Enabling fully automated insulin delivery through meal detection and size estimation using Artificial Intelligence

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Diabetes Technology

Background:

• Automated insulin delivery systems aim to improve glucose control in diabetes management.
• Accurate meal detection and carbohydrate estimation are crucial for effective mealtime insulin dosing.
• Current systems often require manual input, posing a challenge for real-world application.

Purpose of the Study:

• To evaluate the performance of a novel robust artificial pancreas (RAP) system.
• To compare postprandial glucose control between the RAP system and a hybrid model predictive control (MPC) algorithm.
• To assess the efficacy of RAP's automated meal detection and insulin dosing capabilities.

Main Methods:

• A randomized, single-center crossover trial was conducted.
• Participants used both the RAP system and a hybrid MPC algorithm for postprandial glucose control.
• The RAP system incorporates a neural network for automated meal detection and carbohydrate estimation.
• Performance was evaluated over four hours following unannounced meals.

Main Results:

• The RAP system's meal detection algorithm achieved 83.3% sensitivity and a 16.6% false discovery rate with a mean detection time of 25.9 minutes.
• No significant difference in incremental area under the glucose curve was observed between RAP and MPC.
• RAP significantly reduced time above range (glucose >180 mg/dL) by 10.8% (P=0.04).
• RAP showed a trend towards increasing time in range (70-180 mg/dL) by 9.1%.

Conclusions:

• The robust artificial pancreas (RAP) system demonstrates improved glucose control by reducing hyperglycemia.
• Automated meal detection and dosing in RAP enhance usability and potentially improve diabetes management.
• Further research is warranted to optimize RAP performance and explore its long-term benefits.