Skip to Main Content
Ventricular Fibrillation(VF) waveform can represent rapidly worsening chances of defibrillation success, and those of subsequent Return of Spontaneous Circulation (ROSC), during a cardiac arrest. We propose a new method to analyze the chaotic nature of VF using multiple feature extraction and machine learning techniques. Human cardiac arrest data was acquired from the Richmond Ambulance Authority. A Multiple-Domain Model (MDM), which utilizes time-series and wavelet features, was developed. We report two new time-series features that are predictive of countershock (CS) success. Support vector machines were used with a radial basis function to classify 56 CS, 21 successful and 35 unsuccessful, with an average accuracy of 83.9%. Sensitivity and specificity were 71.4% and 91.4%, respectively. ROC area under the curve of 81.4% was achieved. The proposed predictive model performs real-time, short-term analysis of ECG, through signal-processing and machine-learning techniques, and can be accurate enough for clinical application. As more cardiac arrest data is acquired, improved MDM performance is anticipated.