Suboptimal H_∞ hyperglycemia control on T1DM accounting biosignals of exercise and nocturnal hypoglycemia

Abstract

The availability of novel technologies for exogenous insulin release and continuous glucose monitoring have increased the possibilities of developing an artificial pancreas. This contribution tackles the design of a H ∞ -based controller to manage glycemia in Type 1 diabetes mellitus (T1DM) under two scenarios: exercise and nocturnal hypoglycemia. Two biosignals are integrated to the blood glucose control problem: lactate is used in exercise scenario, while adrenaline releasing is used for nocturnal hypoglycemia. The effects of each scenario are represented by weighting transfer functions at the control design. Each weighting function accounts the effect of hepatic glucose production and defines separately the following relations: (a) from plasmatic glucose to lactate during exercise and (b) from plasmatic glucose to adrenaline during nocturnal hypoglycemia. Also, the controller is designed by adding a frequency restriction in control signal to incorporate frequency components by the pancreatic insulin release pattern in a healthy subject. A nonlinear physiological model, including Glucose-Insulin-Glucagon dynamics and counter-regulatory effects, is used to show the time-response of the closed-loop including actuator dynamics and parametric changes. levels due to exogenous events (as, among others, different kinds of carbohydrate load during meal, glucose uptake during exercise) or endogenous events (as changes in hepatic or renal glucose production and uptake). From the control theory viewpoint, the exogenous events can be formulated as disturbances while the endogenous ones can be seen as uncertain parameter changes. Thus, the controlled plant corresponding to the diabetic patient is a nonlinear dynamical system with parametric uncertainties and uncertain disturbance inputs. In this sense, although control design is not an easy task, this problem has been addressed mainly via four control approaches with promissory results: model-based predictive control [10], adaptable techniques [11], fuzzy logic [12], and H ∞ robust control [13,14]. As a matter of fact, there are recent promising in silico and in vivo clinical trials of closed-loop glucose control (mainly classic PI and PID schemes [15], as well as advanced techniques such as nonlinear model predictive control [16][17][18][19]) using subcutaneous insulin delivery and subcutaneous continuous glucose monitoring. Such proofs have highlighted promising results using feedback control schemes in glucose control. Although classical PI or PID schemes have used in clinical devices since 70s, it is known that robustness margin of classical controllers can be improved.Respect to H ∞ approaches, the synthesis for blood glucose control is based on the plant frequency response in such manner that the control design ensures closed-loop performance in the frequency interval where plant is sensitive. The controller captures the frequency components and is capable to ensure closed-loop stability in an optimal or suboptimal sense. Moreover, as parametric uncertainties are incl...

Integrating Multiple Inputs Into an Artificial Pancreas System: Narrative Literature Review

“…MATLAB was used in most of the studies to conduct simulations. Quiroz et al [46,47] simulated the use of invasive inputs based on the Sorenson model [21], the Bergman minimal model [65], the glucose-adrenaline relationship discussed in the study by Schultes et al [66], and the glucose-lactate relationship discussed in the study by Stuart et al [67]. Khan et al [48] and Qaisar et al [49] also used the Bergman minimal model, as well as simulated meals, ECG, and subcutaneous delays.…”

“…A quality assessment of the selected 17 studies were carried out using the Critical Appraisal Skills Programme Tool [35] (Multimedia Appendix 2 [36][37][38][39][40][41][42][43][44][45][46][47][48][49][50][51][52]). It is important to note that the main issues highlighted by the assessment were (1) difficulty in ascertaining the risk of bias in data collection or simulation data and ( 2) the failure of the study reports to provide sufficient information regarding ethical approvals.…”

“…Quiroz and Femat [46] and Quiroz et al [47] identified lactate and adrenaline as 2 important invasive inputs that were directly integrated as inputs in the controller. They are used in detecting exercise and hypoglycemia, respectively, which are important aspects in APS design to address the limitations identified previously.…”

“…They have been (1) used to switch controller modes through activity detection, (2) directly incorporated in the controller for decision-making, and (3) used for the development of intermediate modules for the controller (eg, hypoglycemia and meal detection). [47] MATLAB simulation Fuzzy fusion controller to fuse the additional input to prompt glucagon infusion (dual hormone) PID c controller ECG a , HR b , and skin resistance Khan et al [48] MATLAB simulation Fuzzy fusion controller to fuse the additional input to prompt glucagon infusion (dual hormone)…”

“…Quiroz and Femat [46], Quiroz et al [47], Resalat et al [43], Turksoy et al [36,37], and Hajizadeh et al [38] focused on direct integration of additional inputs in the controller design. Quiroz and Femat [46] and Quiroz et al [47] directly integrated lactate and adrenaline input into their H ∞ control algorithm. Resalat et al [43] developed a run-to-run MPC that used continuous MET data for exercise detection.…”