Neural Network Model Combination for Video-Based Blood Pressure Estimation: New Approach and Evaluation.

Area of Science:

• Biomedical Engineering
• Computer Vision
• Machine Learning

Background:

• Blood pressure is a critical vital sign, but traditional cuff-based monitoring is inconvenient and costly.
• Changes in facial skin color intensity correlate with blood pressure variations.
• Existing methods lack the convenience and accessibility needed for widespread monitoring.

Purpose of the Study:

• To develop a novel, non-invasive blood pressure estimation method using facial video analysis.
• To leverage hybrid deep learning models for accurate blood pressure prediction.
• To validate the proposed approach against existing methods and datasets.

Main Methods:

• Utilized hybrid deep learning models to analyze RGB color channel intensities from facial videos.
• Trained and validated models on the Vision for Vitals (V4V) dataset.
• Introduced a new evaluation metric based on Pearson's correlation coefficient with respiratory rate.

Main Results:

• The proposed hybrid deep learning models demonstrated competitive performance on the V4V dataset.
• The novel approach offers a cuff-less, fast, and comfortable blood pressure estimation.
• The new metric provides an additional layer of performance evaluation.

Conclusions:

• Facial video analysis with hybrid deep learning is a viable method for non-invasive blood pressure estimation.
• This technology has the potential to revolutionize personal health monitoring.
• Future research can further refine the models and explore broader applications.