[A non-contact continuous blood pressure measurement method based on video stream]

Area of Science:

• Biomedical Engineering
• Signal Processing
• Medical Imaging

Context:

• Hypertension poses a significant global health risk.
• Accurate and convenient blood pressure monitoring is crucial for hypertension prevention and management.
• Existing methods can be invasive or inconvenient for continuous monitoring.

Purpose:

• To develop and validate a non-contact, continuous blood pressure measurement method using facial video signals.
• To extract and analyze pulse wave signals from facial videos.
• To establish a robust blood pressure prediction model based on extracted video features.

Summary:

• A novel method extracts video pulse waves from facial regions using color distortion filtering and independent component analysis.
• Multi-dimensional features were extracted in the time-frequency domain, followed by integrated feature selection.
• Blood pressure prediction models were compared using Elman neural networks, SVM, and deep belief networks, with SVM demonstrating superior performance.
• The developed SVM model achieved mean absolute errors of 4.9 mm Hg (systolic) and 4.6 mm Hg (diastolic), meeting AAMI standards.

Impact:

• Presents a viable non-contact approach for continuous blood pressure monitoring.
• Offers a convenient and potentially more accessible alternative to traditional blood pressure measurement devices.
• Potential to improve hypertension screening and management through accessible video technology.