Embedded Model Predictive Control for a Wearable Artificial Pancreas.

Area of Science:

• Biomedical Engineering
• Control Systems Engineering
• Endocrinology

Background:

• Artificial pancreas (AP) systems aim to improve quality of life for individuals with type 1 diabetes mellitus (T1DM).
• Designing convenient, user-friendly AP systems, particularly for active populations like children and adolescents, remains a challenge.
• Existing AP systems can be bulky, impacting user experience and daily activities.

Purpose of the Study:

• To introduce a novel, embeddable design and implementation of model predictive control (MPC) for AP systems in T1DM.
• To significantly reduce the weight and on-body footprint of AP systems.
• To evaluate the performance of the embedded AP system, especially for users with active lifestyles.

Main Methods:

• Development of an embedded zone MPC controller with a simplified safe zone design in the cost function.
• Utilization of alternating minimization algorithms for solving convex programming problems in MPC.
• Offline closed-loop simulations using the FDA-accepted UVA/Padova simulator for algorithm selection and tuning.
• Hardware-in-the-loop (HIL) in silico testing on a resource-constrained Arduino Zero (Feather M0) platform.

Main Results:

• The embedded zone MPC achieved comparable performance to a full-version zone MPC on a desktop system.
• Median time in the euglycemic range ([70, 180] mg/dL) was 84.3% (embedded) vs. 83.1% (full) for announced meals (p=0.0013).
• Median time in the euglycemic range was 66.2% (embedded) vs. 66.0% (full) for unannounced meals (p=0.0028).

Conclusions:

• The proposed embedded zone MPC demonstrates the potential for effective AP control on limited-resource platforms.
• This compact design offers a viable solution for improving AP system convenience and user experience, especially for active individuals.
• The embedded system achieves comparable glycemic control to larger, more resource-intensive AP controllers.