of people at high risk of developing diabetes is utilized for preventing the disease in various aspects. This article presents a chronic illness prediction prototype based on a person's risk feature data to provide an early prediction for diabetes with Fuzzy Entropy random vectors that regulate the development of each tree in the Random Forest. The proposed prototype consists of data imputation, data sampling, feature selection, and various techniques to predict the disease, such as Fuzzy Entropy, Synthetic Minority Oversampling Technique (SMOTE), Convolutional Neural Network (CNN) with Stochastic Gradient Descent with Momentum (SGDM), Support Vector Machines (SVM), Classification and Regression Tree (CART), K-Nearest Neighbor (KNN), and Naïve Bayes (NB). This study uses the existing Pima Indian Diabetes (PID) dataset for diabetic disease prediction. The predictions' true/false positive/negative rate is investigated using the confusion matrix and the receiver operating characteristic area under the curve (ROCAUC). Findings on a PID dataset are compared with machine learning algorithms revealing that the proposed Random Forest Fuzzy Entropy (RFFE) is a valuable approach for diabetes prediction, with an accuracy of 98 percent.

Multiple color representation and fusion for diabetes mellitus diagnosis based on back tongue images

Tongue images have been proved to be effective in diabetes mellitus (DM) diagnosis. Without requirement of collecting blood sample, tongue image based diagnosis approach is non-invasive and convenient for the patients. Meanwhile, the colors of tongues play an important in aiding accurate diagnosis. However, the tongues' colors fall on a small color gamut that makes it difficult for the existing color descripts to identify and distinguish the tiny difference of the tongues. To tackle this problem, we introduce a novel color descriptor by representing the colors with the clustering centers, namely color centroid points, of the color points sampled from tongue images. In order to boost the capacity of the descriptor, we extend it into three color spaces, i.e., RGB, HSV and LAB to mine a rich set of color information and exploit the complementary information among the three spaces. Since there exist correlation and complementarity among the features extracted from the three color spaces, we propose a novel multiple color features fusion method for DM diagnosis. Particularly, two projections are learned to project the multiple features to their corresponding shared and specific subspaces, in which their similarity and diversity are firstly measured by the Euclidean Distance and Hilbert Schmidt Independence Criterion (HSIC), respectively. To fully exploit the similar and complementary information, the two components are jointly transformed to their label vector, efficiently embedding the discriminant prior into the model, leading to significant improvement in the diagnosis outcomes. Experimental results on clinical tongue dataset substantiated the effectiveness of our proposed clustering-based color descriptor and the proposed multiple colors fusion approach. Overall, the proposed pipeline for the diagnosis of DM using back tongue images, achieved an average accuracy of up to 93.38%, indicating its potential toward realization of a clinical diagnostic tool for DM. Without loss generality, we also assessed the performance of the novel multiple features fusion method on two public datasets. The experiments prove the superiority of our multiple features learning model on general real-life application.

Effects of a Novel Blood Glucose Forecasting Feature on Glycemic Management and Logging in Adults With Type 2 Diabetes Using One Drop: Retrospective Cohort Study

Background

Personalized feedback is an effective behavior change technique frequently incorporated into mobile health (mHealth) apps. Innovations in data science create opportunities for leveraging the wealth of user data accumulated by mHealth apps to generate personalized health forecasts. One Drop’s digital program is one of the first to implement blood glucose forecasts for people with type 2 diabetes. The impact of these forecasts on behavior and glycemic management has not been evaluated to date.

Objective

This study sought to evaluate the impact of exposure to blood glucose forecasts on blood glucose logging behavior, average blood glucose, and percentage of glucose points in range.

Methods

This retrospective cohort study examined people with type 2 diabetes who first began using One Drop to record their blood glucose between 2019 and 2021. Cohorts included those who received blood glucose forecasts and those who did not receive forecasts. The cohorts were compared to evaluate the effect of exposure to blood glucose forecasts on logging activity, average glucose, and percentage of glucose readings in range, after controlling for potential confounding factors. Data were analyzed using analysis of covariance (ANCOVA) and regression analyses.

