Abstract 14841: A Video-based Cardiopulmonary Resuscitation Analysis System “See-CPR”
Objectives: Various devices, either expensive or inconvenient, were developed to monitor the quality of cardiopulmonary resuscitation (CPR), which is crucial to resuscitation performance and patient survival in cardiac arrest events. A video-based CPR analysis system, called “See-CPR”, was designed to automatically detect and analyze CPR qualities.
Methods: To automatically detect chest compression (CC) movements in videos documenting CPR, we first estimated the motion of objects using the motion vectors of MPEG videos. We extracted representative features and used a hierarchical detecting scheme, including frame-level detection and group-level classification, to determine the location of CC occurrence in both time and spatial domains. Compression rate, chest compression duration, and hands-off time can then be shown on videos simultaneously. (Figure 1) To determine the precision and recall, the number of detected CC was compared with real CC, which was detected by CPR reporting software (SkillReporter, Laerdal, Norway) connected to a mannequin (Resusci Anne, Laerdal, Norway). (Figure 2)
Results: Five video sequences, which recorded CC performance by different subjects, were used as test data for CC detection. The overall precision and recall achieved 99.7% and 100% respectively. The only one falsely detected CC was intentionally incorrect performance.
Conclusions: A reliable, video-based CPR detection and analysis system for automatically reporting real-time CPR qualities was proposed. It can be used for monitoring, real-time feedback and training of resuscitation.
- © 2011 by American Heart Association, Inc.