Issues in the design of a multirate model-based controller for a nonlinear drug infusion system

Trans-aortic Valvular Ejection Fraction for Monitoring Recovery of Patients with Ventricular Systolic Heart Failure

Durable mechanical circulatory support in the form of left ventricular (LV) assist device (LVAD) therapy is increasingly considered in the context of the recovery of native cardiac function. Progressive improvement in LV function may facilitate LVAD explantation and a resultant reduction in device-related risk. However, ascertaining LV recovery remains a challenge. In this study, we investigated the use of trans-aortic valvular flow rate and trans-LVAD flow rate to assess native LV systolic function using a well-established lumped parameter model of the mechanically assisted LV with pre-existing systolic dysfunction. Trans-aortic valvular ejection fraction (TAVEF) was specifically found to characterize the preload-independent contractility of the LV. It demonstrated excellent sensitivity to simulated pharmacodynamic stress tests and volume infusion tests. TAVEF may prove to be useful in the ascertainment of LV recovery in LVAD-supported LVs with pre-existing LV systolic dysfunction.

Nonlinear System Identification Based on Convolutional Neural Networks for Multiple Drug Interactions

In heart failure patients, hemodynamics can be regulated by therapeutic drugs. Although the cardiovascular responses to these drugs usually include nonlinearity and drug interactions, it is difficult to identify the characteristics of the dynamics under such conditions. This study, therefore, was aimed at evaluating the technique used for nonlinear system identification based on convolutional neural networks (CNN). As an image i.e., pixel values corresponding to time-course data), CNN can be used to treat the complicated relation between previous inputs (i.e., drug infusions) and outputs i.e., hemodynamics). To compare the accuracy of CNN, traditional methods based on the standard neural networks (NN) and fast Fourier transformation FFT were applied to nonlinear system identification with drug interactions. The cardiac output and arterial blood pressure under heart failure were modulated by the drug infusions of an inotropic agent and a vasodilator. CNN accurately predicted the dynamic system responses regardless of the inclusion of nonlinearity and drug interactions. Based on the findings of this study, CNN to carry out nonlinear system identification could clarify complicated pharmacodynamics, and thus could be useful for in appropriate cardiac treatment with multiple therapeutic agents.

A Modeling Tool to Study the Combined Effects of Drug Administration and Lvad Assistance in Pathophysiological Circulatory Conditions

The aim of this work is to develop a tool to study the effect of sodium nitroprusside (SNP) on hemodynamics in conjunction with baroreflex and mechanical circulatory assistance. To this aim, a numerical model of the pharmacodynamic effect of SNP was developed and inserted into a cardiovascular circulatory model integrated with baroreflex and LVAD (continuous flow pump with atrio-aortic connection) sub-models.

The experiments were carried out in two steps. In the first step the model was verified comparing simulations with experimental data acquired from mongrel dogs on mean arterial pressure (MAP), cardiac output (CO), heart rate (HR), peripheral resistance, and left ventricular properties.

In the second step, the combined action of SNP and mechanical circulatory assistance was studied. Data were measured at pump off and at pump on (20000 rpm and 24000 rpm). At pump off, with a 2.5 μg/kg per min SNP infusion in heart failure condition, the MAP was reduced by approximately 8%, CO and HR increased by about 16% and 18%, respectively. In contrast, during assistance (24000 rpm) the changes in MAP, CO and HR were around −9%, +12%, and +20%, respectively. Furthermore, the effects of the drug on hemodynamic parameters at different heart conditions were significantly different. Thus, the model provides insight into the complex interactions between baroreflex, drug infusion, and LVAD and could be a support for clinical decision-making in cardiovascular pathologies.

Design and in silico evaluation of an intraperitoneal-subcutaneous (IP-SC) artificial pancreas

Prandial glucose regulation is a major challenge for the artificial pancreas using subcutaneous insulin (without a feedforward bolus) due to insulin's slow absorption-peak (50-60 min). Intraperitoneal insulin, with a fast absorption peak (20-25 min), has been suggested as an alternative to address these limitations. An artificial pancreas using intraperitoneal insulin was designed and evaluated on 100 in silico subjects and compared with two designs using subcutaneous insulin with and without a feedforward bolus, following the three meal (40-70 g-carbohydrates) evaluation protocol. The design using intraperitoneal insulin resulted in a significantly lower postprandial blood glucose peak (38 mg/dL) and longer time in the clinically accepted region (13%) compared to the design using subcutaneous insulin without a feedforward bolus and comparable results to a subcutaneous feedforward bolus design. This superior regulation with minimal user interaction may improve the quality of life for people with type 1 diabetes mellitus.