Results

Data from a total of 1411 One Drop users with type 2 diabetes and elevated baseline glucose were analyzed. Participants (60.6% male, 795/1311; mean age 50.2 years, SD 11.8) had diabetes for 7.1 years on average (SD 7.9). After controlling for potential confounding factors, blood glucose forecasts were associated with more frequent blood glucose logging (P=.004), lower average blood glucose (P<.001), and a higher percentage of readings in range (P=.03) after 12 weeks. Blood glucose logging partially mediated the relationship between exposure to forecasts and average glucose.

Conclusions

Individuals who received blood glucose forecasts had significantly lower average glucose, with a greater amount of glucose measurements in a healthy range after 12 weeks compared to those who did not receive forecasts. Glucose logging was identified as a partial mediator of the relationship between forecast exposure and week-12 average glucose, highlighting a potential mechanism through which glucose forecasts exert their effect. When administered as a part of a comprehensive mHealth program, blood glucose forecasts may significantly improve glycemic management among people living with type 2 diabetes.

A Multimodal Approach for Real Time Recognition of Engagement towards Adaptive Serious Games for Health

In this article, an unobtrusive and affordable sensor-based multimodal approach for real time recognition of engagement in serious games (SGs) for health is presented. This approach aims to achieve individualization in SGs that promote self-health management. The feasibility of the proposed approach was investigated by designing and implementing an experimental process focusing on real time recognition of engagement. Twenty-six participants were recruited and engaged in sessions with a SG that promotes food and nutrition literacy. Data were collected during play from a heart rate sensor, a smart chair, and in-game metrics. Perceived engagement, as an approximation to the ground truth, was annotated continuously by participants. An additional group of six participants were recruited for smart chair calibration purposes. The analysis was conducted in two directions, firstly investigating associations between identified sitting postures and perceived engagement, and secondly evaluating the predictive capacity of features extracted from the multitude of sources towards the ground truth. The results demonstrate significant associations and predictive capacity from all investigated sources, with a multimodal feature combination displaying superiority over unimodal features. These results advocate for the feasibility of real time recognition of engagement in adaptive serious games for health by using the presented approach.

Multi-step ahead predictive model for blood glucose concentrations of type-1 diabetic patients

Continuous monitoring of blood glucose (BG) levels is a key aspect of diabetes management. Patients with Type-1 diabetes (T1D) require an effective tool to monitor these levels in order to make appropriate decisions regarding insulin administration and food intake to keep BG levels in target range. Effectively and accurately predicting future BG levels at multi-time steps ahead benefits a patient with diabetes by helping them decrease the risks of extremes in BG including hypo- and hyperglycemia. In this study, we present a novel multi-component deep learning model BG-Predict that predicts the BG levels in a multi-step look ahead fashion. The model is evaluated both quantitatively and qualitatively on actual blood glucose data for 97 patients. For the prediction horizon (PH) of 30 mins, the average values for root mean squared error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAPE), and normalized mean squared error (NRMSE) are [Formula: see text] mg/dL, 16.77 ± 4.87 mg/dL, [Formula: see text] and [Formula: see text] respectively. When Clarke and Parkes error grid analyses were performed comparing predicted BG with actual BG, the results showed average percentage of points in Zone A of [Formula: see text] and [Formula: see text] respectively. We offer this tool as a mechanism to enhance the predictive capabilities of algorithms for patients with T1D.

Penalty weighted glucose prediction models could lead to better clinically usage

BACKGROUND AND OBJECTIVE

Numerous attempts to predict glucose value from continuous glucose monitors (CGM) have been published. However, there is a lack of proper analysis and modeling of penalty for errors in different glycemic ranges. The aim of this study was to investigate the potential for developing glucose prediction models with focus on the clinical aspects.

METHODS

We developed and compared six different models to test which approach were best suited for predicting glucose levels at different lead times between 10 and 60 min. The models were: last observation carried forward, linear extrapolation, ensemble methods using LSBoost and bagging, neural networks, one without error-weights and one with error-weights. The modeling and test were based on 225 type 1 diabetes patients with 315,000 h of CGM data.

RESULTS

Results show that it is possible to predict glucose levels based on CGM with a reasonable accuracy and precision with a 30-min prediction lead time. A comparison of different methods shows that there are improvements on performance gained from using more advanced machine learning algorithms (MARD 10.26-10.79 @ 30-min lead time) compared to a simple modeling (MARD 10.75-12.97 @ 30-min lead time). Moreover, the proposed use of error weights could lead to better clinical performance of these models, which is an important factor for real usage. E.g., the percentages in the C-zone of the consensus error grid without error-weights (0.57-0.68%) vs including error-weights (0.28%).

