Using Absorption Models for Insulin and Carbohydrates and Deep Leaning to Improve Glucose Level Predictions

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Endocrinology

Background:

• Diabetes Mellitus (DM) affects over 400 million globally, necessitating advanced management strategies.
• Current diabetes care often struggles with optimizing insulin, carbohydrate intake, and physical activity.
• Predictive modeling of blood glucose levels is crucial for proactive diabetes management.

Purpose of the Study:

• To develop an AI-driven model for predicting future blood glucose concentrations in individuals with type 1 diabetes mellitus (T1DM).
• To enhance prediction accuracy by incorporating realistic insulin and carbohydrate absorption dynamics.
• To provide a tool for optimizing diabetes self-management through accurate glucose level forecasting.

Main Methods:

• A Recurrent Neural Network (RNN) model was employed for blood glucose level prediction.
• The RNN model was integrated with simulated carbohydrate and insulin absorption curves.
• The combined model was validated using real-world data from T1DM patients.

Main Results:

• The proposed RNN model demonstrated encouraging performance in predicting blood glucose levels.
• The model achieved prediction accuracy within 0.510 mmol/L (9.2 mg/dL) in optimal conditions.
• The integration of absorption dynamics improved the predictive capabilities compared to models focusing solely on the algorithm.

Conclusions:

• The study highlights the potential of AI, specifically RNNs combined with physiological absorption modeling, for improving T1DM management.
• Accurate prediction of blood glucose levels can empower patients and clinicians to make informed decisions.
• Further research into pre-processing physiological signals can significantly advance predictive accuracy in diabetes care.