Related Experiment Video
Updated: Jun 7, 2025
10:44
Three-Dimensional Phase Resolved Functional Lung Magnetic Resonance Imaging
Published on: June 21, 2024
447
Feasibility of real-time compression frequency and compression depth assessment in CPR using a "machine-learning"
Hannes Ecker1,2, Niels-Benjamin Adams1,2, Michael Schmitz3
1University of Cologne, Faculty of Medicine, Cologne, Germany.
Resuscitation Plus
|November 19, 2024
Summary
An artificial intelligence tool accurately measured cardiopulmonary resuscitation (CPR) compression frequency in simulated videos. However, the AI showed inaccuracies in detecting compression depth, indicating a need for further development before real-world application in video-assisted CPR.
Area of Science:
- Emergency Medicine
- Artificial Intelligence in Healthcare
- Cardiopulmonary Resuscitation
Background:
- Video-assisted cardiopulmonary resuscitation (V-CPR) improves CPR quality and patient outcomes by enabling Emergency Medical Service (EMS) dispatchers to guide callers via video.
- EMS dispatchers face challenges in V-CPR, including video analysis, real-time feedback, and stress management, necessitating innovative solutions.
- This study investigates the feasibility of using an open-source AI tool to assess compression frequency and depth in simulated V-CPR scenarios.
Purpose of the Study:
- To evaluate the feasibility and accuracy of an open-source AI tool for detecting compression frequency and depth in video footage of simulated CPR.
- To explore the potential of machine learning in providing real-time feedback during V-CPR.
Main Methods:
- MediaPipe Pose Landmark Detection, an open-source AI software, was programmed to analyze nine videos of CPR on a manikin.
- The AI assessed compression frequency and depth for each compression.
- AI-derived measurements were compared against the manikin's internal software (QCPR) using statistical methods including the Wilcoxon matched-pairs signed rank test and Bland Altman analysis.
Main Results:
- MediaPipe Pose Landmark Detection successfully identified and tracked the CPR performer in all video sequences.
- High agreement was observed between AI-calculated compression frequencies and those from the manikin's software.
- Significant inaccuracies were found in the AI's assessment of compression depth, rendering it unreliable.
Conclusions:
- Open-source machine learning tools show potential for real-time feedback in V-CPR video analysis.
- The AI tool accurately assessed CPR compression frequency but requires adjustments for reliable compression depth measurement.
- Further development is needed before this AI technology can be implemented in actual CPR scenarios.