CONCLUSIONS

The results point toward that using error weighting in the training of the models could lead to better clinical performance.

Diabetes self-management in online health communities: an information exchange perspective

Background

Online health communities (OHCs), with a wealth of multi-source information exchange, have provided a convenient way for people with diabetes to actively participate in their self-management and have been widely used. Information exchange assists people with diabetes with health-related decisions to actively engage in their care, and reduce the occurrence of potential complications of diabetes. However, there has been relatively little research on the information exchange behaviors and their effect on health in professional online medical platforms—OHCs.

Objective

Using a social exchange theory, this study focuses on two sources of information (doctors and people with diabetes) to investigate information exchange behaviors and consequences. Moreover, we also examine moderating effects of information price as patients need to pay prices for consulting with doctors to obtain medical information on OHCs.

Methods

By using the Python program, a rich dataset contained 22,746 doctor-patient dialogues from December 2017 to December 2018 is collected from the biggest OHC in China. Then the logistic and ordinal regression models are used to get empirical results.

Results

We found that first information sharing from doctors and other people with diabetes can promote their information sharing behavior. Second, the moderating effects of information price are heterogeneous and change with the exchange participants. Third, rich information exchange supports self-management of people with diabetes and improves their health status.

Conclusion

This study is among the first that tests the information exchange behavior and consequence for diabetes in OHCs and examines the moderating effects of the information price. The present study produces several insights, which have implications for social exchange, patient behavior, online health communities, and information technology in diabetes self-management literature.

Forecasting of Glucose Levels and Hypoglycemic Events: Head-to-Head Comparison of Linear and Nonlinear Data-Driven Algorithms Based on Continuous Glucose Monitoring Data Only

In type 1 diabetes management, the availability of algorithms capable of accurately forecasting future blood glucose (BG) concentrations and hypoglycemic episodes could enable proactive therapeutic actions, e.g., the consumption of carbohydrates to mitigate, or even avoid, an impending critical event. The only input of this kind of algorithm is often continuous glucose monitoring (CGM) sensor data, because other signals (such as injected insulin, ingested carbs, and physical activity) are frequently unavailable. Several predictive algorithms fed by CGM data only have been proposed in the literature, but they were assessed using datasets originated by different experimental protocols, making a comparison of their relative merits difficult. The aim of the present work was to perform a head-to-head comparison of thirty different linear and nonlinear predictive algorithms using the same dataset, given by 124 CGM traces collected over 10 days with the newest Dexcom G6 sensor available on the market and considering a 30-min prediction horizon. We considered the state-of-the art methods, investigating, in particular, linear black-box methods (autoregressive; autoregressive moving-average; and autoregressive integrated moving-average, ARIMA) and nonlinear machine-learning methods (support vector regression, SVR; regression random forest; feed-forward neural network, fNN; and long short-term memory neural network). For each method, the prediction accuracy and hypoglycemia detection capabilities were assessed using either population or individualized model parameters. As far as prediction accuracy is concerned, the results show that the best linear algorithm (individualized ARIMA) provides accuracy comparable to that of the best nonlinear algorithm (individualized fNN), with root mean square errors of 22.15 and 21.52 mg/dL, respectively. As far as hypoglycemia detection is concerned, the best linear algorithm (individualized ARIMA) provided precision = 64%, recall = 82%, and one false alarm/day, comparable to the best nonlinear technique (population SVR): precision = 63%, recall = 69%, and 0.5 false alarms/day. In general, the head-to-head comparison of the thirty algorithms fed by CGM data only made using a wide dataset shows that individualized linear models are more effective than population ones, while no significant advantages seem to emerge when employing nonlinear methodologies.

