Learning Difference Equations With Structured Grammatical Evolution for Postprandial Glycaemia Prediction.

Area of Science:

• Biomedical Engineering
• Computational Biology
• Diabetology

Background:

• Diabetes management requires precise blood glucose monitoring, especially post-meal.
• Accurate glucose prediction is crucial for preventing dangerous complications.
• Existing methods like ANNs lack interpretability for personalized treatments.

Purpose of the Study:

• To develop a novel, interpretable glucose prediction method for post-meal blood glucose levels.
• To provide physicians with explainable models for personalized diabetes management.
• To balance prediction accuracy with model interpretability and computational efficiency.

Main Methods:

• Proposes Interpretable Sparse Identification by Grammatical Evolution (ISIGE) for glucose prediction.
• Combines ISIGE with a pre-meal blood glucose clustering stage.
• Generates finite difference equations for predicting glucose levels up to two hours post-meal in 15-minute intervals.

Main Results:

• The ISIGE method achieved safe predictions, avoiding high-risk zones on the Parkes Error Grid.
• Demonstrated slightly improved accuracy compared to traditional methods and artificial neural networks.
• Provided interpretable mathematical expressions for glucose level predictions.

Conclusions:

• The proposed ISIGE approach offers accurate and interpretable glucose predictions for diabetes management.
• This method provides a promising balance between accuracy, interpretability, and computational efficiency.
• Facilitates personalized treatment strategies by offering explainable prediction models.