Open AccessMultivariate learning framework for long‐term adaptation in the artificial pancreas
Author(s) -
Shi Dawei,
Dassau Eyal,
Doyle Francis J.
Publication year - 2019
Publication title -
bioengineering and translational medicine
Language(s) - English
Resource type - Journals
ISSN - 2380-6761
DOI - 10.1002/btm2.10119
Subject(s) - artificial pancreas , glycemic , hypoglycemia , computer science , multivariate statistics , term (time) , cohort , adaptation (eye) , controller (irrigation) , bayesian probability , basal insulin , machine learning , artificial intelligence , diabetes mellitus , medicine , type 1 diabetes , endocrinology , type 2 diabetes , biology , agronomy , physics , quantum mechanics , neuroscience
The long‐term use of the artificial pancreas (AP) requires an automated insulin delivery algorithm that can learn and adapt with the growth, development, and lifestyle changes of patients. In this work, we introduce a data‐driven AP adaptation method for improved glucose management in a home environment. A two‐phase Bayesian optimization assisted parameter learning algorithm is proposed to adapt basal and carbohydrate‐ratio profile, and key feedback control parameters. The method is evaluated on the basis of the 111‐adult cohort of the FDA‐accepted UVA/Padova type 1 diabetes mellitus simulator through three scenarios with lifestyle disturbances and incorrect initial parameters. For all the scenarios, the proposed method is able to robustly adapt AP parameters for improved glycemic regulation performance in terms of percent time in the euglycemic range [70, 180] mg/dl without causing risk of hypoglycemia in terms of percent time below 70 mg/dl.