A novel few shot learning derived architecture for long-term HbA1c prediction

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Pediatric Endocrinology

Background:

• Regular monitoring of glycated hemoglobin (HbA1c) is crucial for diabetes management, as elevated levels are linked to microvascular complications and comorbidities.
• Predicting future HbA1c levels allows for timely treatment adjustments and lifestyle changes to prevent severe health issues.
• Current research lacks methods for long-term HbA1c prediction using readily available blood glucose measurements.

Purpose of the Study:

• To propose a novel algorithm for the long-term prediction of clinical HbA1c measures using blood glucose data.
• To specifically target the pediatric Type-1 diabetic population for early detection and prevention of complications.
• To investigate the use of Few-Shot Learning (FSL) derived methods for HbA1c prediction.

Main Methods:

• Processing short-term Continuous Glucose Monitoring (CGM) time-series data using novel image transformation and conventional signal processing.
• Utilizing a Convolutional Neural Network (CNN) adapted from a Few-Shot Learning (FSL) model for feature extraction from processed data.
• Implementing a novel normalized FSL-distance (FSLD) metric for feature separation and a K-nearest neighbor (KNN) model for final classification.

Main Results:

• The proposed FSL-derived algorithm achieved a prediction accuracy of 93.2% for clinical HbA1c levels.
• The method successfully fused features extracted through CNN and FSL models.
• The FSLD metric effectively differentiated between various HbA1c level features.

Conclusions:

• This study presents the first literature-based FSL-derived algorithm for long-term HbA1c prediction in pediatric Type-1 diabetes.
• The developed algorithm demonstrates high accuracy, offering a promising tool for proactive diabetes management in children.
• Early prediction capabilities can significantly help in averting life-threatening complications associated with elevated HbA1c levels in young patients.