An Actor-Critic based controller for glucose regulation in type 1 diabetes.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Endocrinology

Background:

• Type 1 diabetes requires continuous glucose monitoring and insulin management.
• Sensor-augmented pump therapy aims to improve glycemic control but faces challenges with mealtime uncertainties.
• Adaptive control strategies are needed for personalized insulin delivery.

Purpose of the Study:

• To propose and evaluate a novel adaptive glucose control system for type 1 diabetes.
• To investigate the effectiveness of an Actor-Critic (AC) learning-based controller for simultaneous basal and bolus insulin adjustment.
• To assess the controller's performance in silico across different age groups (adults, adolescents, children) under varying meal conditions.

Main Methods:

• Development of an AC learning-based controller integrating reinforcement learning and optimal control principles.
• Controller features include simultaneous basal rate and bolus dose adjustment, clinical initialization, and real-time personalization.
• In silico simulations were conducted using open-loop and closed-loop approaches with meal carbohydrate uncertainties (±25%).

Main Results:

• The AC-based controller demonstrated efficient glucose regulation in adults, adolescents, and children.
• Control Variability Grid Analysis (CVGA) showed 100% of data in A+B zones for adults and 93% for adolescents and children.
• Effective glycemic control was maintained despite uncertainties in meal carbohydrate estimation.

Conclusions:

• The proposed AC-based adaptive controller is a promising approach for improving automated insulin delivery in type 1 diabetes.
• The controller's ability to handle meal uncertainties and personalize insulin dosing suggests potential for enhanced glycemic management.
• Further optimization and clinical trials are planned to validate the controller's efficacy in real-world settings.