Deep Learning Strategy for Sliding ECG Analysis during Cardiopulmonary Resuscitation: Influence of the Hands-Off Time...

Area of Science:

• Medical technology
• Artificial intelligence in healthcare
• Emergency medicine

Background:

• Automated external defibrillators (AEDs) are crucial for out-of-hospital cardiac arrest (OHCA) interventions.
• Optimizing shock advisory decisions during cardiopulmonary resuscitation (CPR) is critical for patient outcomes.
• Current AED algorithms may be limited by chest compression interruptions and hands-off time.

Purpose of the Study:

• To develop and evaluate a novel deep learning algorithm for sliding shock advisory decisions during CPR.
• To assess the algorithm's performance as a function of cumulative hands-off time (sHOT).
• To improve the accuracy and efficiency of AED rhythm analysis in real-time CPR scenarios.

Main Methods:

• Retrospective analysis of 13,570 CPR episodes from OHCA interventions.
• Application of three convolutional neural networks (CNNs) with varying ECG input durations (5, 10, 15 s) for sequential analysis.
• Quantification of cumulative chest compression-free time using sliding hands-off time (sHOT).

Main Results:

• A 10-second CNN model demonstrated optimal performance, achieving high sensitivity for ventricular fibrillation (VF) and accuracy across various rhythms (VF, ASYS, ONR, NSR).
• VF sensitivity significantly improved with increased analysis duration from 5s to 10s, while specificity remained consistent.
• sHOT was identified as a key predictor of performance, with reliable rhythm detection achieved at minimal sHOT intervals of 2-3 seconds, meeting AHA standards.

Conclusions:

• The novel deep learning algorithm offers substantial performance improvements for sliding shock advisory decisions during CPR, particularly in short hands-off periods.
• This technology supports improved CPR practices by enabling continuous chest compressions with minimal interruptions for rhythm analysis.
• The algorithm provides a basis for enhanced AED functionality, facilitating earlier VF treatment and more effective non-shockable rhythm management.