Comparison of the Efficacy and Noise Immunity of Algorithms for Recognition of Shockable Cardiac Rhythms

Abstract

According to the terminology of the American Heart Association (AHA), shockable rhythms (SR) are lethal cardiac rhythms that terminate in the patient’s death unless defibrillation is delivered very quickly. Rhythms that must not be shocked are called nonshockable (NSR). Rhythms for which the benefits of defibrillation are limited or uncertain are classified as intermediate (IR) [13]. The efficacy (sensitivity and selectivity) of detection (recognition) of SR is assessed using special test signals. The sensitivity is determined as the ratio of the number of correctly identified SR segments of the test signal to their total number. The selectivity is determined as the ratio of the number of correctly identified NSR segments of the test signal to their total number. Detection of noise-induced change in the sensitivity and selectivity makes it possible to determine the noise immunity of the recognition algorithms. The algorithms for SR recognition can be classified into three groups: time-domain methods, frequencydomain methods, and combined methods. Among linear algorithms of the first group, the following are especially widely used: the method of threshold crossing intervals [4, 17] and the method of linear regression analysis of the autocorrelation function [3]. Spectral analysis of electrocardiosignals [2] and the method of rejection filtering of ventricular fibrillation [14] are the most widely methods of the second group. Linear algorithms of the third group are implemented as combinations of several algorithms [12]. In particular, the complexity measure method [18] belongs to this group. Among the nonlinear algorithms, methods based on various types of neural networks, correlation analysis of signal, and use of reference patterns are widely favored [8, 11, 16]. Nonlinear algorithms are very effective, but rather difficult to implement because of high requirements for computing facilities. The goal of this work was to compare linear recognition algorithms, which are sufficiently simple to be implemented in a portable device, in particular, in an automatic external defibrillator (AED). It is shown that the method of rejection filtering of ventricular fibrillation is optimal for this purpose. The combination of the method of threshold crossing intervals and the spectral analysis method is shown to be optimal for implementation in a portable external defibrillator.