Improving Accuracy for Diabetes Mellitus Prediction by Using Deepnet

Diabetes is a salient issue and a significant health care concern for many nations. The forecast for the prevalence of diabetes is on the rise. Hence, building a prediction machine learning model to assist in the identification of diabetic patients is of great interest. This study aims to create a machine learning model that is capable of predicting diabetes with high performance. The following study used the BigML platform to train four machine learning algorithms, namely, Deepnet, Models (decision tree), Ensemble and Logistic Regression, on data sets collected from the Ministry of National Guard Hospital Affairs (MNGHA) in Saudi Arabia between the years of 2013 and 2015. The comparative evaluation criteria for the four algorithms examined included; Accuracy, Precision, Recall, F-measure and PhiCoefficient. Results show that the Deepnet algorithm achieved higher performance compared to other machine learning algorithms based on various evaluation matrices.

Design and Development of Diabetes Management System Using Machine Learning

This paper describes the design and implementation of a software system to improve the management of diabetes using a machine learning approach and to demonstrate and evaluate its effectiveness in controlling diabetes. The proposed approach for this management system handles the various factors that affect the health of people with diabetes by combining multiple artificial intelligence algorithms. The proposed framework factors the diabetes management problem into subgoals: building a Tensorflow neural network model for food classification; thus, it allows users to upload an image to determine if a meal is recommended for consumption; implementing K-Nearest Neighbour (KNN) algorithm to recommend meals; using cognitive sciences to build a diabetes question and answer chatbot; tracking user activity, user geolocation, and generating pdfs of logged blood sugar readings. The food recognition model was evaluated with cross-entropy metrics that support validation using Neural networks with a backpropagation algorithm. The model learned features of the images fed from local Ghanaian dishes with specific nutritional value and essence in managing diabetics and provided accurate image classification with given labels and corresponding accuracy. The model achieved specified goals by predicting with high accuracy, labels of new images. The food recognition and classification model achieved over 95% accuracy levels for specific calorie intakes. The performance of the meal recommender model and question and answer chatbot was tested with a designed cross-platform user-friendly interface using Cordova and Ionic Frameworks for software development for both mobile and web applications. The system recommended meals to meet the calorific needs of users successfully using KNN (with k = 5) and answered questions asked in a human-like way. The implemented system would solve the problem of managing activity, dieting recommendations, and medication notification of diabetics.

An Ontology-Based Serious Game Design for the Development of Nutrition and Food Literacy Skills

Unhealthy dietary habits constitute a major risk factor for the onset of chronic diseases, such as cardiovascular diseases, cancer, diabetes and other conditions linked to obesity. Effective dietary changes are of paramount importance and can be promoted through empowering individuals with Nutrition Literacy (NL) and Food Literacy (FL) skills. This paper presents a novel serious game aiming at building NL and FL skills in adolescents and young adults. It is based on an innovative conceptual framework, incorporating a recipe ontology and a theory driven game design approach maximizing user attractiveness and promoting sustainable effective dietary changes. The ontological modeling of recipes offers game experience personalization while providing a realistic and diverse simulation environment. Modern game design techniques from three game genres (cooking, roguelike, puzzle) are employed along with a compelling plot for engagement purposes.

User Centered Design to Improve Information Exchange in Diabetes Care Through eHealth : Results from a Small Scale Exploratory Study

Heterogeneity of people with diabetes makes maintaining blood glucose control and achieving therapy adherence a challenge. It is fundamental that patients get actively involved in the management of the disease in their living environments. The objective of this paper is to evaluate the use and acceptance of a self-management system for diabetes developed with User Centered Design Principles in community settings. Persons with diabetes and health professionals were involved the design, development and evaluation of the self-management system; which comprised three iterative cycles: scenario definition, user archetype definition and system development. A comprehensive system was developed integrating modules for the management of blood glucose levels, medication, food intake habits, physical activity, diabetes education and messaging. The system was adapted for two types of principal users (personas): Type 1 Diabetes user and Type 2 Diabetes user. The system was evaluated by assessing the use, the compliance, the attractiveness and perceived usefulness in a multicenter randomized pilot study involving 20 patients and 24 treating professionals for a period of four weeks. Usage and compliance of the co-designed system was compared during the first and the last two weeks of the study, showing a significantly improved behaviour of patients towards the system for each of the modules. This resulted in a successful adoption by both type of personas. Only the medication module showed a significantly different use and compliance (p= 0.01) which can be explained by the different therapeutic course of the two types of diabetes. The involvement of patients to make their own decisions and choices form design stages was key for the adoption of a self-management system for diabetes.

