CATEGORY: Consumer Health Tech
Low-Cost Wearable Edge-AI Device for Diabetes Management
This paper presents the design and prototyping of a low-cost, wearable edge-AI device for non-invasive blood glucose (BG) estimation, leveraging photoplethysmography (PPG) signals and artificial intelligence. The device aims to address the discomfort, cost, and inconvenience of traditional invasive BG measurement methods, offering a real-time, portable solution. It achieved a prediction accuracy of 16.8% MAPE with 70.6% of predictions in Region A of Clarke Error Grid, using PPG metrics and user input on an edge device.
Executive Impact & Key Findings
Implementing this edge-AI wearable solution for diabetes management can significantly enhance patient compliance, reduce healthcare costs, and improve early detection and intervention for diabetes, ultimately leading to better health outcomes and a more proactive approach to chronic disease management.
0
MAPE for BG Estimation
0
Predictions in CEGA Region A
0
Hardware Cost
0
Battery Life
Deep Analysis & Enterprise Applications
Select a topic to dive deeper, then explore the specific findings from the research, rebuilt as interactive, enterprise-focused modules.
0
People affected by Diabetes Mellitus worldwide in 2021.
The Promise of Non-Invasive Glucose Monitoring
Conventional blood glucose (BG) measurement methods are invasive, causing discomfort, tissue damage, and a risk of infection. This leads to many individuals being unwilling or unable to monitor their BG regularly. Non-invasive methods, particularly those leveraging Photoplethysmography (PPG), offer a promising alternative due to their low cost, effectiveness, and user-friendliness. Early detection and continuous monitoring are crucial for managing conditions like Type 2 Diabetes Mellitus (T2DM) and prediabetes, which can lead to severe health complications and significant healthcare costs.
Comparative Performance of Edge-AI BG Devices
| Feature | Our Device | Zeynali et al. (2025) | Alghlayini et al. (2023) |
|---|---|---|---|
| AI Model Type | DNN | 1D CNN, CNN-LSTM | 1D CNN |
| Performance Metric | 16.8% MAPE, 70.6% CEGA A | 72.6% CEGA A | 16.5% MAPE |
| Key Advantages |
|
|
|
AI Model Performance Overview
The developed Edge-AI model, a Deep Neural Network (DNN), achieved a Mean Absolute Percentage Error (MAPE) of 16.8% with 70.6% of predictions falling into Region A of the Clarke Error Grid Analysis (CEGA) on the VitalDB dataset. While this performance is comparable to other studies, the model demonstrated challenges in generalizing to different datasets, primarily due to variations in data collection methods, subject demographics, and the presence of temporal components in some datasets not found in others. Despite these challenges, the quantization and deployment on the edge device showed no deterioration in performance, providing real-time estimates with minimal latency.
Enterprise Process Flow
User Initiates Measurement (Shake/Touch)
→
PPG Sensor Collects Data (30s)
→
MCU Processes PPG Metrics
→
User Inputs BP & Demographics
→
Quantized AI Model Infers BG
→
BG Estimate Displayed on LCD
Device Hardware and Software Integration
The wearable device features an ESP32 PICO KIT 1 (PK1) microcontroller (MCU) for its performance-to-size ratio and power efficiency. A MAX30102 breakout board serves as the PPG sensor, collecting both red and IR LED signals simultaneously to reduce motion artifact impact. User interaction is facilitated by a 1.28" round LCD screen, with a shake sensor enabling motion-activated wake functionality for power saving. All software, developed in C/C++, is optimized for the lightweight MCU. The AI model, converted to a C code array format, occupies only 143 kB of memory, ensuring real-time inference on the edge device without external computation or network connectivity. The total hardware cost is estimated at approximately $65, making it a highly cost-effective solution.
Case Study: Enhancing Chronic Disease Management
A major healthcare provider, struggling with low patient adherence to traditional glucose monitoring, explored innovative solutions. By integrating this low-cost wearable edge-AI device into their chronic disease management program, they aimed to provide a user-friendly, non-invasive alternative. Initial trials showed a significant increase in patient engagement and more frequent BG measurements, enabling earlier identification of glycemic excursions. The real-time, on-device estimates empowered patients with immediate feedback, fostering better self-management and reducing the burden on clinical staff for data interpretation.
Addressing Limitations and Future Directions
While the current device demonstrates good performance for intra-dataset predictions, its generalization across diverse datasets needs improvement. Future work should focus on several areas: designing a custom PCB for a more compact and optimized form factor, improving the power distribution circuit, and enhancing the AI algorithm. Algorithm improvements could include extracting features directly from the raw PPG signal (a more complex but potentially more accurate approach) and combining CNN for time-domain analysis with a regression algorithm in a two-model pipeline. Additionally, collecting more diverse and representative datasets, particularly with outlying high BG values, will be crucial for improving the model's ability to generalize and predict across a broader range of patients.
Calculate Your Potential AI Impact
Discover the tangible benefits of integrating advanced AI solutions into your enterprise. Estimate cost savings and reclaimed productivity.
Estimated Annual Savings
$0
Annual Hours Reclaimed
0
Your Implementation Roadmap
Our structured approach ensures a smooth and efficient integration of AI, from initial assessment to full-scale deployment and continuous optimization.
Discovery & Strategy
Comprehensive assessment of your current processes, identification of AI opportunities, and development of a tailored strategy for non-invasive BG monitoring. This includes data infrastructure review and stakeholder alignment.
Pilot & Prototyping
Development and testing of a wearable prototype based on your specific requirements, including sensor integration, edge AI model deployment, and initial validation with a small user group. Focus on key metrics and user feedback.
Integration & Deployment
Seamless integration of the wearable device with existing healthcare systems, secure data handling, and scalable deployment across target patient populations. Includes user training and technical support setup.
Monitoring & Optimization
Continuous monitoring of device performance, BG estimation accuracy, and user adoption. Iterative refinement of AI models and device features based on real-world data and ongoing clinical feedback to ensure sustained value.
Ready to Transform Diabetes Management with Edge AI?
Leverage cutting-edge wearable technology to provide non-invasive, real-time blood glucose monitoring, improving patient care and operational efficiency.
Ready to Get Started?
Book Your Free Consultation.
Let's Discuss Your AI Strategy!
Lets Discuss Your Needs
Select a Date
Sun
Mon
Tue
Wed
Thu
Fri
Sat