Keywords: Biological systems and controls
Abstract: The paper addresses the design of a H-infinity closed-loop dedicated to Blood Glucose (BG) regulation for patients affected by Type-1 Diabetes Mellitus (T1DM). The closed-loop setup is standard, i.e. the H-infinity feedback controller uses the information provided by a subcutaneous sensor to drive an insulin pump but as opposed to current existing solutions, it is proposed to assess the capabilities of a H-infinity controller to be designed in a patient-independent way. For that purpose, the design is performed on a family of linear models in order to tackle the variability of a cohort of T1DM patients. Worst-case performance and robust margins are next computed with the help of the H-infinity/µ-analysis theory. The solution is finally assessed on the adult cohort of the high-fidelity UVA/Padova benchmark (v3.2), accepted by the US Food and Drug Administration (FDA) as a substitute for pre-clinical testing of control strategies.

Keywords: Biomedical system modelling, Decision support systems for the control of physiological and clinical variables, Identification and validation
Abstract: Reliable prediction of future blood glucose (BG) values is of high relevance for diabetes patients, since it enables the use of predictive glucose alarms (warning the patient about impending situations with dangerously low or high BG), as well as of model-based algorithms for smart glucose control. Control-oriented graybox process models have proven very suitable for such tasks, especially when identified on data from clinical trials under well-defined conditions. The current paper analyzes how such models can also be reliably parametrized using outpatient data of patients on multiple daily injection (MDI) therapy. A dedicated preprocessing algorithm is presented to look for suitable (i.e. complete and sensible) data segments that allow for a reliable system identification. The focus of the current paper is on the prediction of postprandial glucose trajectories, more specifically on predictions made exactly at the time of meal ingestion. This corresponds to a particularly challenging task, but one with high importance for the model-based optimization of insulin doses. It is demonstrated that the identified process models are a suitable choice for predicting such postprandial glucose excursions.

Keywords: Quantification of physiological parameters for diagnosis assessment, Identification and validation, Decision support systems for the control of physiological and clinical variables
Abstract: Insulin sensitivity is an important physiological parameter for determining insulin requirements for patients with type 1 diabetes. In addition to being highly variable between patients, insulin sensitivity increases substantially during exercise and stays elevated for several hours during subsequent recovery. We propose an unscented Kalman filter for estimating insulin sensitivity from continuous glucose monitoring data that does not require the underlying model to capture exercise and relies on average values for patient-specific parameters. Using in silico full-day simulations including exercise and meals, we study how adjusting insulin doses for elevated insulin sensitivity could decrease the risk of hypoglycemia after exercise and improve time-in-range and related metrics.

Keywords: Biomedical system modelling, Decision support systems for the control of physiological and clinical variables, Healthcare management and delivery, disease control, critical care
Abstract: Non-invasive, continuous Endogenous glucose production (EGP) estimation during exercise would help manage and automate insulin and glucose dosing during exercise, providing novel information to more effectively close the loop in managing glucose levels. This study used a combination of new study and literature data to determine relationships between blood lactate concentrations, heart rate (HR), and EGP. From these relationships, EGP can be estimated based on HR, which is continuously and non-invasively available at low cost in exercise. Participants for the exercise protocol were 10 sub-elite athletes who participated in at least 6 hours per week of endurance sports. Lactate as a function of HR during high intensity (HI) exercise has variability R2 = 0.49. The variability in the model curve using independent literature values is R2 = 0.65. Using these results to create a model of EGP as a function of HR gives an R2 = 0.80. This method provides a continuous and non-invasive means of estimating EGP during exercise, including at HI, which is more rarely studied, and can be used to improve diabetes management in exercise.

