A probabilistic neural network as the predictive classifier of out-of-hospital defibrillation outcomes

Introduction:
Although modern defibrillators are nearly always successful in terminating ventricular fibrillation (VF), multiple defibrillation attempts are usually required to achieve return of spontaneous circulation (ROSC). This is potentially deleterious as cardiopulmonary resuscitation (CPR) must be discontinued during each defibrillation attempt which causes deterioration in the heart muscle and reduces the chance of ROSC from later defibrillation attempts. In this work defibrillation outcomes are predicted prior to electrical shocks using a neural network model to analyse VF time series in an attempt to avoid defibrillation attempts that do not result in ROSC.
Methods:
The 198 pre-shock VF ECG episodes from 83 cardiac arrest patients with defibrillation conversions to different outcomes were selected from the Oslo ambulance service database. A probabilistic neural network model was designed for training and testing with a cross validation method being used for the better generalisation performance.
Results:
We achieved an accuracy of 75% in overall prediction with a sensitivity of 84% and a specificity of 65% using VF ECG time series of an order of 1s in length.
Conclusion:
Pre-shock VF ECG time series can be classified according to the defibrillation conversion to a return of spontaneous circulation (ROSC) or No-ROSC.