Blood Pressure Morphology Assessment from Photoplethysmogram and Demographic Information Using Deep Learning with...

Area of Science:

• Biomedical Engineering
• Cardiovascular Physiology
• Artificial Intelligence in Medicine

Background:

• Arterial blood pressure (ABP) is a critical vital sign for assessing cardiovascular health and diagnosing conditions like hypertension.
• Current cuff-based methods provide only intermittent systolic and diastolic blood pressure (SBP and DBP) readings.
• Routine monitoring of ABP morphology can offer deeper insights into cardiovascular status.

Purpose of the Study:

• To develop and validate a novel cuff-free method for estimating the average ABP pulse (ABPM¯) morphology.
• To utilize deep learning, specifically a seq2seq architecture with an attention mechanism, for PPG signal analysis.
• To integrate demographic information (DI) to improve the accuracy of ABP estimation.

Main Methods:

• A deep learning model employing a seq2seq architecture with an attention mechanism was developed.
• The model processes raw photoplethysmogram (PPG) signals from the finger.
• Categorical and continuous demographic information (age, gender) was integrated into the model.

Main Results:

• The model achieved mean absolute errors (MAE) of 6.57 ± 0.20 mmHg for DBP and 14.39 ± 0.42 mmHg for SBP.
• For ABPM¯ estimation, a correlation coefficient (R) of 0.98 ± 0.001 and an MAE of 8.89 ± 0.10 mmHg were obtained.
• Integrating demographic information improved the accuracy of the ABP pulse estimation.

Conclusions:

• The proposed methodology effectively transforms PPG signals into ABP pulse morphology.
• This cuff-free approach demonstrates high accuracy in estimating SBP, DBP, and ABPM¯.
• The method shows potential for widespread use in wearable devices for continuous, non-invasive cardiovascular monitoring.