Keywords: Kinetic modelling and control of biological systems, Biological systems and controls, Decision support systems for the control of physiological and clinical variables
Abstract: For people with type 1 diabetes and some with type 2 diabetes, the problem of insulin titration, i.e. finding an adequate basal rate of insulin, is a complex and time-consuming task. This paper proposes a simple model-free algorithm and a procedure for fast initial titration in people with type 1 diabetes (T1D). A modified proportional-integral-derivative (PID) controller (i) updates the estimated insulin basal rate, and (ii) administers micro-boli of insulin every 5 minutes using glucose measurements from a continuous glucose monitor (CGM). A bolus calculator mitigates the effect of meals and reduces postprandial peaks. We evaluate the performance of our system qualitatively and numerically using a virtual clinic of 1,000 T1D patients with a broad inter-patient variability representative of a real population of people with T1D. We let the titration phase run for three consecutive days, followed by a three-day test phase using the newly computed basal insulin infusion rate. The proposed algorithm is able to provide a safe titration and individualized treatment for people with T1D.

Keywords: Healthcare management and delivery, disease control, critical care, Biomedical system modelling, Decision support systems for the control of physiological and clinical variables
Abstract: Glycemic control (GC) has improved outcomes for intensive care unit (ICU) patients. However, the increased risk of hypoglycemia and glycemic variability due to inter- and intra- patient variability make safe, effective GC difficult. Stochastic TARgeted (STAR) GC framework is a unique, patient-specific, risk-based dosing protocol directly accounting for both inter- and intra- patient variability using a stochastic model of future patient variability. A new tri-variate (3D) stochastic model, developed and validated in virtual trials to provide more accurate future predictions of insulin sensitivity (SI), is clinically evaluated. STAR-3D was implemented as standard care at the Christchurch Hospital ICU, New Zealand, between April 2019 and January 2021. In total, 567 patients (33276 hours) were treated. The overall median [IQR] BG achieved was 6.7 [6.0 7.8] mmol/L with 76% BG in the 4.4-8.0 mmol/L target band. Importantly, there were only 0.3% BG < 4.0 mmol/L (mild hypoglycemia) and no incidence of severe hypoglycemia (BG < 2.2 mmol/L). These outcomes were achieved with median [IQR] 4.0 [2.0 6.0] U/h insulin and median [IQR] nutrition delivery of 99 [80 100]% goal feed (GF). Similar safety and performance BG outcomes were obtained at a per-patient level, suggesting STAR-3D successfully provided safe, effective control for all patients, regardless of patient condition. Compared to the original version of STAR, STAR-3D provided improved safety and efficacy, while achieving higher nutrition delivery. The new 3D stochastic model in STAR-3D provided higher safety and efficacy for all patients in this large clinical trial, despite using higher insulin rates than its predecessor to provide greater nutrition delivery. STAR-3D thus better captured patient-specific condition and variability to provide improved GC outcomes.

Keywords: Decision support systems and feedback control, Kinetic modelling and control of biological systems, Simulation and visualization,
Abstract: In type 2 diabetes (T2D), injections with long-acting insulin can become necessary to regulate blood glucose and avoid long-term complications. However, finding a safe and effective insulin dose, a process known as titration, is both challenging and time demanding. In this paper, we propose a new method for safe and rapid identification of a personalized insulin dose with long-acting insulin through short-term use of fast-acting insulin in an artificial pancreas (AP). To illustrate this novel concept, we simulate our method by modelling the glucose response to fast- and long-acting insulin in people with T2D. We apply a simple control-algorithm for the AP to adjust the insulin infusion rate during fasting periods. In this case-study, we simulate an insulin naïve T2D patient on AP treatment for one week, gradually adjusting the insulin infusion rate. After one week, we convert the insulin infusion rate, unit-to-unit, to a daily injection of long-acting insulin. We compare our method to titration with the standard of care 2-0-2 algorithm. Our simulations indicate that we can reduce the titration period from five weeks to a single week, whilst easing the burden on the patient.

