Estimating Blood Pressure from the Photoplethysmogram Signal and Demographic Features Using Machine Learning...

Area of Science:

• Biomedical Engineering
• Health Informatics
• Machine Learning

Background:

• Hypertension is a significant health risk, necessitating continuous blood pressure (BP) monitoring.
• Traditional cuff-based BP measurements are discrete and uncomfortable, limiting continuous monitoring.
• A need exists for noninvasive, continuous BP monitoring solutions.

Purpose of the Study:

• To develop and evaluate a cuff-less, continuous, and noninvasive BP measurement system.
• To utilize photoplethysmograph (PPG) signals and demographic features with machine learning (ML) algorithms for BP estimation.
• To identify the most effective ML model for accurate systolic BP (SBP) and diastolic BP (DBP) prediction.

Main Methods:

• Acquired PPG signals from 219 subjects for preprocessing and feature extraction.
• Extracted time, frequency, and time-frequency domain features from PPG and derivative signals.
• Employed feature selection techniques and trained various ML algorithms, including Gaussian process regression (GPR), for SBP and DBP estimation.

Main Results:

• Gaussian process regression (GPR) combined with ReliefF feature selection demonstrated superior performance.
• The best model achieved a root mean square error (RMSE) of 6.74 for SBP estimation.
• The model achieved an RMSE of 3.59 for DBP estimation, indicating high accuracy.

Conclusions:

• The proposed ML model enables accurate, cuff-less, and continuous BP monitoring using PPG signals.
• This noninvasive system offers a comfortable and practical alternative to conventional BP measurement methods.
• Implementation in hardware can facilitate real-time BP tracking, potentially preventing critical health events.