Blood Pressure Prediction by a Smartphone Sensor using Fully Convolutional Networks.

Area of Science:

• Biomedical Engineering
• Cardiovascular Health
• Artificial Intelligence in Medicine

Background:

• Heart disease and stroke are leading global causes of mortality.
• Hypertension significantly elevates the risk of cardiovascular events.
• Accurate and convenient blood pressure monitoring is crucial for managing hypertension.

Purpose of the Study:

• To develop a feature-extraction-free method for predicting blood pressure (BP) using fully convolutional neural networks (CNNs).
• To investigate the optimal combination of photoplethysmography (PPG) signal wavelengths for accurate BP prediction.
• To assess the feasibility of using smartphone-acquired PPG signals for non-invasive BP estimation.

Main Methods:

• Acquisition of single multi-wave photoplethysmography (PPG) signals via smartphone.
• Application of fully convolutional neural networks (CNNs) for direct BP prediction from PPG signals.
• Systematic evaluation of different PPG signal wavelength combinations for BP prediction performance.

Main Results:

• The CNN-based BP predictor achieved optimal performance using a combination of green PPG time signals and instantaneous frequency signals.
• The best performing model demonstrated a mean absolute error (MAE) of 5.28 mmHg for systolic BP and 4.92 mmHg for diastolic BP.
• The proposed method achieved results comparable to existing approaches utilizing single PPG signals.

Conclusions:

• A novel CNN-based approach enables accurate blood pressure prediction using smartphone PPG signals without manual feature extraction.
• Combining green PPG signals with instantaneous frequency offers a promising strategy for enhanced BP estimation.
• This method provides a convenient and potentially scalable solution for remote and frequent blood pressure monitoring.