Subject-Based Model for Reconstructing Arterial Blood Pressure from Photoplethysmogram

Area of Science:

• Biomedical Engineering
• Cardiovascular Physiology
• Artificial Intelligence in Medicine

Background:

• Continuous, non-invasive arterial blood pressure (ABP) monitoring is crucial for cardiovascular disease management.
• Photoplethysmography (PPG) signals offer potential for ABP reconstruction due to shared physiological origins and signal similarities.
• Existing PPG-to-ABP reconstruction methods often lack global similarity evaluation, limiting accuracy.

Purpose of the Study:

• To develop and validate a deep learning model for accurate, continuous, non-invasive ABP waveform reconstruction from PPG signals.
• To address limitations in existing methods by incorporating global similarity assessment.
• To improve cardiovascular disease prevention and treatment through enhanced ABP monitoring.

Main Methods:

• A novel deep learning W-Net architecture, comprising a U-Net encoder and decoder, was proposed for ABP signal reconstruction from PPG.
• The model was trained and tested using 500 records of varying lengths.
• Performance was evaluated using Pearson correlation, root mean square error (RMSE), normalized dynamic time warping (NDTW) distance, and mean absolute errors for systolic blood pressure (SBP) and diastolic blood pressure (DBP).

Main Results:

• The W-Net model achieved high similarity between reconstructed and reference ABP signals.
• Average similarity measures included a Pearson correlation of 0.995, RMSE of 2.236 mmHg, and NDTW distance of 0.612 mmHg.
• Mean absolute errors for SBP and DBP were 2.602 mmHg and 1.450 mmHg, respectively, demonstrating significant accuracy.

Conclusions:

• The proposed W-Net deep learning model effectively reconstructs ABP signals from PPG with high fidelity.
• This non-invasive method shows promise for continuous ABP monitoring and clinical applications in cardiovascular health.
• The model's ability to capture global signal similarity enhances its reliability for predicting ABP waveforms.