A multimodal physiological dataset for non-invasive blood glucose estimation

Area of Science:

• Biomedical Engineering
• Wearable Technology
• Diabetes Research

Background:

• Diabetes affects millions globally, necessitating effective blood glucose monitoring.
• Current continuous glucose monitors (CGMs) are invasive.
• Physiological signals change with blood glucose fluctuations.

Purpose of the Study:

• To introduce PhysioCGM, an open-source dataset for non-invasive glucose monitoring.
• To facilitate the development of machine learning models for predicting glucose levels.
• To improve diabetes management through advanced sensing technologies.

Main Methods:

• Collected multimodal physiological data (ECG, PPG, EDA, skin temperature, accelerometry) using wearable sensors.
• Acquired concurrent ground-truth continuous glucose monitor (CGM) data.
• Recorded data from 10 participants with Type 1 Diabetes over 17 days in ambulatory settings.

Main Results:

• PhysioCGM provides raw, high-fidelity physiological recordings synchronized with CGM data.
• The dataset captures physiological changes related to blood glucose excursions.
• Enables training of machine learning models on comprehensive, multimodal data.

Conclusions:

• PhysioCGM addresses limitations of existing datasets for non-invasive glucose monitoring research.
• The dataset supports the advancement of machine learning algorithms for diabetes management.
• Promotes development of non-invasive wearable solutions for continuous glucose tracking.