BioScan: An IoT System for Non-Invasive Blood Glucose Monitoring Using PPG Sensors

BioScan is an IoT system for non-invasive blood glucose monitoring based on PPG sensors, developed and open-sourced by Pruthvi Raj (GitHub link: https://github.com/pruthvi9053/BioScan-IoT-Enabled-Glucose-Detection-System-Using-Machine-Learning, released on 2026-06-06). This system combines ESP32 microcontroller, MAX30102 photoplethysmography (PPG) sensor, and machine learning algorithms to achieve non-invasive estimation of blood glucose levels by collecting fingertip photoelectric signals, aiming to address the pain points of traditional invasive blood glucose monitoring.

Diabetes is a global chronic health issue affecting hundreds of millions of people. Traditional blood glucose monitoring requires pricking the skin to collect blood, causing pain and increasing infection risks. Non-invasive blood glucose monitoring has become an important research direction. BioScan uses PPG sensors to collect fingertip optical signals and combines ML algorithms to achieve non-invasive blood glucose estimation, improving patient compliance and enabling continuous monitoring.

BioScan is an end-to-end IoT system consisting of four layers:

1. Hardware layer: ESP32 drives the MAX30102 sensor to collect data, which is preprocessed and uploaded via WiFi; the MAX30102 has features such as dual-wavelength LEDs and high-sensitivity detectors.
2. Backend layer: A RESTful API built with Flask, responsible for data reception and storage (SQL Server), preprocessing, model inference, and user management.
3. Frontend layer: A Web UI that provides functions like real-time data display, historical records, trend analysis, and alarm settings.
4. Machine learning layer: The core is to extract features from PPG signals and build prediction models.

PPG technology is based on the selective absorption of light of specific wavelengths by blood. The MAX30102 operates by alternating light emission, detecting reflected light, converting it into digital signals, and transmitting them to the ESP32. Changes in blood glucose concentration affect the optical properties of blood:

1. Changes in scattering characteristics (glucose alters the scattering pattern of red blood cells);
2. Shift in absorption spectrum (indirectly reflects blood glucose changes, such as blood viscosity and tissue perfusion rate);
3. Multi-parameter fusion (time-domain, frequency-domain, and dual-wavelength features).

Model design includes:

• Feature engineering: Time-domain (peak amplitude, cycle, etc.), frequency-domain (HRV, power ratio, etc.), and dual-wavelength (absorption ratio, correlation, etc.) features.
• Model selection: Regression (linear, SVR), ensemble (Random Forest, XGBoost), neural networks (MLP, CNN, LSTM).
• Evaluation metrics: RMSE, MAE, Clarke Error Grid Analysis (CEG), where Zone A is clinically accurate and Zone B is acceptable.

Implementation details:

• Data collection: ESP32 initializes the sensor, samples at 100Hz for 30-60 seconds, preprocesses the data, and uploads it in JSON format.
• Backend processing: Receives data, stores it, extracts features, performs model inference, and returns results. Challenges and solutions:

1. Signal instability: Adaptive filtering, signal quality assessment, averaging of multiple data sets;
2. Individual differences: Personalized calibration, transfer learning, online optimization;
3. Privacy and security: HTTPS encryption, field encryption, user isolation;
4. Real-time performance: Lightweight models, edge computing, asynchronous processing.

Application scenarios:

• Personal management: Daily monitoring, continuous tracking, data sharing;
• Clinical assistance: Outpatient screening, inpatient monitoring, surgical monitoring;
• Health warning: High/low blood glucose risk prediction, abnormal alarms;
• Scientific research: Large-scale dataset collection, algorithm validation. Limitations: Lower accuracy than invasive methods, need for regular calibration, influence of individual differences, environmental interference. Future directions: Multi-modal fusion, deep learning optimization, personalized models, edge AI deployment, clinical validation.

BioScan demonstrates the innovative application of IoT, sensor technology, and ML in the healthcare field, providing a feasible solution for non-invasive blood glucose monitoring. Although there are current challenges such as accuracy, with technological progress, it is expected to become an important tool for diabetes management. Its open-source implementation provides reference value for research in this field.