Miniaturized Neural Networks for Deploying Fully Closed Loop Insulin Delivery Systems: A Pilot Study Featuring...

Area of Science:

• Biomedical Engineering
• Endocrinology
• Diabetes Technology

Background:

• Automated insulin delivery (AID) systems are advancing towards fully closed-loop (FCL) control to minimize user intervention.
• Next-generation AID systems aim to reduce mealtime dosing burden through enhanced automation.
• Assessing the safety and feasibility of novel FCL control strategies is crucial for diabetes management.

Purpose of the Study:

• To evaluate the safety and feasibility of a next-generation FCL AID system (AIDANET) utilizing a miniature neural network.
• To assess glycemic control and user input reduction in free-living conditions.
• To test hybrid closed-loop modalities, including carbohydrate counting and an easy-bolus strategy.

Main Methods:

• A randomized crossover trial involving six adults with type 1 diabetes.
• Participants completed seven days of usual care and seven days using the AIDANET system.
• AIDANET was tested in FCL mode, with optional carbohydrate counting and easy-bolus strategies for hybrid closed-loop evaluation.

Main Results:

• AIDANET system showed comparable mean glucose levels (161.3 ± 16.7 mg/dL) to usual care (168 ± 24.3 mg/dL).
• Time-in-range (TIR) 70-180 mg/dL was slightly improved with AIDANET (66.4% ± 8.3%) versus usual care (63.3% ± 14.9%).
• Hybrid bolusing options demonstrated safe glycemic control, with Easy Bolus achieving 70.5% TIR.

Conclusions:

• The AIDANET system demonstrated safe glycemic control in a pilot feasibility study.
• The system's potential for reduced user input in automated insulin delivery was highlighted.
• Larger, statistically powered trials are necessary to validate these preliminary findings and assess generalizability.