Carbohydrate Content Classification Using Postprandial Heart Rate Responses from Non-Invasive Wearables

Area of Science:

• Biomedical Engineering
• Wearable Technology
• Diabetes Management

Background:

• Rising type 2 diabetes incidence necessitates improved dietary monitoring tools.
• Current technologies track meal timing and content, but non-invasive carbohydrate classification is underexplored.

Purpose of the Study:

• To investigate carbohydrate content classification using postprandial heart rate (PHR) responses from non-invasive wearables.
• To develop and validate a system for estimating dietary carbohydrate load.

Main Methods:

• Developed the timeStampr iOS app for data labeling and ground truth collection.
• Conducted a pilot study with 23 participants using Empatica E4 devices to record heart rate.
• Collected data during consumption of low-carbohydrate and carbohydrate-rich meals.
• Trained a Light Gradient Boosting Machine (LGBM) model for classification.

Main Results:

• Classifiers achieved over 84% accuracy, precision, recall, and AUCROC within a 60-second window.
• Demonstrated robust performance in distinguishing between low- and high-carbohydrate meals.
• Successfully utilized PHR signals for dietary carbohydrate content classification.

Conclusions:

• Postprandial heart rate responses from non-invasive wearables show potential for classifying dietary carbohydrate content.
• This approach could enhance diabetes management by providing objective dietary intake data.
• Further research is warranted to refine the technology and address limitations such as sensor performance across diverse skin tones.