Automated Cardiac Drug Infusion System Using Adaptive Fuzzy Neural Networks Controller

This paper presents a fuzzy neural network (FNN) control system to automatically manage the hemodynamic variables of patients with hypertension and congestive heart failure (CHF) via simultaneous infusion of cardiac drugs such as vasodilators and inotropic agents. The developed system includes two FNN sub-controllers for regulating cardiac output (CO) and mean arterial pressure (MAP) by cardiac drugs, considering interactive pharmacological effects. The adaptive FNN controller was tested and evaluated on a cardiovascular model. Six short-term therapy conditions of hypertension and CHF are presented under different sensitivities of a vasodilator drug. The results of the automated system showed that root mean square errors were ≤ 5.56 mmHg and ≤ 0.22 L min-1 for regulating MAP and CO, respectively, providing short settling time responses of MAP (≤ 10.9 min) and CO (≤ 8.22 min) in all therapy conditions. The proposed FNN control scheme can significantly improve the performance of cardiac drug infusion System.

Systems engineering medicine: engineering the inflammation response to infectious and traumatic challenges

The complexity of the systemic inflammatory response and the lack of a treatment breakthrough in the treatment of pathogenic infection demand that advanced tools be brought to bear in the treatment of severe sepsis and trauma. Systems medicine, the translational science counterpart to basic science's systems biology, is the interface at which these tools may be constructed. Rapid initial strides in improving sepsis treatment are possible through the use of phenomenological modelling and optimization tools for process understanding and device design. Higher impact, and more generalizable, treatment designs are based on mechanistic understanding developed through the use of physiologically based models, characterization of population variability, and the use of control-theoretic systems engineering concepts. In this review we introduce acute inflammation and sepsis as an example of just one area that is currently underserved by the systems medicine community, and, therefore, an area in which contributions of all types can be made.

The next generation of artificial pancreas control algorithms

Creating a wearable artificial pancreas (AP) by closing the loop between a glucose sensor and an insulin infusion pump has the potential to significantly impact the complications associated with and improve the quality of life of diabetic individuals. Despite recent progress on glucose sensor and insulin infusion technologies, control algorithms built on the simple glucose value efferent and insulin dose afferent model are not efficient and reliable. Based on glucose regulatory mechanisms known to date, their impairment in the diabetic state, and fundamental principles of control theory, some corrections to the present course of research are proposed to facilitate the removal of this barrier. A greater emphasis on model predictive controllers or controllers that exploit a mathematical representation, or model, of the patient's own physiology is proposed. Whole-body physiologically based pharmacokinetics-pharmacodynamics-type models hold the best odds for enabling a successful closed-loop AP. However, two major improvements to the diabetes modeling state of the art are required to make them practical for daily care: integrating hypothalamus-pituitary-adrenal axis and gastrointestinal tract submodels. Although there are simple representations of these in current existence, large concerted efforts between experimentalists and modelers will be required to enhance their accuracy. Finally, changes in hardware that complements controller performance are suggested. For instance, the development of dual control inputs of insulin and glucagon could relax tolerances on controller accuracy.