Hypoglycemia Prevention via Personalized Glucose-Insulin Models Identified in Free-Living Conditions

BACKGROUND

The objective of this research is to show the effectiveness of individualized hypoglycemia predictive alerts (IHPAs) based on patient-tailored glucose-insulin models (PTMs) for different subjects. Interpatient variability calls for PTMs that have been identified from data collected in free-living conditions during a one-month trial.

METHODS

A new impulse-response (IR) identification technique has been applied to free-living data in order to identify PTMs that are able to predict the future glucose trends and prevent hypoglycemia events. Impulse response has been applied to seven patients with type 1 diabetes (T1D) of the University of Amsterdam Medical Centre. Individualized hypoglycemia predictive alert has been designed for each patient thanks to the good prediction capabilities of PTMs.

RESULTS

The PTMs performance is evaluated in terms of index of fitting (FIT), coefficient of determination, and Pearson's correlation coefficient with a population FIT of 63.74%. The IHPAs are evaluated on seven patients with T1D with the aim of predicting in advance (between 45 and 10 minutes) the unavoidable hypoglycemia events; these systems show better performance in terms of sensitivity, precision, and accuracy with respect to previously published results.

CONCLUSION

The proposed work shows the successful results obtained applying the IR to an entire set of patients, participants of a one-month trial. Individualized hypoglycemia predictive alerts are evaluated in terms of hypoglycemia prevention: the use of a PTM allows to detect 84.67% of the hypoglycemia events occurred during a one-month trial on average with less than 0.4% of false alarms. The promising prediction capabilities of PTMs can be a key ingredient for new generations of individualized model predictive control for artificial pancreas.

What do healthcare professionals need to turn risk models for type 2 diabetes into usable computerized clinical decision support systems? Lessons learned from the MOSAIC project

Background

To understand user needs, system requirements and organizational conditions towards successful design and adoption of Clinical Decision Support Systems for Type 2 Diabetes (T2D) care built on top of computerized risk models.

Methods

The holistic and evidence-based CEHRES Roadmap, used to create eHealth solutions through participatory development approach, persuasive design techniques and business modelling, was adopted in the MOSAIC project to define the sequence of multidisciplinary methods organized in three phases, user needs, implementation and evaluation. The research was qualitative, the total number of participants was ninety, about five-seventeen involved in each round of experiment.

Results

Prediction models for the onset of T2D are built on clinical studies, while for T2D care are derived from healthcare registries. Accordingly, two set of DSSs were defined: the first, T2D Screening, introduces a novel routine; in the second case, T2D Care, DSSs can support managers at population level, and daily practitioners at individual level. In the user needs phase, T2D Screening and solution T2D Care at population level share similar priorities, as both deal with risk-stratification. End-users of T2D Screening and solution T2D Care at individual level prioritize easiness of use and satisfaction, while managers prefer the tools to be available every time and everywhere. In the implementation phase, three Use Cases were defined for T2D Screening, adapting the tool to different settings and granularity of information. Two Use Cases were defined around solutions T2D Care at population and T2D Care at individual, to be used in primary or secondary care. Suitable filtering options were equipped with “attractive” visual analytics to focus the attention of end-users on specific parameters and events. In the evaluation phase, good levels of user experience versus bad level of usability suggest that end-users of T2D Screening perceived the potential, but they are worried about complexity. Usability and user experience were above acceptable thresholds for T2D Care at population and T2D Care at individual.

Conclusions

By using a holistic approach, we have been able to understand user needs, behaviours and interactions and give new insights in the definition of effective Decision Support Systems to deal with the complexity of T2D care.

Electronic supplementary material

The online version of this article (10.1186/s12911-019-0887-8) contains supplementary material, which is available to authorized users.

