Detecting Mistakes in CPR Training with Multimodal Data and Neural Networks.

Area of Science:

• Medical Education Technology
• Biomedical Engineering
• Human-Computer Interaction

Background:

• Cardiopulmonary Resuscitation (CPR) is a critical life-saving skill requiring precise technique.
• Current CPR training manikins offer limited feedback on certain performance aspects.
• Objective assessment of CPR skills is essential for effective training and improved patient outcomes.

Purpose of the Study:

• To evaluate the efficacy of a multimodal data system for detecting errors during CPR training.
• To compare the performance of the multimodal system against existing CPR manikin feedback.
• To identify additional CPR performance indicators detectable by multimodal sensing.

Main Methods:

• Utilized a multi-sensor system (Microsoft Kinect, Myo armband) integrated with a CPR manikin.
• Collected multimodal data from 11 medical students performing chest compressions (CCs).
• Trained five neural networks to classify five performance indicators, including rate, depth, release, arm, and body position.

Main Results:

• Multimodal data achieved accurate mistake detection, exceeding the baseline performance of the ResusciAnne manikin.
• The system successfully identified errors in arm and body positioning, which are typically assessed manually.
• The study demonstrated the potential for automated detection of nuanced CPR performance errors.

Conclusions:

• Multimodal data fusion offers a robust approach to enhancing CPR training feedback.
• The developed Multimodal Tutor for CPR can provide comprehensive, objective assessments of trainee performance.
• Future implementations could significantly improve the quality and consistency of CPR education.