A multistage deep neural network model for blood pressure estimation using photoplethysmogram signals

Area of Science:

• Biomedical Engineering
• Signal Processing
• Artificial Intelligence

Background:

• Cuff-based and invasive methods for blood pressure (BP) measurement present significant limitations.
• Non-invasive and easily accessible bio-signals are crucial for overcoming these challenges in BP monitoring.

Purpose of the Study:

• To propose and validate a multistage deep neural network model for estimating systolic and diastolic blood pressures.
• To utilize photoplethysmogram (PPG) signals for accurate, non-invasive BP estimation.

Main Methods:

• A two-stage deep learning approach combining Convolutional Neural Networks (CNNs) for feature extraction and Long Short-Term Memory (LSTM) for temporal dependency analysis.
• The model incorporates the dynamic relationship between systolic and diastolic blood pressures to enhance estimation accuracy.

Main Results:

• The proposed model achieved performance meeting the Association for the Advancement of Medical Instrumentation (AAMI) standards.
• Evaluation on 200 subjects resulted in Grade A performance for both systolic and diastolic BP estimation according to British Hypertension Society (BHS) standards.
• Mean error and standard deviation for systolic and diastolic BP estimations were +1.91±5.55mmHg and +0.67±2.84mmHg, respectively.

Conclusions:

• The developed multistage model demonstrates effective feature extraction from PPG signals.
• The model offers consistent and accurate estimation of blood pressure, highlighting its potential for non-invasive BP monitoring.