Keywords: Healthcare management and delivery, disease control, critical care, Devices and sensors, Biological systems and controls
Abstract: Insulin pumps are the most consistent and accurate means of regulating blood glucose levels in Types 1 and 2 diabetes. However, the technology is underutilised due to very high costs. A typical insulin pump costs US6500, which makes this gold standard of care inaccessible to many, reducing equity of access to care. Since insulin pumps were first introduced, the simple hardware has not changed significantly. Pump manufacturers couple the low-cost and simple hardware with their own software, removing consumer choice and locking value into the product. Using both a traditional motor-driven and novel spring-loaded approach, insulin pump hardware can be replicated for US100. Initial testing of the traditional motor-driven prototype proves the low-cost approach has comparable accuracy to commercially available pumps, with 85.1% of basal doses delivered within 5% of target. The results obtained indicate pump hardware and software can be separated with no significant loss in accuracy. If a separate market for pump software is established, costs can be driven down through market pressure given the increasing access to relatively extensive mobile and cloud computing. If open-source software is made accessible, a complete pump could be offered for US100. A 98.5% cost reduction would drastically improve pump accessibility, particularly in developing nations, and could lead to a global improvement in diabetes treatment, outcomes and costs.

Keywords: Biomedical system modelling, Identification and validation, Quantification of physiological parameters for diagnosis assessment
Abstract: The prospect of ex-vivo functional evaluation of donor hearts is considered. Particularly, the dynamics of a synthetic cardiac afterload model are compared to those of normal physiology. A method for identification of continuous-time transfer functions from sampled data is developed and verified against results from the literature. The method relies on exact gradients and Hessians obtained through automatic differentiation. This also enables straightforward sensitivity analyses. Such analyses reveal that the 4-element Windkessel model is not practically identifiable from representative data while the 3-element model underfits the data. Direct comparison of aortic pressure-flow relations, without relying on matching of fitted Windkessel model parameters, is therefore suggested as an alternative for comparing afterload dynamics.

Keywords: Cellular, metabolic, cardiovascular, neurosystems, Quantification of physiological parameters for diagnosis assessment
Abstract: The Seidel-Herzel model of the autonomic-cardiorespiratory system is used in this work to derive three observers for the estimation of the concentrations of the neurotransmitters and external excitations of the sympathetic and parasympathetic systems. The bservers require noninvasive measurements of respiration and blood pressure which simplifies the development of patient-specific applications. The errors stemming from model approximations and uncertain parameters are analyzed. The applicability of the approach is discussed in a simulation study.

Keywords: Biomedical system modelling, Biological systems and controls, Artificial organs and biomechanical systems
Abstract: LifeTec Group has developed a Cardiac Biosimulator where a dead porcine heart is used to mimic a beating heart in a simulated environment. This can be used for assessment of medical devices or as a training platform for medical professionals. The research presented aims at extending this simulator by designing feedback controllers for the time-averaged relevant pressures and flow, which reduces the startup time and potentially increases stability of the simulator. To achieve this, both the continuous and time-averaged models of the simulator are presented together with their state-space representations. The proposed controller consists of three independent PI controllers, which are presented along with simulation and measurement results and show promising system performance. Lastly, the controller implementation was tested on the Biosimulator with a pathological heart (Mitral prolapse), which showed no significant decrease in performance compared to the physiological heart.

Keywords: Artificial intelligence support in diagnosis and decision making systems, Biosignal analysis and interpretation,, Biomedical system modelling
Abstract: Drowsy driving is a fatal problem that may cause traffic accidents. Although many driver drowsiness detection methods have been proposed, most of them have problems in input data availability, robustness to real driving environments, or detection precision. A drowsiness detection method based on heart rate variability (HRV), which is an R-R interval (RRI) fluctuation obtained from an electrocardiogram (ECG), has been proposed since ECG is easy to measure by using a wearable sensor. HRV is related to the autonomic nervous system (ANS) and is affected by drowsiness. However, its drowsiness detection performance was not always satisfactory. This study proposes a new drowsiness detection method using raw RRI data instead of HRV to improve the drowsiness detection performance. The proposed method uses raw RRI time series as inputs, and a drowsiness detection model is trained based on long short-term memory (LSTM) and autoencoder (AE), which are types of neural networks. RRI data during driving were collected from 25 participants using a driving simulator. The drowsiness detection model was trained following the proposed method. The experimental result showed that the proposed method achieved an AUC of 0.88, a sensitivity of 81%, and a specificity of 91%, which was higher than the HRV-based method. The result suggests that it is better to use raw RRIs as inputs than HRV features.

