Unannounced Meal Detection for Artificial Pancreas Systems Using Extended Isolation Forest

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Diabetes Technology

Background:

• Artificial pancreas systems require accurate meal detection for effective glucose control.
• Unannounced meals pose a significant challenge to maintaining glycemic targets.
• Existing detection methods may lack the precision needed for real-world application.

Purpose of the Study:

• To develop and evaluate an unannounced meal detection method for artificial pancreas systems.
• To leverage an advanced machine learning algorithm, Extended Isolation Forest (EIF), for enhanced detection capabilities.
• To utilize individual Continuous Glucose Monitoring (CGM) profiles for personalized detection.

Main Methods:

• Implementation of an Extended Isolation Forest (EIF) algorithm for anomaly detection.
• Feature engineering based on individual Continuous Glucose Monitoring (CGM) data.
• Application of a two-threshold decision rule for robust detection.
• Validation using simulated data from virtual diabetic patients.

Main Results:

• The EIF-based method demonstrated high accuracy in detecting unannounced meals.
• The proposed approach achieved acceptable detection delays, crucial for real-time control.
• Experiments confirmed the superiority of EIF over standard Isolation Forest for this task.
• The method effectively utilized personalized CGM features for improved performance.

Conclusions:

• The developed EIF-based method offers a promising solution for unannounced meal detection in artificial pancreas systems.
• This advancement can contribute to more precise glucose control and improved quality of life for individuals with diabetes.
• Further clinical validation is warranted to translate these findings into practice.