Sammendrag
Type 1 diabetes is a metabolic disorder characterized by high blood glucose levels as a consequence of deficiency of the hormone insulin, requiring the patient to follow a strict personalized protocol of food intake, subcutaneous insulin administration as a lifelong treatment for the high blood glucose levels, and exercise. This condition leads to acute complications, damaging several organs and tissues throughout the patient's body. Despite years of research and clinical trials, no cure for type 1 diabetes exists yet.
New technologies have impacted current research for type 1 diabetes, changing how the disease is treated and leading to vast improvements in patient's quality of life. Among others, the artificial pancreas for automatically regulating blood glucose levels has gained importance in recent years, becoming the holy grail of the diabetes research.
The artificial pancreas has opened doors for new research fields, and recent advances are focused on automated insulin delivery systems for blood glucose control. This has resulted in the application of machine learning techniques competing with traditional control approaches. Concretely, reinforcement learning methods have emerged as a promising and personalized solution for the blood glucose regulation problem in type 1 diabetes.
The objective of this thesis is to develop control algorithms to automatically adjust insulin delivery based on data from both, the blood glucose concentrations and the administered insulin, to improve diabetes management in the artificial pancreas system for type 1 diabetes patients. Concretely, this work explores the use of reinforcement learning algorithms as an alternative approach to the traditional control methods used in the artificial pancreas system for the blood glucose control task. Specifically, the effort is dedicated to recognize the challenges and the opportunities in the artificial pancreas system, analyze the state-of-the-art in diabetes blood glucose control using reinforcement learning approaches, identify the existing problems, and provide solutions based on reinforcement learning.
Har del(er)
Paper I: Tejedor, M., Woldaregay, A.Z. & Godtliebsen, F. (2020). Reinforcement learning application in diabetes blood glucose control: A systematic review. Artificial Intelligence in Medicine, 104, 101836. Also available at https://doi.org/10.1016/j.artmed.2020.101836.
Paper II: Tejedor, M. & Myhre, J.N. (2020). Controlling Blood Glucose For Patients With Type 1 Diabetes Using Deep Reinforcement Learning - The Influence Of Changing The Reward Function. Proceedings of the Northern Lights Deep Learning Workshop, 1. Also available at https://doi.org/10.7557/18.5166.
Paper III: Myhre, J.N., Tejedor, M., Launonen, I.K., El Fathi, A. & Godtliebsen, F. (2020). In-Silico Evaluation of Glucose Regulation Using Policy Gradient Reinforcement Learning for Patients with Type 1 Diabetes Mellitus. Applied Sciences, 10(18), 6350. Also available in Munin at https://hdl.handle.net/10037/20656.
Paper IV: Ngo, P., Tejedor, M., Tayefi, M., Chomutare, T. & Godtliebsen, F. (2020). Risk-Averse Food Recommendation Using Bayesian Feedforward Neural Networks for Patients with Type 1 Diabetes Doing Physical Activities. Applied Sciences, 10(22), 8037. Also available in Munin at https://hdl.handle.net/10037/19856.
English
norsk
