Evaluation of an artificial pancreas in in silico patients with online-tuned internal model control. (March 2018)

• Record Type: Journal Article
• Authors: Bhattacharjee, Arpita; Easwaran, Arvind; Leow, Melvin Khee-Shing; Cho, Namjoon
• Is Part Of: Biomedical signal processing and control. Volume 41(2018)
• Page Start: 198
• Page End: 209
• Publication Date: 2018-03
• Subjects: Type 1 diabetes mellitus -- Artificial pancreas -- Internal model control -- Volterra model -- Recursive least squares; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2017.12.002
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: An internal model control (IMC) algorithm that can tune the controller both offline and online in the presence of intra- and inter-patient variabilities and unannounced disturbances is evaluated. The IMC is developed online using frequency domain Volterra kernel of Volterra model. The IMC algorithms have been evaluated for different scenarios in the UVA/Padova metabolic simulator. The IMC compensates unannounced meal disturbances with low insulin infusion that reduces the risk of hypoglycemic events. The controller performs robustly with variations in insulin sensitivity, in the presence of irregular amounts of meal disturbances at random times, and in the presence of very high noise levels in the sensors and actuators. Abstract: A fully-automated controller in the artificial pancreas (AP) system designed to regulate blood glucose concentration can give better lifestyle to a type 1 diabetic patient. This paper deals with evaluating the benefit of fully-automated online-tuned controller for the AP system over offline-tuned and semi-automated controller based on internal model control (IMC) strategy. The online-tuned controller is fully-automatic in the sense that it can automatically deal with intra- and inter-patient variabilities and compensate for unannounced meal disturbances without any prior knowledge of patient parameters, patient specific characteristics or patient specific input–output data. A data driven Volterra model of patients is used to design IMC algorithms. For online-tuned controller, the Volterra kernels of the model are computed online by recursive least squares algorithm. The IMC algorithms are evaluated using different scenarios in the UVA/Padova metabolic simulator for validation, comparison with a fully-automatic zone model predictive controller and robustness analysis. Unlike offline-tuned IMC and semi-automated IMC, the online-tuned IMC in the AP system performs satisfactorily for every patient condition without patients' intervention. Experimental results show that the online-tuned IMC compensates unannounced meal disturbances with low frequency of hypoglycemic events and most importantly, with low insulin infusion even with variations in insulin sensitivity, in the presence of irregular amounts of meal disturbances at random times, and in the presence of very high noise levels in the sensors and actuators. Patients experience hypoglycemia 0.46%, 1.01% and 20% of the time using online-tuned, offline-tuned and semi-automated IMC respectively when the insulin sensitivity is increased by +20%.