Reinforcement learning for bolus insulin dosing for people with type 1 diabetes

Full Text
embargat.txt embargoed access
Request a copy
When filling the form you are requesting a copy of the article, that is deposited in the institutional repository (DUGiDocs), at the autor or main autor of the article. It will be the same author who decides to give a copy of the document to the person who requests it, if it considers it appropriate. In any case, the UdG Library doesn’t take part in this process because it is not authorized to provide restricted articles.
Share
Type 1 Diabetes (T1D) is a chronic metabolic disorder caused by destruction of the insulin producing beta cells in the islets of Langerhans within the pancreas due to an autoimmune reaction. T1D is distinguished by elevated levels of blood glucose (BG) owing to the deficiency of insulin, a hormone responsible for the regulation of BG within the normal range of 70-180 mg/dL. T1D is associated with various micro-vascular and macro-vascular complications such as nephropathy, neuropathy, retinopathy, coronary heart disease, cerebrovascular disease, peripheral artery disease etc. People with T1D rely on the administration of exogenous insulin to restrict the BG in a healthy range. The insulin treatment strategies for T1D can be broadly divided into two categories i.e., multiple daily injections (MDI) and continuous subcutaneous insulin infusion (CSII) to avoid the T1D complications. In the past decade, a significant effort has been made by researchers to reproduce the behavior of beta cells and automate the insulin delivery for the management of T1D paving a way for the rapid development of the artificial pancreas (AP) technology. Integration of a continuous glucose monitor (CGM) with closed-loop control (CLC) algorithms to compute the continuous insulin dosing rate constitute an AP system. The preclinical validation and evaluation of the insulin dosing strategies developed by researchers are performed in the simulation environments that represent virtual patients (VPs) with T1D. The work presented in this thesis provides three contributions. Firstly, a methodology is introduced to generate a cohort of VPs with T1D to replicate the BG metrics of a real cohort of people with T1D from the Hospital Clinic de Barcelona. The clinical data of meals, meal times and insulin (basal and bolus) was utilized to derive realistic scenarios for the generation of VPs. The exercise sessions were introduced as disturbances and were derived from the BG profile of the real patients. The proposed methodology is capable of adopting the daily variations of BG profile from real patients and thus provide a realistic and challenging simulation environment for the validation and evaluation of therapeutic strategies developed for the management of T1D. Secondly, a Q-Learning based Reinforcement Learning (RL) algorithm is proposed for the bolus insulin calculation in patients with T1D and validated on the generated cohort of VPs with T1D. Usually the bolus insulin calculation is based on carbohydrates (CHO) in meal, CHO to insulin ratio (CR) and the insulin sensitivity based correction factor (CF). On the contrary, the proposed algorithm is independent of the CHO content in meals, CR and CF with an aim to avoid the CHO estimation and counting errors and the management burden on patients with T1D ​
​ADVERTIMENT. Tots els drets reservats. L'accés als continguts d'aquesta tesi doctoral i la seva utilització ha de respectar els drets de la persona autora. Pot ser utilitzada per a consulta o estudi personal, així com en activitats o materials d'investigació i docència en els termes establerts a l'art. 32 del Text Refós de la Llei de Propietat Intel·lectual (RDL 1/1996). Per altres utilitzacions es requereix l'autorització prèvia i expressa de la persona autora. En qualsevol cas, en la utilització dels seus continguts caldrà indicar de forma clara el nom i cognoms de la persona autora i el títol de la tesi doctoral. No s'autoritza la seva reproducció o altres formes d'explotació efectuades amb finalitats de lucre ni la seva comunicació pública des d'un lloc aliè al servei TDX. Tampoc s'autoritza la presentació del seu contingut en una finestra o marc aliè a TDX (framing). Aquesta reserva de drets afecta tant als continguts de la tesi com als seus resums i índexs.