A Deep-Learning-Based CPR Action Standardization Method.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Emergency Medical Services

Background:

• Standardized cardiopulmonary resuscitation (CPR) is critical in emergencies.
• Current automated external defibrillators (AEDs) lack mechanisms for real-time CPR quality assessment.
• Inconsistent CPR quality can negatively impact patient outcomes.

Purpose of the Study:

• To develop and validate a novel deep-learning-based method for standardizing CPR actions.
• To enhance the accuracy and real-time feedback of CPR performance monitoring.
• To optimize CPR quality assessment for integration into emergency medical devices.

Main Methods:

• Utilized OpenPose for skeletal point recognition to detect correct CPR posture.
• Developed a CPR-Detection algorithm to identify a marker wristband and measure compression depth, count, and frequency.
• Employed a depth algorithm for precise compression depth analysis.
• Optimized the algorithm for edge devices to ensure rapid, real-time processing.

Main Results:

• The CPR-Detection algorithm achieved a mean Average Precision (mAP0.5) of 97.04% with reduced computational load (0.20 M parameters, 132.15 K FLOPs).
• The depth measurement solution demonstrated 90% accuracy with a <1 cm error margin.
• Compression count and frequency measurements achieved 98% accuracy with a <2 count error margin.
• Real-time processing speed on edge devices increased from 8 fps to 25 fps.

Conclusions:

• The deep-learning-based CPR action standardization (DLCAS) method effectively standardizes CPR quality.
• The system meets real-time processing requirements for medical scenarios.
• This technology has the potential to significantly improve CPR effectiveness and patient survival rates.