Predicting patterns of epicardial potentials during ventricular fibrillation

Area of Science:

• Cardiology
• Computational Biology
• Biomedical Engineering

Background:

• Ventricular fibrillation (VF) is a life-threatening cardiac arrhythmia.
• VF involves uncoordinated activation wavefronts in the ventricular myocardium.
• Accurate prediction of VF dynamics is crucial for therapeutic interventions.

Purpose of the Study:

• To assess the feasibility of short-term prediction of epicardial potential fields during VF.
• To evaluate the accuracy of linear prediction models at different stages of VF.
• To investigate factors influencing prediction accuracy, such as spatial complexity and VF duration.

Main Methods:

• VF was induced in pigs, and unipolar electrograms were recorded using a high-density epicardial electrode array.
• Karhunen-Loeve decomposition was employed to identify optimal spatial basis functions (modes).
• Linear autoregressive (AR) models were developed using a few dominant spatial modes to predict future potential fields.

Main Results:

• Linear AR models generated qualitatively similar patterns to observed epicardial potentials.
• Predictions achieved a temporal accuracy of 0.256 seconds into the future.
• Prediction accuracy, measured by normalized mean squared error, ranged from 0.14 to 1.23.
• Predictability was inversely related to spatial complexity and increased significantly within the first minute of VF.

Conclusions:

• Linear AR models can provide short-term predictions of epicardial potential fields during VF.
• Predictability is influenced by the spatial complexity of VF and improves early in the episode.
• Limitations of linear models highlight the need for more advanced techniques for long-term forecasting.