Personalized Blood Pressure Estimation Using Photoplethysmography: A Transfer Learning Approach

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Cardiovascular Monitoring

Background:

• Accurate blood pressure (BP) monitoring is crucial for cardiovascular health management.
• Photoplethysmogram (PPG) signals offer a non-invasive method for potential BP estimation.
• Limited individual data hinders the development of personalized BP estimation models.

Purpose of the Study:

• To develop a personalized deep learning approach for estimating BP from PPG signals.
• To introduce a transfer learning technique to overcome data scarcity in individual BP prediction.
• To validate the accuracy and clinical applicability of the proposed BP estimation method.

Main Methods:

• A hybrid neural network (convolutional, recurrent, fully connected layers) was designed to process raw PPG time series.
• Transfer learning was employed, personalizing a pre-trained network using limited individual data.
• The MIMIC III database, containing invasive arterial BP and PPG data, was utilized for model development and analysis.

Main Results:

• The proposed BP-CRNN-Transfer model achieved a Mean Absolute Error (MAE) of 3.52 mmHg for systolic BP (SBP) and 2.20 mmHg for diastolic BP (DBP).
• The method met both British Hypertension Society (BHS) and Association for the Advancement of Medical Instrumentation (AAMI) standards for BP measurement.
• Accurate personalized models were trainable with as few as 50 data samples per individual.

Conclusions:

• The personalized deep learning approach effectively estimates BP from PPG signals with high accuracy.
• Transfer learning significantly enhances personalized BP estimation, even with minimal individual data.
• The developed method demonstrates clinical viability and adherence to established BP measurement standards.