Diabetes mellitus is a chronic metabolic disorder requiring meticulous blood glucose regulation to minimize both acute complications and long-term vascular damage. Traditional glucose monitoring approaches—such as finger-prick tests and continuous glucose monitoring (CGM)—primarily support reactive interventions, often falling short in enabling proactive management. This study proposes a deep learning-based predictive framework for blood glucose level estimation using historical CGM data. The model’s performance was evaluated using standard metrics including Root Mean Squared Error (RMSE), Mean Absolute Percentage Error (MAPE), and R-squared (R2) score. Experimental results across multiple patients reveal that the model achieved RMSE values ranging from 19.37 to 28.57, and MAPE values between 7.76 and 13.31. The highest predictive accuracy was observed for Patient 570 (RMSE: 20.38, MAPE: 7.76), while the model struggled with higher variability in Patient 559. These findings demonstrate the model’s potential in delivering personalized, anticipatory glycemic control, thereby supporting more effective diabetes management strategies.

This paper introduces an integrated Hardware-In-the-Loop (HIL) testing framework, combining the UVA/Padova Type 1 Diabetes Simulator with AndroidAPS, an open-source artificial pancreas system. This integration forms a testing environment capable of evaluating insulin regulation algorithms under both virtual and real hardware conditions. The FDA-approved UVA / Padova Simulator models glucose-insulin dynamics and meal digestion. Paired with AndroidAPS, the system can actuate real-world insulin pumps to test insulin delivery control algorithms. The framework is tied together by various REST APIs and uses the Flask framework for efficient data exchange and system connectivity. The HIL approach provides a robust platform for functional and reliability testing of these algorithms. The developed APIs are open-source: https://github.com/OE-Diab/AP-HIL

Physical and psychological stressors have substantial effects on metabolism, challenging treatment decisions for people with type 1 diabetes (TID). Incorporating physical activity (PA) in multivariable automated insulin delivery (mv AID) algorithms using physiological signals improves diabetes treatment decisions, but acute psychological stress (APS) in diabetes therapies has not been explicitly studied. In this work, we develop machine learning (ML) models using clinical experiment data to detect physical and psychological stressors and incorporate this information in mvAID. The ML models can detect PA and APS in independent test data with 97.8% and 96.1 % accuracy, respectively. We develop a mathematical model that characterizes the effects of PA and APS on glycemia and use it to design a predictive control algorithm to regulate blood glucose concentrations (BGC) by manipulating insulin dosing in response to detected PA and APS. In silico studies demonstrate that the mv AID informed of PA and APS results in 2.36 % improvement in time spent in the target glucose range (BGC of 70–180 mg/dL) while reducing the time spent in hypoglycemia (BGC <70 mg/dL) by 3.11 % compared to a mv AID considering PA only. The results show that insulin dosing algorithms that explicitly consider PA and APS in their decisions can improve glucose control and improve the lives of people with TID.

Activities and events during daily living can have substantial effects on blood glucose levels in patients with type 1 diabetes. We evaluate the glycemic effects of different physical and psychological stress inducements and combined activities based on clinical data. The average change in blood glucose levels were −18, −15 and −10 mg/dL for cycling, running and resistance exercise, respectively. In some scenarios, psychological stress inducement activities showed an increase in glucose levels that warrant further exploration. The developed model is able to reproduce the self-reported insulin-to-carb ratios with a Pearson correlation coefficient of 0.86. Understanding the glycemic response to physical and psychological stressors will enable more effective therapy for people with type 1 diabetes.

Reinforcement learning (RL) has shown promise in controlling blood glucose levels in a personalized way in type 1 diabetic patients. In this study, we investigate the impact of different activation functions and layer numbers on RL performance in blood glucose control. We train RL agents with various combinations of activation functions and layer numbers on a virtual patient model. The RL agents are evaluated based on their ability to maintain blood glucose levels within a target range while minimizing the frequency and magnitude of hypoglycemia and hyperglycemia events. Our results show that the choice of activation function and layer number significantly affects the RL performance. Specifically, the agents with ReLU activation functions and two or three hidden layers outperform the other agents, achieving a higher percentage of time in the target range and fewer hypoglycemia and hyperglycemia events. These findings provide valuable insights for the development of RL-based blood glucose control systems in type 1 diabetic patients.

Despite the increasing availability and reliability of artificial pancreas devices, many people with type 1 diabetes mellitus are still on multiple daily injections therapy consisting of a daily basal insulin injection and mealtime boluses. The use of an insulin bolus advisor may improve glycaemic control as well as reduce the burden of the disease on these patients. This paper investigates the application of an impulsive model predictive controller in a bolus advisor system. The bolus calculator is assessed in closed-loop simulations using different basal insulin scenarios. The simulations show that a model predictive controller can achieve good glycaemic control and may be suitable to give bolus recommendations to people with diabetes. Higher basal insulin levels can result in better times in the target range but also carry a higher risk of hypoglycemia.