Keywords: Healthcare management and delivery, disease control, critical care, Model formulation, Biomedical system modelling
Abstract: This paper presents a set of criteria used to identify the parameters describing a tube-load model of the arterial system to estimate central blood pressure (BP). The criteria are generalizable to accommodate for inter- and intra-subject variability encountered in the ICU. The proposed single measurement transfer function (SMTF) requires only a single peripheral pressure measurement, commonly available in the ICU, for central pressure estimation, removing the need for an additional measurement of pulse transit time, common to other central BP estimation models. The method was tested using data from six (6) porcine experiments where septic shock was induced and subsequent treatment was performed. Systolic pressure (SP), pulse pressure (PP) and root-mean-squared (RMSE) errors relative to invasive measurements of aortic pressure were used to assess accuracy. The SMTF method produced mean errors <5mmHg across all metrics with 84.4, 88.7 and 63.2% of SP, PP, and RMSE, respectively, within this bound of measured central pressure. Peripheral BP accuracy was also assessed as it is commonly used as a surrogate for central BP in clinical settings. Finally, two alternate methods utilizing the same model equations with additional inputs, one using the measured pulse transit time and the other minimizing the RMSE with measured aortic pressure, were implemented to compare the SMTF accuracy to best case outputs, given the model equations.

Keywords: Biomedical system modelling, Identification and validation, Model formulation
Abstract: A fractional-order vascular model representation for emulating arterial hemodynamics has been recently presented as an alternative to the well-known integer-order arterial Windkessel. The model uses a fractional-order capacitor (FOC) to describe the complex and frequency-dependent arterial compliance. This paper presents a two-stage algorithm based on modulating functions for finite-time simultaneous estimation of the model's input and the fractional differentiation order. The proposed approach is validated using in-silico human data. Results show the prominent potential of this method for calibrating arterial models and enhancing cardiovascular mechanics research as well as clinical practice.

Keywords: On demand product manufacturing and quality control, Artificial intelligence for decision support systems, Devices and sensors
Abstract: Pill defects encountered during the manufacturing process may cause in low quality product and high timeline delays, and costs. In this paper, an improved convolutional neural network is proposed for automatic pill defects detection during pill manufacturing. In the first step, Gauss filtering and smoothing techniques is implemented for complex background- weakening purpose. Then, Hog feature extraction is executed to simplify the representation of the image that contains only the most important information about the image. The aim of this sub-process is to reduce the computation burden. Lastly, an improved YOLO model is proposed for online detection of pill defects and it was validated on our experiment platform in the laboratory for online pill defect detection. The proposed approach obtains robust quantification of internal pill cracks. This proposed approach is effective tool implemented into the industrial pill manufacturing system.

Keywords: Biomedical system modelling, Biological systems and controls, Model formulation
Abstract: This work presents a global sensitivity analysis and simulations of a perfusion bioreactor process using the method of high-dimensional model representation (HDMR). This method was developed to express the input–output relationships of a complex model with a high dimensional input space. The comprehensive mathematical model of convection and diffusion in a perfusion bioreactor, combined with cell growth kinetics, is developed and implemented using Computational Fluid Dynamics (with the commercial software COMSOL Multiphysics v5.5). The model describes the spatio-temporal evolution of glucose concentration, oxygen concentration, lactate concentration and cell density within a 3D polymeric scaffold. A quantitative analysis of the complex kinetic mechanisms using recent development of advanced mathematical approaches to global sensitivity and uncertainty analysis through HDMR can be exploited to investigate the important features of the perfusion bioreactor process as well as possible factors underlying qualitative discrepancies.

