Predicting defibrillation outcomes by combining ventricular fibrillation and defibrillation waveforms: a...

Area of Science:

• Cardiology
• Biomedical Engineering
• Artificial Intelligence in Medicine

Background:

• Defibrillation success prediction is crucial for cardiac arrest management.
• Current methods, like amplitude spectrum area (AMSA), have limitations in predicting defibrillation outcomes.
• Integrating pre-shock ventricular fibrillation (VF) and post-shock defibrillation (DF) waveforms may offer improved predictive capabilities.

Purpose of the Study:

• To validate if combining VF and DF waveforms enhances the accuracy of predicting DF outcomes.
• To assess the performance of a deep learning model utilizing combined VF and DF waveforms against traditional metrics.

Main Methods:

• Electrocardiographic waveforms from VF and DF events were recorded.
• DF waveforms were modeled, and errors were assessed.
• A convolutional neural network combined uncorrupted VF and modeled DF waveforms.
• Model performance was evaluated using area under the receiver operating characteristic curve (AUC) for termination of VF (TOVF), return of organized rhythm (ROOR), and return of potentially perfusing rhythm (RPPR).

Main Results:

• The combined waveform model demonstrated significantly higher AUC values compared to AMSA for TOVF (0.627 vs. 0.578), ROOR (0.854 vs. 0.804), and RPPR (0.873 vs. 0.836).
• The most substantial improvements were observed in shocks with AMSA values between 4.6 and 15.0 mVHz.
• AUC increases for TOVF, ROOR, and RPPR were 9.0, 7.3, and 5.1 points, respectively, in this range.

Conclusions:

• Combining VF and DF waveforms via a deep learning approach significantly improves the prediction accuracy of defibrillation outcomes.
• This method proves effective regardless of the specific criteria used to define defibrillation success.
• The findings suggest a promising new avenue for real-time prediction of defibrillation efficacy in cardiac arrest patients.