ABSTRACTS
Video-Based Analysis for Carotid Pulse DetectionAuthor: Ryo Sagisaka | Lecturer | Research Institute of Disaster Management and Emergency Medical System, Kokushikan University Associate Authors: Inoue, Yumemi, Ms. | Nakagawa, Koshi, Dr.
Objective Out-of-hospital cardiac arrest (OHCA) affects approximately 140,000 cases annually in Japan, with bystander CPR at 51.8% and one-month survival 6.4%. A critical barrier to bystander CPR implementation is the difficulty in rapid cardiac arrest recognition by laypersons. Traditional assessment methods focusing on breathing and chest movement are unreliable due to clothing thickness and patient positioning. We investigated smartphone-based video analysis for carotid pulse detection as a novel approach to enhance cardiac arrest recognition accuracy. Methods We enrolled 30 healthy university students for carotid artery analysis and 25 for radial artery comparison (22 males total, mean age 22.2±1.3 years). Using smartphone cameras (Galaxy S22 Ultra), we recorded 60-second neck videos (1920×1080 pixels, 15fps) under standardized conditions with custom stabilizing stands. To evaluate pulse detection accuracy, we recorded radial artery videos with and without tourniquet compression to simulate pulseless conditions. Video analysis employed Eulerian Video Magnification (EVM) with 4-level Laplacian pyramid decomposition and Butterworth bandpass filtering (0.67-1.17Hz) with 10-fold amplification, followed by optical flow analysis. Three motion indicators were calculated: mean motion, maximum motion, and principal component analysis (PC1). Statistical analysis included Pearson’s correlation with physiological parameters and logistic regression using generalized estimating equations. Results EVM processing successfully visualized carotid pulsations in all participants. Mean motion showed significant negative correlation with body fat percentage (r = -0.44, p = 0.014) and positive correlation with pulse pressure (r = 0.65, p < 0.001). Carotid and radial artery mean motion values demonstrated moderate correlation (r = 0.50, p = 0.012). However, logistic regression analysis revealed that none of the motion indicators were effective predictors for pulse detection (Mean motion OR: 0.71, 95% CI: 0.43-1.17; Max motion OR: 0.99, 95% CI: 0.99-1.001; PC1 OR: 0.99, 95% CI: 0.88-1.13). Conclusions While video-based carotid pulse visualization is technically feasible using smartphone technology, current motion analysis algorithms are insufficient for reliable cardiac arrest detection. The significant correlations with physiological parameters and moderate inter-arterial correlation suggest potential for algorithm refinement. Future research requires validation with actual cardiac arrest patients and optimization for emergency conditions.
|