A Comprehensive Medical Decision–Support Framework Based on a Heterogeneous Ensemble Classifier for Diabetes Prediction

Early diagnosis of diabetes mellitus (DM) is critical to prevent its serious complications. An ensemble of classifiers is an effective way to enhance classification performance, which can be used to diagnose complex diseases, such as DM. This paper proposes an ensemble framework to diagnose DM by optimally employing multiple classifiers based on bagging and random subspace techniques. The proposed framework combines seven of the most suitable and heterogeneous data mining techniques, each with a separate set of suitable features. These techniques are k-nearest neighbors, naïve Bayes, decision tree, support vector machine, fuzzy decision tree, artificial neural network, and logistic regression. The framework is designed accurately by selecting, for every sub-dataset, the most suitable feature set and the most accurate classifier. It was evaluated using a real dataset collected from electronic health records of Mansura University Hospitals (Mansura, Egypt). The resulting framework achieved 90% of accuracy, 90.2% of recall = 90.2%, and 94.9% of precision. We evaluated and compared the proposed framework with many other classification algorithms. An analysis of the results indicated that the proposed ensemble framework significantly outperforms all other classifiers. It is a successful step towards constructing a personalized decision support system, which could help physicians in daily clinical practice.

A Two-Level Food Classification System For People With Diabetes Mellitus Using Convolutional Neural Networks

Accurate estimation of food's macronutrient content for people with Diabetes Mellitus (DM) is of great importance, as it determines postprandial insulin dosage. This paper introduces a classification system for food images that is adjusted to the nutritional needs of people with DM. A two-level image classification scheme, exploiting Convolutional Neural Networks (CNNs), is proposed, in order to classify an image in one of eight broad food categories with similar macronutrient content and then assign it to a specific food within that category. To this end, a visual dataset, namely NTUA-Food 2017, has been designed, consisting of 3248 images organized in eight broad food categories of totally 82 different foods. Moreover, a novel evaluation metric is proposed, which penalizes classification errors proportionally to the discrepancy in postprandial blood sugar levels between the actual and predicted class. The proposed system achieves 84.18% and 85.94% classification accuracy at the first and second level of classification, respectively, on the NTUA-Food 2017 dataset. The algorithm developed for the first level of classification on the NTUA-Food 2017 dataset improves classification accuracy on the benchmark Food Image Dataset (FID) to 97.08% outperforming previous approaches. The algorithm's mean error in terms of carbohydrate content estimation on the NTUA-Food 2017 dataset is less than 2 g per food serving.

An ARIMA Model With Adaptive Orders for Predicting Blood Glucose Concentrations and Hypoglycemia

The continuous glucose monitoring system is an effective tool, which enables the users to monitor their blood glucose (BG) levels. Based on the continuous glucose monitoring (CGM) data, we aim at predicting future BG levels so that appropriate actions can be taken in advance to prevent hyperglycemia or hypoglycemia. Due to the time-varying nonstationarity of CGM data, verified by Augmented Dickey-Fuller test and analysis of variance, an autoregressive integrated moving average (ARIMA) model with an adaptive identification algorithm of model orders is proposed in the prediction framework. Such identification algorithm adaptively determines the model orders and simultaneously estimates the corresponding parameters using Akaike Information Criterion and least square estimation. A case study is conducted with the CGM data of diabetics under daily living conditions to analyze the prediction performance of the proposed model together with the early hypoglycemic alarms. Results show that the proposed model outperforms the adaptive univariate model and ARIMA model.

Limits to the Evaluation of the Accuracy of Continuous Glucose Monitoring Systems by Clinical Trials

Systems for continuous glucose monitoring (CGM) are evolving quickly, and the data obtained are expected to become the basis for clinical decisions for many patients with diabetes in the near future. However, this requires that their analytical accuracy is sufficient. This accuracy is usually determined with clinical studies by comparing the data obtained by the given CGM system with blood glucose (BG) point measurements made with a so-called reference method. The latter is assumed to indicate the correct value of the target quantity. Unfortunately, due to the nature of the clinical trials and the approach used, such a comparison is subject to several effects which may lead to misleading results. While some reasons for the differences between the values obtained with CGM and BG point measurements are relatively well-known (e.g., measurement in different body compartments), others related to the clinical study protocols are less visible, but also quite important. In this review, we present a general picture of the topic as well as tools which allow to correct or at least to estimate the uncertainty of measures of CGM system performance.

DMTO: a realistic ontology for standard diabetes mellitus treatment

BACKGROUND

Treatment of type 2 diabetes mellitus (T2DM) is a complex problem. A clinical decision support system (CDSS) based on massive and distributed electronic health record data can facilitate the automation of this process and enhance its accuracy. The most important component of any CDSS is its knowledge base. This knowledge base can be formulated using ontologies. The formal description logic of ontology supports the inference of hidden knowledge. Building a complete, coherent, consistent, interoperable, and sharable ontology is a challenge.

