Abstract 15563: Analysis of the Ventricular Fibrillation Waveform Can Guide Resuscitation in Cardiac Arrest by Identifying the Precipitating Cause and Estimating the Underlying Hemodynamics

Abstract

Background: Patients presenting with out of hospital cardiac arrest (OHCA) and prolonged (>30 min) cardiopulmonary resuscitation (CPR) face nearly 100% mortality. In patients presenting with ventricular fibrillation (VF), coronary ischemia is thought to be the cause in up to 60%. As conventional CPR does not reverse a coronary ischemia, CPR often must be done until a patient can be brought to a cardiac catheterization laboratory (CCL) where the underlying cause (acute ischemia) can be treated. In a porcine model of cardiac arrest, we have previously demonstrated that machine learning (ML) targeting invasive hemodynamic monitoring, could significantly improve CPR quality. To advance the clinical relevance of the machine-controlled CPR methods, we developed ML algorithms that can diagnose coronary ischemia during VF, and infer the underlying hemodynamics while performing CPR. Methods: In a porcine model of cardiac arrest, VF was induced via high-grade coronary occlusion (ischemia) or electrical stimulation (dysrhythmia). The animals (n = 72) were left in VF for 3 minutes prior to intervention and the ECG waveform was collected. The VF signal was converted to the frequency domain using a fast-fourier transformation and an ensemble classifier was developed to detect underlying ischemia. CPR was then initiated and the VF waveform was continuously monitored along with the coronary perfusion pressure (CPP). We then developed a 20-layer convolutional neural network (CNN) to non-invasively infer the CPP from a 10-second VF waveform. Results: Animal studies were split into training and test sets. When using the developed ML technique to detect ischemia from the VF waveform, evaluation of the test set yielded an area under the curve (AUC) of 0.81 on a ROC. When we used the CNN to infer the underlying hemodynamics, we demonstrated a high correlation between measured and predicted CPP in the test set. The AUC for the ROC when predicting a CPP of >15 mmHg was 0.85. Conclusion: Survival for patients with OHCA secondary to coronary ischemia will improve with rapid diagnosis and high quality CPR. Our results demonstrate the proof-of-concept that the VF waveform contains features that can guide resuscitation for cause identification and hemodynamic improvement.