Predicting Hypoglycemia in Diabetic Patients Using Time-Sensitive Artificial Neural Networks

Abstract

Type-One Diabetes Mellitus (T1DM) is a chronic disease characterized by the elevation of glucose levels within patient's blood. It can lead to serious complications including kidney and heart diseases, stroke, and blindness. The proper treatment of diabetes, on the other hand, can lead to a normal longevity. Yet such a treatment requires tight glycemic control which increases the risk of developing hypoglycemia; a sudden drop in patients' blood glucose levels that could lead to coma and possibly death. Continuous Glucose Monitoring (CGM) devices placed on a patient body, measure glucose levels every few minutes. These devices can also detect hypoglycemia. Yet detecting hypoglycemia sometimes is too late for a patient to take proper action, so a better approach is to predict the hypoglycemic events ahead of time and alarm the patient to such occurrences. In this research, the authors develop a system that involves a special type of Artificial Neural Networks (ANN), the Time-Sensitive ANN (TS-ANN), to predict hypoglycemia events ahead of time and within a prediction horizon of thirty minutes. This period should be long enough to enable diabetic patients to avoid hypoglycemia by taking a proper action. A TS-ANN based system that is able to predict hypoglycemia events have been developed and tested with high accuracy results (average specificity of 98.2%, average accuracy of 97.6% and average sensitivity of 80.19% with a maximum value reaching 93%).

Sense and Learn: Recent Advances in Wearable Sensing and Machine Learning for Blood Glucose Monitoring and Trend-Detection

“…In the case of imbalanced data, oversampling techniques can address this and improve the accuracy ( Mayo et al, 2019 ). Acquiring data from the same participant for longer periods allows the machine learning algorithm to combat intra-individual differences and increases overall prediction performance ( Eljil et al, 2016 ). The reliance on data annotated by the participants is an issue in acquiring accurate information as it depends solely on their commitment ( Bertachi et al, 2020 ).…”

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

“…There are various types of ANNs used in solving BG classification and anomaly detection tasks: hypoglycemia, hyperglycemia, and GV classification and detection. Regarding hypoglycemia classification and detection, for instance, Eljil et al [48], had proposed a special type of ANN known as the time-sensitive ANN and compared the result with a time delay neural network, NARX, distributed time delay neural network, and NAR. San et al [37,49] proposed an evolvable BBNN and compared the result with feedforward ANNs and multiple regression.…”

Using machine learning to predict severe hypoglycaemia in hospital

“…The keywords mapped to concepts that identify an increased risk of hypoglycaemia: prior hypoglycaemia, type 2 diabetes mellitus, type 1 diabetes mellitus, renal failure, cirrhosis, frailty, and dementia. The concepts were based on prior literature and clinical experience that identified each to be associated with hypoglycaemia 8,14 . A keyword search was also used on the electronic nursing notes to approximate a patient's oral intake.…”