Machine Learning Based Continuous Glucose Monitoring System

Abstract

One of the most encouraging advancements to track blood sugar levels in people with diabetes who require insulin treatment is aclosed-loop insulin delivery system (also known as the artificial pancreas). Such a system incorporates continuous glucose monitoring (CGM), insulin (with or without glucagon) infusion, and a control algorithm to constantly direct blood glucose levels. In this model we incorporated machine learning based models to anticipate and forecast future glucose levels in the blood based on two study populations (CGM based and CGM- and accelerometery-based glucose predictions. We used data from The Maastricht Study, an observational, imminent, populace-based accomplice study. The Maastricht Study is broad phenotyping study that focuses on the etiology of type 2 diabetes (T2DM), its classic complications, and its arising co morbidities.

Models trained with CGM data were capable to accurately anticipate glucose values at 15 (RMSE: 0.19mmol/L; rho: 0.96) and 60 minutes (RMSE: 0.59mmol/L, rho: 0.72). Model performance was comparable in individuals with type 2 diabetes. Incorporation of accelerometer data only slightly improved prediction. Prediction models translated well to individuals with type 1 diabetes, which is reflected by high precision (RMSEs for 15 and 60 minutes of 0.43 and 1.73 mmol/L, respectively) and clinical safety. Hence machine learning models are able to predict the future glucose levels accurately and precisely than the traditional non-invasive methods like closed loop monitoring.