Cascade Skip-Connection BiLSTM Autoencoder for CPR Artifact Removal Prior to AED Shock Advisory

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Cardiovascular Research

Background:

• Current automated external defibrillators (AEDs) necessitate interruptions in cardiopulmonary resuscitation (CPR) for electrocardiogram (ECG) rhythm analysis.
• These pauses reduce chest compression fraction and delay critical defibrillation, negatively impacting patient outcomes.

Purpose of the Study:

• To develop an automated method for removing CPR-induced artifacts from ECG signals in real-time.
• To enable continuous rhythm analysis for AEDs without requiring pauses in CPR.

Main Methods:

• A skip-connection BiLSTM autoencoder (SBAE) architecture was designed to directly process 1D ECG signals, eliminating the need for auxiliary signals or time-frequency transformations.
• A cascade approach utilized balanced and biased denoising models, with a conservative routing strategy to identify indeterminate cases.

Main Results:

• The SBAE achieved high performance metrics, including 97.51% sensitivity for ventricular fibrillation and 99.30% specificity for normal sinus rhythm, exceeding American Heart Association guidelines.
• The system demonstrated excellent performance across various cardiac rhythms, with an indeterminate rate of 3.63%.

Conclusions:

• The proposed SBAE framework offers a compact, reference-free solution for real-time shock advisory decision support in AEDs.
• Further optimization and platform-specific engineering are required for deployment on embedded AED hardware.