RESULTS

This paper introduces the first version of the newly constructed Diabetes Mellitus Treatment Ontology (DMTO) as a basis for shared-semantics, domain-specific, standard, machine-readable, and interoperable knowledge relevant to T2DM treatment. It is a comprehensive ontology and provides the highest coverage and the most complete picture of coded knowledge about T2DM patients' current conditions, previous profiles, and T2DM-related aspects, including complications, symptoms, lab tests, interactions, treatment plan (TP) frameworks, and glucose-related diseases and medications. It adheres to the design principles recommended by the Open Biomedical Ontologies Foundry and is based on ontological realism that follows the principles of the Basic Formal Ontology and the Ontology for General Medical Science. DMTO is implemented under Protégé 5.0 in Web Ontology Language (OWL) 2 format and is publicly available through the National Center for Biomedical Ontology's BioPortal at http://bioportal.bioontology.org/ontologies/DMTO . The current version of DMTO includes more than 10,700 classes, 277 relations, 39,425 annotations, 214 semantic rules, and 62,974 axioms. We provide proof of concept for this approach to modeling TPs.

CONCLUSION

The ontology is able to collect and analyze most features of T2DM as well as customize chronic TPs with the most appropriate drugs, foods, and physical exercises. DMTO is ready to be used as a knowledge base for semantically intelligent and distributed CDSS systems.

Comparison of Machine Learning Approaches Toward Assessing the Risk of Developing Cardiovascular Disease as a Long-Term Diabetes Complication

The estimation of long-term diabetes complications risk is essential in the process of medical decision making. Guidelines for the management of Type 2 Diabetes Mellitus (T2DM) advocate calculating the Cardiovascular Disease (CVD) risk to initiate appropriate treatment. The objective of this study is to investigate the use of sophisticated machine learning techniques toward the development of personalized models able to predict the risk of fatal or nonfatal CVD incidence in T2DM patients. The important challenge of handling the unbalanced nature of the available dataset is addressed by applying novel ensemble strategies. Hybrid Wavelet Neural Networks (HWNNs) and Self-Organizing Maps (SOMs) constitute the primary models for building ensembles following a subsampling approach. Different methods for combining the decisions of the primary models are applied and comparatively assessed. Data from the 5-year follow up of 560 patients with T2DM are used for development and evaluation purposes. The highest discrimination performance (Area Under the Curve (AUC): 71.48%) is achieved by taking into account both the HWNN- and SOM- based primary models' outputs. The proposed method is superior to the Binomial Linear Regression (BLR) model justifying the need to apply more sophisticated techniques in order to produce reliable CVD risk scores.

A review of personalized blood glucose prediction strategies for T1DM patients

This paper presents a methodological review of models for predicting blood glucose (BG) concentration, risks and BG events. The surveyed models are classified into three categories, and they are presented in summary tables containing the most relevant data regarding the experimental setup for fitting and testing each model as well as the input signals and the performance metrics. Each category exhibits trends that are presented and discussed. This document aims to be a compact guide to determine the modeling options that are currently being exploited for personalized BG prediction.

Effect of acrylodan conjugation and forced oxidation on the structural integrity, conformational stability, and binding activity of a glucose binding protein SM4 used in a prototype continuous glucose monitor

Continuous glucose monitoring (CGM) devices offer diabetes patients a convenient approach to assist in controlling blood glucose levels. A prototype CGM has been developed that uses the emission profile of a polarity-sensitive fluorophore (acrylodan) conjugated to a glucose/galactose-binding protein (SM4-AC) to measure the concentration of glucose in vivo. During development, a decrease in the devices signal intensity was observed in vivo over time, which was postulated to be result of oxidative degradation of SM4-AC. A comprehensive physicochemical analysis of SM4-AC was pursued to identify potential mechanisms of signal intensity loss in this CGM during in vitro forced oxidation studies. An assessment of the structural integrity and conformational stability of SM4-AC indicated a relatively decreased polarity and lower tertiary structure stability compared to unconjugated protein (SM4). The stability and polarity of SM4-AC was also altered in the presence of H₂ O₂ . Furthermore, a time-dependent loss in the fluorescence signal of SM4-AC was observed when incubated with H₂ O₂ . An LC-MS peptide mapping analysis of these protein samples indicated that primarily two Met residues in SM4-AC were susceptible to oxidation. When these two residues were genetically altered to an amino acid not prone to oxidation, the glucose binding ability of the protein was retained and no loss of acrylodan fluorescence was observed in the presence of H₂ O₂ . Genetic alteration of these two residues is proposed as an effective approach to increase the long-term stability of SM4-AC within this prototype CGM in vivo.

