Page 77 - Control Theory in Biomedical Engineering

64 Control theory in biomedical engineering

body, and a powerful control algorithm are required to develop a safe and
reliable AP system. The AP system, tested in many simulation and clinical
studies, can reduce the risks of immediate life-threatening conditions, such
as severe hypoglycemia and ketoacidosis, and long-term health complica-
tions, such as cardiovascular disease, nephropathy, neuropathy, and retinop-
athy (Turksoy et al., 2017; Peyser et al., 2014; Esposito et al., 2018; Bekiari
et al., 2018).
Complex nonlinear dynamical systems such as the metabolic processes in
the human body are particularly challenging to model and control due to
time-varying characteristics, nonlinear behavior, presence of stochastic
and unknown disturbances, uncertain time-varying delays, and variations
between and within peoples’ metabolic activities. The glycemic models pro-
posed in the literature can be categorized as physiological and data-driven
models (Silvia et al., 2017). Physiological models consist of simultaneous
differential equations describing the insulin and glucose interactions based
on mass exchange between compartments representing various organs of
depots. Data-driven models with relatively simpler structures that can char-
acterize the relationships among the measured variables generally need less
computational load. Subspace-based system identification methods can
readily identify linear state-space models from multiinput, multioutput sam-
pled data of a dynamic system. However, physiological and data-driven
models with fixed parameters cannot accurately describe the dynamic
behavior of BGC variations in the human body over wide ranges of real-
world conditions. Therefore, the models need to be appropriately adapted
online to characterize the current dynamics of the individuals and make
accurate short-term predictions of glucose concentration measurements.
For this purpose, adaptive system identification approaches are proposed
to determine linear, time-varying models and effectively characterize the
evolving glycemic dynamics, thus allowing the adaptive models to be valid
over a diverse range of daily conditions. In our previous work, an adaptive-
personalized modeling approach considering the effects of unannounced
meals and exercise on transient glycemic dynamics was proposed and applied
to 15 clinical data sets involving closed-loop experiments of the AP systems
(Hajizadeh et al., 2017b, 2018a, b, d).
To minimize the risk of hypoglycemia, AP systems need safety con-
straints to avoid insulin overdosing. Quantifying the amount of active insu-
lin present in the body is difficult due to lack of sensors to measure plasma
insulin concentration (PIC) in the bloodstream. However, accurate PIC
estimates can be obtained by using CGM measurements and infused insulin