Microcontroller Implementation of Support Vector Machine for Detecting Blood Glucose Levels Using Breath Volatile...

Area of Science:

• Biomedical Engineering
• Analytical Chemistry
• Embedded Systems

Background:

• Current blood glucose monitoring methods are invasive.
• Volatile organic compounds (VOCs) in breath correlate with blood glucose levels.
• Non-invasive glucose monitoring is a significant unmet clinical need.

Purpose of the Study:

• To develop and validate an embedded system for non-invasive blood glucose estimation using breath VOCs.
• To implement a Support Vector Machine (SVM) model on a microcontroller for real-time analysis.
• To assess the accuracy of the system in categorizing high and low blood glucose levels.

Main Methods:

• Utilized sensor arrays to detect VOCs in artificial breath samples.
• Trained a Support Vector Machine (SVM) model on a computer using an existing library.
• Implemented the optimized SVM model on an ATMega microprocessor-based embedded system.
• Validated the system using artificial breath mimicking VOC profiles of different glucose levels.

Main Results:

• The embedded system accurately categorized blood glucose levels from breath VOCs.
• Achieved a classification accuracy of 97.1% for artificial breath samples.
• Demonstrated the feasibility of implementing a sophisticated machine learning model on a low-power microcontroller.

Conclusions:

• The developed embedded system offers a promising non-invasive approach for blood glucose monitoring.
• This technology contributes to the development of portable, accurate, and patient-friendly glucose detection devices.
• Further research can explore in-vivo validation and integration into wearable devices.