Selection of Noninvasive Features in Wrist-Based Wearable Sensors to Predict Blood Glucose Concentrations Using...

Area of Science:

• Biomedical Engineering
• Data Science
• Diabetes Technology

Background:

• Diabetes Mellitus (DM) management relies on accurate blood glucose level (BGL) monitoring.
• Traditional Self-Monitoring Blood Glucose (SMBG) methods are invasive, requiring finger pricks.
• Noninvasive Continuous Glucose Monitoring (NICGM) offers a less invasive alternative for BGL tracking.

Purpose of the Study:

• To investigate the feasibility of a novel Noninvasive Continuous Glucose Monitoring (NICGM) approach.
• To utilize multiple off-the-shelf wearable sensors and machine learning (ML) models for BGL prediction.
• To assess the accuracy of noninvasive biometric measurements in predicting blood glucose.

Main Methods:

• Employed two datasets: OhioT1DM and a custom UofM dataset.
• The UofM dataset included fourteen features from six wearable sensors.
• Applied a machine learning pipeline with linear and nonlinear models to predict BGLs from noninvasive features.

Main Results:

• Demonstrated that fourteen noninvasive biometric measurements combined with ML algorithms can achieve accurate BGL predictions.
• Results indicate predictions fall within the clinically acceptable range.
• This pilot study highlights the potential of the developed NICGM approach.

Conclusions:

• The combination of noninvasive wearable sensor data and machine learning shows promise for accurate BGL prediction.
• Further validation with larger datasets is necessary to confirm the feasibility and reliability of this NICGM method.
• This approach could significantly improve diabetes management by offering a noninvasive monitoring solution.