Assessing Elevated Blood Glucose Levels Through Blood Glucose Evaluation and Monitoring Using Machine Learning and...

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Metabolic Disease Research

Background:

• Diabetes mellitus is a major global health challenge, affecting 10.5% of adults, with many undiagnosed.
• Current screening methods are invasive, opportunistic, and require professional administration, leading to delayed interventions.
• A growing population at risk for diabetes strains healthcare resources.

Purpose of the Study:

• To develop an artificial intelligence (AI) solution for noninvasive recognition of elevated blood glucose levels (≥7.8 mmol/L).
• To evaluate diabetic risk using repeated measurements from a wearable device.

Main Methods:

• 500 participants were recruited, with blood glucose measured using a conventional glucometer and a wrist-worn wearable.
• Photoplethysmography (PPG) sensor data underwent extensive feature engineering to identify glucose-sensitive features.
• Explainable AI was used to analyze the contribution of selected features to predictions.

Main Results:

• Machine learning models were trained and validated using demographic data and PPG features.
• A support vector machine achieved the highest performance: 84.7% accuracy, 81.05% sensitivity, and 88.3% specificity.
• Key features from PPG signals were identified as important predictors for glucose levels.

Conclusions:

• Photoplethysmography (PPG) measurements show potential for identifying elevated blood glucose.
• This noninvasive approach can aid in screening individuals for diabetes risk.
• AI-powered analysis of wearable data offers a novel avenue for proactive diabetes management.