Non-Invasive Blood Pressure Monitoring and Estimation using Dual-Wavelength PPG and Machine Learning

This study introduces a photoplethysmography (PPG)-based method for monitoring blood pressure(BP) using finger-tip signals. This approach enables the real-time, non-invasive estimation of systolic and diastolic blood pressure through the analysis of pulse wave signals. The comprehensive framework presents for non-invasive BP estimation utilizing dual-wavelength PPG signals in conjunction with machine learning (ML) algorithms. In the proposed sys-tem, an embedded pulse oximetry and heart-rate sensing module is built using the MAX30101 optical sensor and the MAX32664 ultra-low-power biometric sensor hub to facilitate the synchronized PPG signals at 660 nm and 880 nm wavelengths. The acquired signals are then processed in real-time by a microcontroller executing a Blood Pressure Tracking (BPT) algorithm, which estimates systolic and diastolic BP. Multiple regression models are trained and tested with quantitative output driven by derived features based on PPG combined with demographic data of 575 individuals. The predictive accuracy of the proposed models was compared to PPM reference BP values and the linear regression model had the best predictive accuracy with Mean Absolute Error (MAE) and 3.25 mmHg and a Mean Squared Error (MSE) 16.47 mmHg2. This metrics suggest that the proposed system has a robust, scalable solution to the continuous, non-invasive BP monitoring, appropriate in healthcare applications.