Efficient Closed Loop Simulation of Do-It-Yourself Artificial Pancreas Systems

Area of Science:

• Biomedical Engineering
• Computational Biology
• Diabetes Technology

Background:

• In silico trials are crucial for artificial pancreas (AP) system approval.
• The UVA/Padova Type 1 Diabetes Metabolic Simulator (T1DMS) is a key simulation tool.
• Previous simulations of the AndroidAPS system using T1DMS were limited by slow execution speed.

Purpose of the Study:

• To significantly accelerate the simulation speed of the AndroidAPS system.
• To achieve this acceleration by reimplementing the AndroidAPS algorithm in MATLAB/Simulink and integrating it with T1DMS.

Main Methods:

• Re-implementation of the AndroidAPS algorithm in MATLAB/Simulink and subsequent verification.
• Integration of the new MATLAB/Simulink implementation into the T1DMS environment.
• Evaluation using a 2-day real-time scenario with 30 virtual patients, comparing results to existing literature.

Main Results:

• Unit and integration tests confirmed the equivalence of the new and original AndroidAPS code.
• The new implementation achieved a speed-up factor of approximately 1000x compared to real time.
• Simulation results closely matched previous findings, with minor discrepancies attributed to different virtual populations and parameter harmonization.

Conclusions:

• The MATLAB/Simulink implementation drastically reduces runtime for in silico trials of AndroidAPS.
• This offers a cost-effective and rapid method for testing new AndroidAPS algorithm versions.
• Facilitates more extensive testing and validation before community release.