Assessment of Blood Glucose Level Prediction Using LSTM Deep Learning Method

Abstract

Continuous glucose monitoring has significantly improved the situations of patients with T1DM in this day and age of technological advancement as new procedures and technologies in clinical medicine have been developed. An artificial pancreas was recently developed to help these people manage their glucose levels. An artificial pancreas is a three-part device that mimics how the body's functional pancreas maintains blood glucose, also known as blood sugar. A synthetic pancreas is mostly used to help people with type 1 diabetes mellitus (T1DM). However, to create a successful artificial pancreas, the first step is to predict the future blood glucose level of diabetes patients to enable proactive and precise control of insulin delivery to them, with the application of artificial intelligence (AI) to accelerate and simplify this process. Thus, a study on deep learning which is a subset of the AI was conducted for the blood glucose level time-series prediction based on the Cobelli model of T1DM and is presented in this paper. In this study, Long Short-Term Memory (LSTM) deep learning model was built to predict the blood glucose levels for 3- and 6-minute prediction horizon (PH) using Python Programming. Besides, the time-series data of the blood glucose level used to train the model was generated from the T1DM-typed Cobelli model-based open-loop insulin delivery system developed in MATLAB Simulink software. From the results, the 3-minute PH generally outperformed the 6-minute PH in terms of mean absolute error (MAE), mean squared error (MSE) and root mean squared error (RMSE), indicating better prediction performance. In conclusion, to some extent, the finding contributes to the ongoing efforts in developing advanced technologies for diabetes management.