Data-Driven Blood Glucose Pattern Classification and Anomalies Detection: Machine-Learning Applications in Type 1 Diabetes

Background Diabetes mellitus is a chronic metabolic disorder that results in abnormal blood glucose (BG) regulations. The BG level is preferably maintained close to normality through self-management practices, which involves actively tracking BG levels and taking proper actions including adjusting diet and insulin medications. BG anomalies could be defined as any undesirable reading because of either a precisely known reason (normal cause variation) or an unknown reason (special cause variation) to the patient. Recently, machine-learning applications have been widely introduced within diabetes research in general and BG anomaly detection in particular. However, irrespective of their expanding and increasing popularity, there is a lack of up-to-date reviews that materialize the current trends in modeling options and strategies for BG anomaly classification and detection in people with diabetes. Objective This review aimed to identify, assess, and analyze the state-of-the-art machine-learning strategies and their hybrid systems focusing on BG anomaly classification and detection including glycemic variability (GV), hyperglycemia, and hypoglycemia in type 1 diabetes within the context of personalized decision support systems and BG alarm events applications, which are important constituents for optimal diabetes self-management. Methods A rigorous literature search was conducted between September 1 and October 1, 2017, and October 15 and November 5, 2018, through various Web-based databases. Peer-reviewed journals and articles were considered. Information from the selected literature was extracted based on predefined categories, which were based on previous research and further elaborated through brainstorming. Results The initial results were vetted using the title, abstract, and keywords and retrieved 496 papers. After a thorough assessment and screening, 47 articles remained, which were critically analyzed. The interrater agreement was measured using a Cohen kappa test, and disagreements were resolved through discussion. The state-of-the-art classes of machine learning have been developed and tested up to the task and achieved promising performance including artificial neural network, support vector machine, decision tree, genetic algorithm, Gaussian process regression, Bayesian neural network, deep belief network, and others. Conclusions Despite the complexity of BG dynamics, there are many attempts to capture hypoglycemia and hyperglycemia incidences and the extent of an individual’s GV using different approaches. Recently, the advancement of diabetes technologies and continuous accumulation of self-collected health data have paved the way for popularity of machine learning in these tasks. According to the review, most of the identified studies used a theoretical threshold, which suffers from inter- and intrapatient variation. Therefore, future studies should consider the difference among patients and also track its temporal change over time. Moreover, studies should also give more emphasis on the types of inputs used and their associated time lag. Generally, we foresee that these developments might encourage researchers to further develop and test these systems on a large-scale basis.

Reconstruction of CGM data using spline interpolation and a rough feature elimination, using fast SEPCOR algorithm. Support vector machine Sensitivity, and specificity [13] 21 Real BG, Meal, Rate of decrease from a peak and absolute level of the BG at the decision point Diagnostic (professional) CGM devices N/A Decision trees Accuracy, sensitivity, and specificity [14] 1 Real (Male) Glucose levels right before meals (G1), Glucose levels after more than 5 hours (G2), Time interval (T), Average Fasting glucose level (AG1), The rate of decrease in [Glu], Ratio of current level to average (RBF), exponential radial basis function (ERBF) and polynomial function)-hybrid particle swarm optimization [38] 15 Real (children) BG, heart rate (HR), corrected QT (QTc), change in the heart rate (ΔHR) and change in the QTc interval (ΔQTc) Compumedics system, BGLs were acquired using Yellow Spring Instruments Normalization Hybrid particle swarm optimization based normalized radial basis function neural network (NRBFNN)-hybrid particle swarm optimization with wavelet mutation (HPSOWM) Sensitivity and specificity [39] & [40] 15 Real (children) BG, heart rate (HR) and corrected QT interval (QTc) Compumedics system, BGLs were acquired using Yellow Spring Instruments Normalization Variable translation wavelet neural network (VTWNN)hybrid particle swarm optimization with wavelet mutation (HPSOWM) Sensitivity and specificity [41] & [42] 15 Real (children) BG, heart rate (HR) and the corrected QT interval (QTc) Compumedics system, BGLs were acquired using Yellow Spring Instruments Normalization Evolvable block based neural network (BBNN)-hybrid particle swarm optimization with wavelet mutation (HPSOWM) Sensitivity, specificity, ROC Curve, and geometric mean value [43] 15 Real (children) BG, heart rate (HR) and corrected QT (QTc) Compumedics system, BGLs were acquired using Yellow Spring Instruments N/A Adaptive neural fuzzy inference system (ANFIS)-hybrid particle swarm optimization with wavelet mutation (HPSOWM) Sensitivity and specificity [44] 15  Compared support vector regression (SVR) and Gaussian process (GP).
[8] Fuzzy neural network estimator algorithm (FNNE) predicted the onset of hypoglycemia episodes with a mean error of 0.071 (p < 0.03) The FNNE algorithm was developed as a parallel combination of fuzzy inference mechanism (FIM) and a multi-layered neural network architecture. [9] & [10] Support vector regression (SVR) -with an event-based sensitivity of 100%, the algorithm produced only one false hypoglycemia detection. The sample-based sensitivity and specificity levels were 78% and 96%, respectively Developed an android based system to detect hypoglycemia incidence using CGM and other information.
[16] & [17] & [18] Genetic algorithm based multiple regression with fuzzy inference -Sensitivity (75%) and specificity (over 50%) Genetic algorithm is used to optimize regression and fuzzy rules. Compared various order multiple Regression Fuzzy Inference System and Linear multiple regression with various number of inputs.
Bayesian neural network -Sensitivity (83.46%) and specificity (63.88%) Investigated the applicability of Bayesian neural network to detect hypoglycemia from real time physiological parameters.
Investigated towards predicting hypoglycemia incidence during intravenous (IV) insulin infusion for ICU patients.
[53] Feed forward multi-layer neural network -Sensitivity (70.59%), specificity (65.38%) and geometric mean (67.94%) Compared the ANN model with Linear Discriminant Analysis (LDA) and K-Nearest Neighbors (KNN) on hyperglycemia detection. [52,54] Hidden Markov model (HMM) -The simulation result show that the proposed model is capable of detecting anomalies (i.e., no false positives) from the CGM readings based on historical data (in the presence of reasonable changes in the patient's daily routine).
Investigated the applicability of Hidden Markov model (HMM) in anomalies detection from the change in the patient's daily lifestyle.
[55] Naïve Bayes classifier -matched the physicians' classifications 85% of the time that they were internally consistent and in agreement with each other.
Investigated into the applicability of characterizing blood glucose variability using new metrics with CGM data using Naïve Bayes classifier. [56] SVR models -When applied to 262 different CGM plots as a screen for excessive GV (accuracy (90.1%), sensitivity (97.0%), and specificity (74.1%).
Investigated on an automatic glycemic variability detection and compared Naive Bayes (NB), Multilayer Perceptron (MP), and Support Vector Machine (SVM) models using CGM data.
[58] Artificial Neural network -Average Accuracy (90%), Average sensitivity (72.23%) and Average specificity (92%) Developed Artificial Neural Network integrated with physiological model for both blood glucose prediction and classification of hypoglycemia and further compared the result with existing models. [59] Bayesian regularized neural network -Sensitivity (73%) and specificity (60%) Investigated and tested a feed-forward neural network trained with Bayesian regularization algorithm.