Related Experiment Video
Updated: May 11, 2026
08:19
Hollow Microneedle-based Sensor for Multiplexed Transdermal Electrochemical Sensing
Published on: June 1, 2012
14.3K
A Deep Sparse Capsule Network for Non-Invasive Blood Glucose Level Estimation Using a PPG Sensor
Narmatha Chellamani1, Saleh Ali Albelwi1, Manimurugan Shanmuganathan1
1Faculty of Computers and Information Technology, University of Tabuk, Tabuk 71491, Saudi Arabia.
Sensors (Basel, Switzerland)
|April 28, 2025
Summary
This study presents a non-invasive method for monitoring blood glucose levels (BGLs) using deep learning and photoplethysmography (PPG) signals. The developed Deep Sparse Capsule Network (DSCNet) model offers accurate BGL estimation, improving diabetes management.
Area of Science:
- Biomedical Engineering
- Artificial Intelligence
- Medical Informatics
Background:
- Diabetes mellitus necessitates frequent blood glucose level (BGL) monitoring.
- Current invasive BGL monitoring methods present challenges and discomfort for patients.
- Non-invasive BGL monitoring techniques are highly sought after to improve patient compliance and quality of life.
Purpose of the Study:
- To develop and evaluate a non-invasive deep learning (DL) approach for estimating blood glucose levels (BGLs) using photoplethysmography (PPG) signals.
- To introduce a novel Deep Sparse Capsule Network (DSCNet) model for accurate and robust BGL monitoring.
- To compare the performance of a baseline DSCNet model against an enhanced version incorporating self-attention mechanisms.
Main Methods:
- A hardware module comprising a PPG sensor and Raspberry Pi was utilized for patient data collection.
- Signal preprocessing involved applying Savitzky-Golay and moving average filters to denoise PPG signals while preserving essential waveform characteristics.
- A Deep Sparse Capsule Network (DSCNet), including a variant with self-attention, was employed for BGL prediction based on processed PPG data.
Main Results:
- The DSCNet model demonstrated high performance in BGL estimation.
- Key performance metrics included Mean Absolute Percentage Error (MAPE) of 3.022, Mean Absolute Error (MAE) of 0.05, and a coefficient of determination (R²) of 0.98.
- The enhanced DSCNet model with self-attention showed improved accuracy and robustness in predicting blood glucose levels.
Conclusions:
- The proposed non-invasive DL-based approach using PPG signals offers a promising alternative for BGL monitoring.
- The DSCNet model, particularly the self-attention enhanced version, provides accurate and reliable BGL estimations.
- This technology has the potential to significantly improve diabetes management and patient care by reducing the need for invasive procedures.