Automatic regulation of hemodynamic variables in acute heart failure by a multiple adaptive predictive controller based on neural networks

Automated drug-delivery systems that can tolerate various responses to therapeutic agents have been required to control hemodynamic variables with heart failure. This study is intended to evaluate the control performance of a multiple adaptive predictive control based on neural networks (MAPC_NN) to regulate the unexpected responses to therapeutic agents of cardiac output (CO) and mean arterial pressure (MAP) in cases of heart failure. The NN components in the MAPC_NN learned nonlinear responses of CO and MAP determined by hemodynamics of dogs with heart failure. The MAPC_NN performed ideal control against unexpected (1) drug interactions, (2) acute disturbances, and (3) time-variant responses of hemodynamics [average errors between setpoints (+35 ml kg⁻¹ min⁻¹ in CO and ±0 mmHg in MAP) and observed responses; 6.4, 3.7, and 4.2 ml kg⁻¹ min⁻¹ in CO and 1.6, 1.4, and 2.7 mmHg (10.5, 20.8, and 15.3 mmHg without a vasodilator) in MAP] during 120-min closed-loop control. The MAPC_NN could also regulate the hemodynamics in actual heart failure of a dog. Robust regulation of hemodynamics by the MAPC_NN was attributable to the ability of on-line adaptation to adopt various responses and predictive control using the NN. Results demonstrate the feasibility of applying the MAPC_NN using a simple NN to clinical situations.

Modeling, identification and nonlinear model predictive control of type I diabetic patient

Patients with type I diabetes nearly always need therapy with insulin. The most desirable treatment would be to mimic the operation of a normal pancreas. In this work a patient affected with this pathology is modeled and identified with a neural network, and a control strategy known as Nonlinear Model Predictive Control is evaluated as an approach to command an insulin pump using the subcutaneous route. A method for dealing with the problems related with the multiple insulin injections simulation and a multilayer neural network identification of the patient model is presented. The controller performance of the proposed strategy is tested under charge and reference disturbances (setpoint). Simulating an initial blood glucose concentration of 250 mg/dl a stable value of 97.0 mg/dl was reached, with a minimum level of 76.1 mg/dl. The results of a simulated 50 g oral glucose tolerance test show a maximum glucose concentration of 142.6 mg/dl with an undershoot of 76.0 mg/dl. According to the simulation results, stable close-loop control is achieved and physiological levels are reached with reasonable delays, avoiding the undesirable low glucose levels. Further studies are needed in order to deal with noise and robustness aspects, issues which are out of the scope of this work.

A critical assessment of algorithms and challenges in the development of a closed-loop artificial pancreas

The development of an artificial pancreas is placed in the context of the history of the field of feedback control systems, beginning with the water clock of ancient Greece, and including a discussion of current efforts in the control of complex systems. The first generation of artificial pancreas devices included two manipulated variables (insulin and glucose infusion) and nonlinear functions of the error (difference between desired and measured glucose concentration) to minimize hyperglycemia while avoiding hypoglycemia. Dynamic lags between insulin infusion and glucose measurement were relatively small for these intravenous-based systems. Advances in continuous glucose sensing, fast-acting insulin analogs, and a mature insulin pump market bring us close to commercial realization of a closed-loop artificial pancreas. Model predictive control is discussed in-depth as an approach that is well suited for a closed-loop artificial pancreas. A major challenge that remains is handling an unknown glucose disturbance (meal), and an approach is proposed to base a current insulin infusion action on the predicted effect of a meal on future glucose values. Better "meal models" are needed, as a limited knowledge of the effect of a meal on the future glucose values limits the performance of any control algorithm.

A model-based algorithm for blood glucose control in type I diabetic patients

A model-based predictive control algorithm is developed to maintain normoglycemia in the Type I diabetic patient using a closed-loop insulin infusion pump. Utilizing compartmental modeling techniques, a fundamental model of the diabetic patient is constructed. The resulting nineteenth-order nonlinear pharmacokinetic-pharmacodynamic representation is used in controller synthesis. Linear identification of an input-output model from noisy patient data is performed by filtering the impulse-response coefficients via projection onto the Laguerre basis. A linear model predictive controller is developed using the identified step response model. Controller performance for unmeasured disturbance rejection (50 g oral glucose tolerance test) is examined. Glucose setpoint tracking performance is improved by designing a second controller which substitutes a more detailed internal model including state-estimation and a Kalman filter for the input-output representation. The state-estimating controller maintains glucose within 15 mg/dl of the setpoint in the presence of measurement noise. Under noise-free conditions, the model-based predictive controller using state estimation outperforms an internal model controller from literature (49.4% reduction in undershoot and 45.7% reduction in settling time). These results demonstrate the potential use of predictive algorithms for blood glucose control in an insulin infusion pump.