Contactless Blood Pressure Measurement Via Remote Photoplethysmography With Synthetic Data Generation Using Generative Adversarial Networks | JoVE Visualize

Area of Science:

Biomedical Engineering
Computer Vision
Artificial Intelligence

Background:

Remote photoplethysmography (rPPG) enables contactless vital sign monitoring, including blood pressure (BP).
Existing rPPG methods for BP estimation often rely on time/phase differences or single-signal waveform analysis.
Developing accurate and robust contactless BP measurement remains a significant challenge.

Purpose of the Study:

To develop a novel deep learning model for accurate contactless blood pressure estimation using rPPG.
To extract and utilize time-difference features from rPPG signals for improved BP prediction.
To enhance model generalizability and prevent overfitting through multi-model design, synthetic data generation, and subject information integration.

Main Methods:

Extraction of time-difference features from 30 FPS rPPG signals.
Proposal of a deep learning architecture for BP estimation from extracted features.
Leveraging multi-model design, synthetic data generation (InfoGAN), and subject information (real and estimated) to combat data scarcity and overfitting.
Evaluation on self-constructed and TVGH datasets, including dynamic and nighttime BP measurements.

Main Results:

Significant reduction in Mean Absolute Error (MAE) for systolic blood pressure (SBP) from 15.49 to 8.78 mmHg and diastolic blood pressure (DBP) from 10.56 to 6.16 mmHg on a self-constructed dataset.
On the TVGH dataset for nighttime BP measurement, MAE for SBP decreased from 21.58 to 11.12 mmHg (48.47% improvement) and for DBP from 9.74 to 7.59 mmHg (22.07% improvement).
The proposed method outperforms state-of-the-art approaches, demonstrating robustness with estimated and real subject information.

Conclusions:

The developed deep learning model effectively estimates blood pressure using rPPG signals by incorporating time-difference features and subject-specific information.
The combination of advanced feature extraction, synthetic data generation, and subject information significantly enhances BP measurement accuracy and reliability.
This contactless rPPG-based BP monitoring system shows promising results for both general and challenging nighttime conditions, offering a non-invasive alternative for health assessment.