A Novel Controller Architecture for Intelligent Artificial Pancreas

The extensive research conducted in the past decades has led to the development of safe and reliable Artificial Pancreas (AP). However, there are still several challenges hindering successful fully automated insulin delivery such as maintaining time-in-range (TIR) above 90% throughout treatment and continuously administering exogenous insulin to patients with severe insulin resistance. The traditional strategy has been to implement an algorithm capable of handling the full complexity of the insulin-glucose physiology (IGP) while keeping blood glucose (BG) within the desired TIR. This paper proposes a novel architecture that adopts independent agents specialized in the control of each critical compartment of the IGP and a Decision Engine (DE) that coordinates the agents to achieve near-optimal performance. The approach allows for the development of treatments tailored to the patient's physiology through custom tuning based on patient-specific metabolic parameters. The preliminary results led to superior TIR, predictive carbohydrate (CHO) and insulin behavior, and greater system robustness.