Keywords: Identification and validation, Biosignal analysis and interpretation,, Biological systems and controls
Abstract: This paper presents a parameter free approach to identify the importance of reactions (and involved species) in biochemical reaction networks for the purpose of model order reduction. The new methodology is based on the structure of the network assuming that at steady state the flux distribution is preserved. Our method employs Flux Balance Analysis (FBA) to calculate these flux distributions. The ranking is based on the comparison of the FBA results of the original and reduced networks. The main purpose of this identification step is to guide the selection of species that could be deleted by model order reduction methods. Our ranking method is illustrated in a model for Glycolysis, where model reduction is performed via the Kron reduction method considering the elimination of the less sensitive species proposed by the ranking method. The reduced model shows that the global dynamics could be accurately reproduced with a smaller number of species.

Keywords: Pharmaceutical product on demand, design, manufacturing and quality assessment
Abstract: Despite manufacturing innovations and the technologies on the rise, solid oral dosage in the pharmaceutical industry is still mass production. Although this is efficient and costeffective, it is typically based on a ‘one-size-fits-all’ product concept and lacks the flexibility and agility required to fully meet the needs of the individual patient. Nowadays pharmaceutical industry is experiencing a paradigm shift from batch to continuous manufacturing. This will lead to increased flexibility to target diverse populations as well as more consistent product quality to ensure best efficacy. Continuous processing integrated with online/inline monitoring tools coupled with an efficient automatic feedback control system is highly desired by the pharmaceutical industry. To facilitate the transition from the batch wise production to continuous manufacturing in the pharma industry engineering tools are needed. Hence, the aim of this paper is to enhance the advantage of modeling and control techniques in the field of pharmaceutical applications.

Keywords: Biological systems and controls, Pharmacokinetics and drug delivery, Biomedical system modelling
Abstract: This paper is aimed to develop and test one novel and unexplored enhancement of the classical model reduction method applied to a class of biochemical networks. Both methods, being (i) the standard quasi-steady-state approximation (QSSA), and (ii) the so-called delayed-QSSA methods are extensively presented. Specially, the numerical issues related to the setting of constant delays are discussed. Finally, for one slightly modified version of an enzyme-substrate reaction network (Michaelis-Menten kinetics), the comparison of the full non-reduced system behavior with respective variants of reduced model is presented and future prospects are proposed.

Keywords: Kinetic modelling and control of biological systems, Biosignal analysis and interpretation,, Biological systems and controls
Abstract: Inside living cells, proteins or mRNA can show oscillations even without a periodic driving force. Such genetic oscillations are precise timekeepers for cell-cycle regulations, pattern formation during embryonic development in higher animals, and daily cycle maintenance in most organisms. The synchronization between oscillations in adjacent cells happens via intercellular coupling, which is essential for periodic segmentation formation in vertebrates and other biological processes. While molecular mechanisms of generating sustained oscillations are quite well understood, how do simple intercellular coupling produces robust synchronizations are still poorly understood? To address this question, we investigate two models of coupled gene oscillators - activator-based coupled oscillators (ACO) and repressor-based coupled oscillators (RCO) models. In our study, a single autonomous oscillator (that operates in a single cell) is based on a negative feedback, which is delayed by intracellular dynamics of an intermediate species. For the ACO model (RCO), the repressor protein of one cell activates (represses) the production of another protein in the neighbouring cell after a intercellular time delay. We investigate the coupled models in the presence of intrinsic noise due to the inherent stochasticity of the biochemical reactions. We analyze the collective oscillations from stochastic trajectories in the presence and absence of explicit coupling delay and make careful comparison between two models. Our results show no clear synchronizations in the ACO model when the coupling time delay is zero. However, a non-zero coupling delay can lead to anti-phase synchronizations in ACO. Interestingly, the RCO model shows robust in-phase synchronizations in the presence and absence of the coupling time delay. Our results suggest that the naturally occurring intercellular couplings might be based on repression rather than activation where in-phase synchronization is crucial.