Introduction: Survival from out-of-hospital cardiac arrest (OHCA) relies heavily on early identification and defibrillation of ventricular fibrillation (VF). Therefore, the aim of this study is to develop and test an automated method based on a novel machine learning technique to detect VF.

Materials and methods: The dataset contained ECG segments from 169 OHCA patients treated by Tualatin Valley Fire & Rescue (Tigard, OR, USA) using the Philips HeartStart MRx monitor/defibrillator. The dataset was composed of 596 10-s ECG segments, 144 shockable and 452 non-shockable, annotated by consensus by a pool of four emergency medicine doctors. The dataset was split patient-wise into training (60%) and test (40%) sets. Each ECG segment was band-pass filtered (1-30 Hz), waveform features were calculated and fed to a state of the art machine learning algorithm, a support vector machine (SVM) classifier with radial basis function kernel. The SVM diagnosed each segment as shockable or non-shockable. The training set was used to select the most discriminative feature subset (incremental selection of maximum 5 features), and to tune the hyperparameters of the SVM through patient-wise 10-fold cross validation. The test set was used to compute the performance of the method in terms of sensitivity (SE) and specificity (SP). This procedure was repeated 500 times to estimate the distributions of SE and SP.

Results:  The SVM showed a mean (standard deviation) SE and SP of 96.5% (2.5) and 97.0% (1.4), respectively. The method met the minimum performance requirements of the American Heart Association (SE>90% and SP>95%). The method required on average only 279 (36) ms per segment in a standard platform.

Conclusion: An automated method based on a state of the art machine learning technique accurately detects VF during OHCA. Its low computational cost makes it suitable for implementation into current defibrillators.

6 - A machine learning approach for detecting ventricular fibrillation during out-of-hospital cardiac arrest