Modeling, identification and nonlinear model predictive control of type I diabetic patient

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Endocrinology

Background:

• Type I diabetes management necessitates insulin therapy, ideally mimicking a healthy pancreas.
• Current insulin delivery methods face challenges in precise glucose regulation.

Purpose of the Study:

• To develop and evaluate a Nonlinear Model Predictive Control (NMPC) strategy for artificial pancreas systems.
• To utilize neural networks for patient-specific modeling and insulin pump control.

Main Methods:

• A multilayer neural network was employed for patient model identification.
• Nonlinear Model Predictive Control (NMPC) was implemented to command an insulin pump via subcutaneous route.
• Simulations addressed challenges in multiple insulin injection modeling.

Main Results:

• Simulated blood glucose stabilized at 97.0 mg/dl from an initial 250 mg/dl, with a minimum of 76.1 mg/dl.
• A 50g oral glucose tolerance test resulted in a peak of 142.6 mg/dl and undershoot of 76.0 mg/dl.
• The NMPC strategy demonstrated stable closed-loop control, achieving physiological glucose levels with minimal delay and avoiding hypoglycemia.

Conclusions:

• The proposed NMPC strategy, combined with neural network patient modeling, shows promise for artificial pancreas systems.
• The simulation results indicate effective glucose regulation, approaching physiological norms.
• Further research is recommended to address noise and robustness for clinical application.