Non-Invasive Blood Pressure Sensing via Machine Learning.

Area of Science:

• Biomedical Engineering
• Machine Learning in Healthcare
• Signal Processing

Background:

• Non-invasive blood pressure monitoring is crucial for telemedicine.
• Photoplethysmography (PPG) signals offer a potential non-invasive source for blood pressure estimation.
• Existing methods require further refinement for clinical accuracy and telemedicine applications.

Purpose of the Study:

• To develop and validate machine learning models for non-invasive blood pressure estimation using PPG signals.
• To assess the performance of eXtreme Gradient Boost (XGBoost) and Neural Network (NN) models.
• To ensure the developed models meet established medical device standards.

Main Methods:

• Extracted novel features from PPG signals using Maximal Overlap Discrete Wavelet Transform (MODWT).
• Selected optimal features using the Minimum Redundancy Maximum Relevance (MRMR) algorithm.
• Trained and compared XGBoost and NN regression models for estimating systolic blood pressure (SBP) and diastolic blood pressure (DBP).

Main Results:

• XGBoost models demonstrated superior accuracy over NN models for both SBP and DBP estimation.
• Achieved a Root Mean Square Error (RMSE) of 5.67 mmHg for SBP and 3.95 mmHg for DBP.
• The XGBoost model's SBP estimation performance surpassed existing literature benchmarks.

Conclusions:

• The developed XGBoost regression model accurately estimates blood pressure non-invasively from PPG signals.
• The model meets the stringent requirements of the Association for the Advancement of Medical Instrumentation (AAMI) and the British Hypertension Society (BHS) grade A standard.
• This approach is suitable for integration into telemedicine health-care monitoring systems.