Neural network based glucose - insulin metabolism models for children with Type 1 diabetes

Area of Science:

• Biomedical Engineering
• Computational Biology
• Endocrinology

Background:

• Type 1 diabetes management requires accurate glucose level prediction.
• Simulating glucose-insulin metabolism is crucial for developing effective treatment strategies.
• Existing models may not fully capture the complex dynamics of pediatric Type 1 diabetes.

Purpose of the Study:

• To develop and evaluate hybrid computational models for simulating glucose-insulin metabolism in children with Type 1 diabetes.
• To compare the predictive performance of different artificial neural network architectures for short-term glucose forecasting.
• To integrate compartmental models with neural networks for enhanced metabolic simulation.

Main Methods:

• Development of two hybrid models combining Compartmental Models (CMs) and artificial Neural Networks (NNs).
• Utilized data from four children with Type 1 diabetes, including continuous glucose monitoring, insulin, and food intake.
• Compared Feed-Forward Neural Networks (FFNNs) and Recurrent Neural Networks (RNNs) for glucose prediction.

Main Results:

• CMs successfully estimated the influence of insulin on plasma insulin and food intake on blood glucose.
• The hybrid models provided short-term predictions of glucose values.
• Recurrent Neural Networks (RNNs) achieved the best prediction performance compared to FFNNs.

Conclusions:

• Hybrid models integrating CMs and NNs are effective for simulating glucose-insulin metabolism in pediatric Type 1 diabetes.
• RNNs offer superior accuracy for short-term glucose level prediction in this population.
• These models hold promise for improving diabetes management tools.