Detection of shockable rhythm using multi-parameter fusion identification and BP neural network

Abstract

The widening application of automated external defibrillators (AEDs) present very strong requirements for reliable shockable rhythm detection. In this study, we developed a BP neural network to differentiate well between shockable and nonshockable rhythm. A total of 18 metrics were extracted from the ECG signals. Each one of these metrics respectively characteristics each aspect of the signals, such as morphology, gaussianity, spectra, variability, complexity, and so on. These metrics were regarded as the input vector of the BP neural network. After the training, a classifier used for shockable and nonshockable rhythm classification was obtained. The constructed BP neural network was tested with the database of VFDB and CUDB, the sensitivity and specificity reached up to 93.04% and 97.43 %, respectively.