Convolution Neural Network Algorithm for Shockable Arrhythmia Classification Within a Digitally Connected Automated...

Area of Science:

• Cardiology
• Biomedical Engineering
• Artificial Intelligence

Background:

• Automated external defibrillators (AEDs) are crucial for out-of-hospital cardiac arrest (OHCA) outcomes.
• Accurate diagnosis of shockable rhythms by AED algorithms is essential for timely defibrillation.
• Machine learning, specifically CNNs, offers advanced capabilities for arrhythmia detection on ECGs.

Purpose of the Study:

• To develop and validate a CNN algorithm for diagnosing shockable arrhythmias.
• To integrate this CNN into a novel, miniaturized AED for real-time analysis.
• To assess the CNN's diagnostic performance and robustness against noise and misclassification.

Main Methods:

• Utilized a dataset of 26,464 single-lead ECGs (7-s duration).
• Trained and tested a CNN model on partitioned ECG data (training, validation, test sets).
• Evaluated CNN performance using ROC analysis, F1 scores, sensitivity, and specificity; assessed processing time and robustness.

Main Results:

• The CNN achieved an AUC of 0.995, 98% sensitivity, and 100% specificity for shockable rhythms.
• F1 scores were 0.990 for shockable and 0.995 for nonshockable rhythms.
• The algorithm processed ECGs in 383±29 ms, outperformed human readers in specific classifications, and showed robustness to noise.

Conclusions:

• A CNN algorithm demonstrates high diagnostic accuracy for shockable and nonshockable arrhythmias in an AED.
• The developed CNN is suitable for integration into digitally connected, miniaturized AEDs.
• This AI-driven approach has the potential to enhance OHCA resuscitation outcomes.