Closed-Loop Glucose Control Using Multi-Model Predictive Dosing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current artificial pancreas systems lack effective control methods to manage abrupt and slow variations in glucose dynamics, relying on limited control algorithms that struggle with rapid changes and do not adequately account for insulin dynamics in the human body.
Innovation Solution
A closed-loop system incorporating a Multi-Model Predictive Controller (MMPC) algorithm that uses multiple state vectors and models to predict and adjust insulin delivery, integrating data from continuous glucose monitors and user input to optimize basal insulin deviation and glucagon dosing for improved glycemic control.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If conventional control algorithms are used in artificial pancreas systems, then the system structure remains simple, but the system cannot effectively handle abrupt and slow variations in glucose dynamics
Solution Approach 1:
The patent implements a dynamic model of insulin action in the human body that adapts to both abrupt and slow variations in glucose dynamics. The controller continuously updates insulin sensitivity estimates and adjusts control parameters in real-time, transforming the static control algorithm into a dynamic system that responds to changing physiological conditions.
Solution Approach 2:
The control algorithm is segmented into multiple functional components: a glucose prediction module, an insulin action model, a control parameter optimization module, and a safety monitoring module. This segmentation allows each component to specialize in handling specific aspects of glucose dynamics, improving overall adaptability while maintaining manageable complexity through modular design.
2Adaptability or versatility
If a single fixed insulin model is used, then the control algorithm is simple to implement, but it cannot adequately control glucose concentrations when glucose dynamics are rapidly changing
Solution Approach 1:
The patent replaces the static single fixed insulin model with a dynamic insulin action model that continuously adapts to changing glucose dynamics. The model estimates insulin sensitivity in real-time and adjusts its parameters based on observed glucose responses, enabling it to effectively control glucose concentrations during rapid changes while maintaining a computationally efficient structure.
Solution Approach 2:
The insulin model parameters are changed dynamically based on physiological conditions. The controller estimates insulin sensitivity as a time-varying parameter and adjusts the model parameters accordingly, allowing the same model structure to adapt to both rapid and slow glucose dynamics changes without requiring multiple complex models.
3Adaptability or versatility
If non-adaptive controllers are used, then the control system is stable and easy to implement, but it is restricted to operating within a small range of variation
Solution Approach 1:
The patent transforms the non-adaptive controller into an adaptive system that continuously monitors glucose dynamics and adjusts control parameters accordingly. The controller estimates insulin sensitivity and other physiological parameters in real-time, expanding the operating range from a small fixed range to a broad adaptive range that covers various physiological states and glucose dynamics patterns.
Solution Approach 2:
The controller implements continuous feedback by monitoring glucose measurements and comparing them with predicted values. Based on the prediction errors and observed glucose responses, the controller updates its estimates of insulin sensitivity and adjusts control parameters, enabling operation across a wide range of variations while maintaining stability through controlled adaptation.
Data Source
Figure 1
Figure 1A
Figure 2
AI summary
The present disclosure relates to a system to control glycemia in a patient. In one arrangement, the system comprises a user interface for inputting patient data including meal data. A controller receives the patient data, defines a state vector and an associated model, propagates the state vector, corrects the propagated state vector, determines a dose request based, at least in part, on the corrected-state vector, and transmits the dose request.