Profiling intra-patient type I diabetes behaviors

BACKGROUND

The large intra-patient variability in type 1 diabetic patients dramatically reduces the ability to achieve adequate blood glucose control. A novel methodology to identify different blood glucose dynamics profiles will allow therapies to be more accurate and tailored according to patient's conditions and to the situations faced by patients (exercise, week-ends, holidays, menstruation, etc).

MATERIALS AND METHODS

A clustering methodology based on the normalized compression distance is applied to identify different profiles for diabetic patients. First, the methodology is validated using "in silico" data from 10 patients in 3 different scenarios: days without exercise, poor controlled exercise days and days with well-controlled exercise. Second, we perform a series of in vivo experiments using data from 10 patients assessing the ability of the proposed methodology in real scenarios.

RESULTS

In silico experiments show that the methodology is able to identify poor and well-controlled days in theoretical scenarios. In vivo experiments present meaningful profiles for working days, bank days and other situations, where different insulin requirements were detected.

CONCLUSIONS

A tool for profiling blood glucose dynamics of patients can be implemented in a short term to enhance existing analysis platforms using combined CGM-CSII systems. Besides coping with the information overload, the tool will assist physicians to adjust and improve insulin therapy and patients in the self-management of the disease.

Modeling Glucagon Action in Patients With Type 1 Diabetes

The dual-hormone artificial pancreas is an emerging technology to treat type 1 diabetes (T1D). It consists of a glucose sensor, infusion pumps, and a dosing algorithm that directs hormonal delivery. Preclinical optimization of dosing algorithms using computer simulations has the potential to accelerate the pace of development for this technology. However, current simulation environments consider glucose regulation models that either do not include glucagon action submodels or include submodels that were proposed without comparison to other candidate models. We consider here nine candidate models of glucagon action featuring a number of possible characteristics: insulin-independent glucagon action, insulin/glucagon ratio effect on hepatic glucose production, insulin-dependent suppression of glucagon action, and the effect of rate of change of glucagon. To assess the models, we use measurements of plasma insulin, plasma glucagon, and endogenous glucose production collected from experiments involving eight subjects with T1D who receive four subcutaneous glucagon boluses. We estimate each model's parameters using a Bayesian approach, and the models are contrasted based on the deviance information criterion. The model achieving the best fit features insulin-dependent suppression of glucagon action and incorporates effects of both glucagon levels and its rate of change.

Non-invasive detection of fasting blood glucose level via electrochemical measurement of saliva

Machine learning techniques such as logistic regression (LR), support vector machine (SVM) and artificial neural network (ANN) were used to detect fasting blood glucose levels (FBGL) in a mixed population of healthy and diseased individuals in an Indian population. The occurrence of elevated FBGL was estimated in a non-invasive manner from the status of an individual’s salivary electrochemical parameters such as pH, redox potential, conductivity and concentration of sodium, potassium and calcium ions. The samples were obtained from 175 randomly selected volunteers comprising half healthy and half diabetic patients. The models were trained using 70 % of the total data, and tested upon the remaining set. For each algorithm, data points were cross-validated by randomly shuffling them three times prior to implementing the model. The performance of the machine learning technique was reported in terms of four statistically significant parameters—accuracy, precision, sensitivity and F1 score. SVM using RBF kernel showed the best performance for classifying high FBGLs with approximately 85 % accuracy, 84 % precision, 85 % sensitivity and 85 % F₁ score. This study has been approved by the ethical committee of All India Institute of Medical Sciences, New Delhi, India with the reference number: IEC/NP-278/01-08-2014, RP-29/2014.