Smart, Secure, Non Invasive Wearable System for Proactive Detection of Hypoglycemic Events
Grant
Overview
Affiliation
View All
Overview
abstract
An estimated 415 million people have diabetes world-wide according to the International Diabetes federation, of which 35.4 million people are from the Middle East and North Africa (MENA) region including Qatar. Diabetes prevalence in Qatar is approximately two times the prevalence in the world with up to 35% of adults aged 55 and over costing the Qatari economy about 240 billion per year [2]. One of the potentially life-threatening conditions for patients under treatment for diabetes is hypoglycemia (low blood glucose) which could result from incorrect medication, over-exercise, stress, excessive alcohol consumption, delayed meals, and fasting. Symptoms of hypoglycemia are tremors, sweating, tiredness, lightheadedness, disorientedness, and palpitations [3]. Frequent hypoglycemic episodes not only affect patient quality of life but also could result in coma, stupor, behavior change (e.g., confusion or bizarre behavior), dizziness, seizures, and death. According to American Diabetes Association (ADA), about 13% of type 1 diabetes deaths in the United States are due to hypoglycemic events. Existing solutions for glucose monitoring such as continuous glucose monitors (CGMs) are invasive, costly, and reactive. Self-monitored glucose approaches are intermittent and miss hypoglycemic events, especially at night. Also, a significant portion of the type 2 diabetes patients currently do not use CGMs and thus cannot monitor for hypoglycemic events. We propose to develop a wearable, non-invasive, reliable, inexpensive, and proactive device to detect and prevent hypoglycemic events by detecting early onsets of hypoglycemic tremor. Hypoglycemia tremor is categorized as physiological tremor with frequencies ranging from 6-12 Hz. The device will use compact high-precision accelerometers and will have two novel configurations to support arm/wrist and finger-based use to capture low frequency physiologic tremors. Currently, remote sensors to detect tremor in a wearable configuration do not exist and will be a contribution of the proposed effort. Existing approaches to tremor detection are based on quantification of movement over a time period. The innovation of our approach is to develop machine learning-based pattern detection algorithms to explicitly detect and characterize the frequency and amplitude of tremor. Innovative ensemble model fusion based on multiple machine learning and neural network models, and feature extraction will be used to improve the sensitivity and specificity of tremor detection. Predictive algorithms will be developed for personalized hypoglycemia event risk characterization based on the detected tremor, activity monitoring, and individual baseline characterization..........
