WSEAS Transactions on Biology and Biomedicine
Print ISSN: 1109-9518, E-ISSN: 2224-2902
Volume 21, 2024
Chest Compression Evaluation based on Pose Estimation
Authors: , , , , ,
Abstract: Correct and prompt performance of cardiopulmonary resuscitation yields improvements in mortality and social return rates. Chest compression, a vital cardiopulmonary resuscitation technique, requires regular re-education for skill maintenance. Training with a manikin is feasible for chest compression, but assessing proficiencies without an expert presents challenges. This study aims to facilitate autonomous chest compression training even without expert supervision based on pose estimation. Twenty subjects were recruited for the training and successive performance evaluation of chest compression on a sensor-equipped training manikin, and the corresponding videos were recorded simultaneously. A system was developed to analyze chest compression movements through pose estimation on recorded videos for evaluating interruption presence, compression count, compression tempo, compression depth, and compression recoil. Through comparing three pose estimation models, OpenPose demonstrated the best performance, achieving accuracy rates of 67.08%, 56.67%, 61.25%, 39.17%, and 33.75% for the detection of interruption presence, compression count, appropriate tempo count, appropriate depth count, and appropriate recoil count, respectively. Additionally, posture analysis during compression, unattainable with the sensor-equipped manikin, revealed effectiveness in shooting at a position shifted 45 degrees from the front. The proposed method may serve as a tool for completely automated CPR chest compression training, anticipating an increase in citizens proficient in cardiopulmonary resuscitation.
Search Articles
Keywords: cardiopulmonary resuscitation, chest compression, compression count, deep learning, pose estimation, and sudden cardiac death
Pages: 323-330
DOI: 10.37394/23208.2024.21.32