Edge-AI Enabled Wearable Device for Non-Invasive Type 1 Diabetes Detection Using ECG Signals.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Wearable Technology

Background:

• Traditional diabetes monitoring is invasive and impacts patient quality of life.
• There is a need for non-invasive, real-time diabetes detection methods.
• Edge Artificial Intelligence (Edge-AI) offers potential for on-device health monitoring.

Purpose of the Study:

• To design and validate an innovative Edge-AI system for real-time, non-invasive diabetes detection.
• To analyze electrocardiogram (ECG) signals for diabetes presence using a microcontroller-based system.
• To optimize the system for low power consumption and minimal memory footprint suitable for wearable devices.

Main Methods:

• Developed a microcontroller-based system for real-time ECG acquisition.
• Implemented a spectrogram-based preprocessing technique combined with a 1-Dimensional Convolutional Neural Network (1D-CNN).
• Applied quantization for model optimization, balancing memory usage and accuracy.
• Designed and validated a custom Printed Circuit Board (PCB) for real-world testing.

Main Results:

• Achieved 89.52% accuracy in diabetes detection.
• Obtained average precision of 0.91 and recall of 0.90.
• Demonstrated a minimal memory footprint (347 kB flash, 23 kB RAM).
• Validated the system's feasibility on a custom PCB with low power consumption.

Conclusions:

• The developed Edge-AI system enables non-invasive, real-time diabetes detection.
• The system is suitable for resource-constrained wearable embedded devices.
• This approach significantly enhances the accessibility and practicality of diabetes monitoring.