Video Based Cocktail Causal Container for Blood Pressure Classification and Blood Glucose Prediction.

Area of Science:

• Biomedical Engineering
• Computer Vision
• Machine Learning

Background:

• Modern cameras on portable devices enable physiological signal acquisition.
• Non-invasive, hemodynamically-based video processing for health monitoring is emerging but limited.
• Existing methods face challenges in extracting discriminative features from video data.

Purpose of the Study:

• To propose an end-to-end framework, 'cocktail causal container,' for multi-task learning.
• To fuse multiple physiological representations and reconstruct frequency-temporal correlations.
• To enable unobtrusive blood pressure classification and blood glucose prediction using smartphone cameras.

Main Methods:

• Developed a 'cocktail causal container' framework for processing hematologic reflex information.
• Introduced a token feature fusion block to create a unified discrete frequency space from multi-view representations.
• Utilized a causal net for analyzing fused information and a pair-wise temporal frequency map for photoplethysmograph (PPG) extraction.

Main Results:

• Achieved a root mean square error of 1.329±0.167 for blood glucose prediction on a clinical dataset.
• Attained a precision of 0.89±0.03 for blood pressure classification on the same dataset.
• Validated the framework using a clinical dataset and the PPG-BP benchmark.

Conclusions:

• The 'cocktail causal container' framework effectively fuses multi-modal physiological data from videos.
• The proposed method demonstrates high accuracy in blood pressure classification and blood glucose prediction.
• This approach offers a promising direction for non-invasive, at-home physiological monitoring.