Like many physiological systems, the various mathematical models describing the glucose-insulin system can only have non-negative inputs. Thus, the control method — often model predictive control in artificial pancreas systems—must provide a non-negative control signal. Existing solutions include saturation and constrained optimization. In this paper, we propose a dynamic extension of the patient model as a way of ensuring the positivity of the control signal, a method previously applied in tumor growth control. We evaluate the controller in a closed-loop simulation and compare the results with a controller using saturation. Our simulations show that control performance with the extended model can reach or exceed the performance achieved with saturation.

One of the most challenging area of diabetes research is to provide such automated insulin delivery systems – so called artificial pancreas systems – that have robust and adaptive capabilities in a highly sophysticated way. I.e. they are able to provide robust insulin delivery actions at the beginning of the therapy to satisfy the requirements of the patients without knowing the users daily lifestyle and preferences however adaptive on the short-term to learn these patient specifics to increase the quality of the therapy. One possible solution is the closed-loop systems that have self-learning features. In the present study, we have examined a glucose regulatory problem using direct reinforcement learning based controller. The approach represents the fully automatic insulin administration as the timepoint and the carbohydrate content of the meals were unknown and randomized. We constructed a virtual environment of the patient with type 1 diabetes by applying a mathematical model. Proximal policy optimization learning model with continuous action space was used. Furthermore, we evaluated the effect of different training lengths on the test scenario.

The effects of individual physiological phenomena play an important role considering the accuracy of artificial pancreas systems. An example of these phenomena is the heart rate which is easy to measure. There is a connection between heart rate and endogenous glucose production that significantly influences the blood glucose level. The proper implementation of heart rate could lead to defining physical activity in TIDM models. The aim of the current study is to examine how the change in endogenous glucose production influences the fitting accuracies in model versions with different complexity. Our extensions include a heart rate dependent endogenous glucose production equation, modeling the effect of physical activity, and a part defining the effect of insulin on endogenous glucose production. The joint effect of the mentioned extensions was also considered.

Remote photoplethysmography (RPPG) is a camera-based optical technique for detecting volumetric changes of organs. This technique enables the non-contact measurement of respiration and pulse. Monitoring newborn infants is a challenging task, due to the weak pulse signals, the rather irregular respiratory pattern, and the frequent movements. Therefore, heavy optimization of the sensing and evaluation process is a must in a resource-limited embedded vision system. This is a two-faceted study, with a special focus on low computational complexity. In the field of respiration monitoring, the paper introduces an optimized convolutional neural network (CNN) and a novel, light Long Short-term Memory (LSTM) motion classifier with a narrow CNN layer. From heart rate measurement point of view, a skin segmentation based algorithm is presented. The performance of each algorithm is evaluated on a database collected at the IstDept. of Neonatology of Pediatrics, Dept of Obstetrics and Gynecology, Semmelweis University, Budapest, Hungary.

In case of physiological systems state and parameter estimation is a crucial question. It is key to describe given patient population with appropriate accuracy. Furthermore, state feedback kind of applications also require some sort of estimation procedure in order to get internal information about the controlled system. Linear parameter varying (LPV) framework is beneficial for controller design as well. However, to realize the necessary scheduling parameters, estimation of both state variables and model parameters is needed. A possible solution is the application of Dual Extended Kalman Filter (DEKF) which is able to estimate these signals. The developed framework can be used to design LPV based controller in our further work. In this study we introduce our developed DEKF solution by using the widely applied Cambridge Type 1 Diabetes Mellitus (TIDM) model for virtual patient generation. We have found that our solution is able to estimate the state variables with good accuracy. The variation of parameters can also be tracked by using the proposed solution.

The utilization of modern and advanced control engineering related methods for the control, estimation and assessment of physiological applications is widespread. It is also well-known that this engineering apparatus is executed on digital computers. The current insufficiency of available and accurate discretized models, especially in case of Diabetes Mellitus (DM), provides incentive for this research. The researchers typically approximate the continuous solutions which may not be the best alternative in many cases, in particular considering numerical stability and cost-effectiveness. In this paper we performed an analysis of the available discretization options in order to develop discrete models with a special focus on the Linear Parameter Varying (LPV) systems. LPV techniques are very useful frameworks which allow the application of linear controller, observer and estimator design. In this study, three LPV discretization and two Jacobian based discretization methods are introduced and analyzed to provide a basis for our further investigations in the topic.