A time-series approach for shock outcome prediction using machine learning

Chances of successful defibrillation, and that of subsequent return of spontaneous circulation (ROSC), worsen rapidly with passage of time during cardiac arrest. The Electrocardiogram (ECG) signal of ventricular fibrillation (VF) has been analyzed for certain characteristics which may be predictive of successful defibrillation. Time-series features were extracted. A total of 59 counter shocks (CS) were analyzed. They were best classified as successful or unsuccessful by employing the Random Tree method. An average accuracy of 71% was achieved for 6 randomized runs of 6-fold cross validation. Classification could be performed on ECG tracings of 40 seconds. Real-time, short-term analysis of ECG, through signal-processing and machine-learning techniques, may be valuable in determining CS success.