Deep Neural Network Approach for Continuous ECG-Based Automated External Defibrillator Shock Advisory System During...

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Devices

Background:

• Automated external defibrillators (AEDs) require pauses in cardiopulmonary resuscitation (CPR) for rhythm analysis, interrupting vital chest compressions.
• Minimizing CPR interruptions is crucial for improving patient survival rates during cardiac arrest.
• CPR artifacts in electrocardiogram (ECG) signals complicate automated rhythm interpretation.

Purpose of the Study:

• To develop and validate a deep-learning algorithm for classifying shockable versus nonshockable rhythms.
• To assess the algorithm's performance in the presence and absence of CPR-induced artifacts.
• To enable AEDs to perform rhythm analysis without pausing CPR.

Main Methods:

• A deep-learning model incorporating convolutional layers, residual networks, and bidirectional long short-term memory was employed.
• The model was trained and tested on ECG data from 40 subjects, including 1131 shockable and 2741 nonshockable samples with 43 types of CPR artifacts.
• Performance was evaluated using 4-fold cross-validation and leave-one-subject-out validation.

Main Results:

• The deep-learning model achieved a sensitivity of 95.21% and specificity of 86.03% for rhythm classification across the entire dataset.
• For ECG data with CPR artifact, sensitivity was 94.21% and specificity was 86.14%.
• Leave-one-subject-out validation yielded a sensitivity of 92.71% and specificity of 97.6%.

Conclusions:

• The developed deep-learning model reliably distinguishes between shockable and nonshockable rhythms, irrespective of CPR artifact presence.
• The model meets the American Heart Association's sensitivity requirement of >90% for AEDs.
• This technology has the potential to allow continuous CPR during rhythm analysis, improving resuscitation outcomes.