Feature Selection Methods to Extract Knowledge and Enhance Analysis of Ventricular Fibrillation Signals

Background

Fibrillatory rhythms, as atrial and ventricular fibrillation, have traditionally been analysed with cardiac mapping systems (Nash et al., 2006). These systems capture signals in hundreds of points of the cardiac tissue by means of an electrode array placed either in the epicardium or in the endocardium. The electrode array simultaneously captures a high number of electrograms, which can be used to extract some parameters to describe the fibrillatory process. Such description is usually performed through parameters that measure the regularity, despite fibrillation has been traditionally considered a chaotic and irregular process (Jalife, 2000). The regularity can be measured in time and frequency domains. Besides, there are other parameters related to the fibrillatory frequency content that can provide information related to the activation rate of the tissue.

The Dominant Frequency (DF) of a fibrillation process is defined as the highest peak on its power spectrum, and it is both related to the mean activation period and to the cardiocytes refractoriness period (Sanders et al., 2005). DF is one of the most commonly used parameters in fibrillatory rhythms due to its direct relation with these physiological characteristics.

Other spectral parameter is commonly used in order to measure the spectral regularity of a fibrillatory electrogram: the Normalized Energy (NE). It is defined as the ratio between the spectral energy in a frequency window (DF±1Hz) and the spectral energy in the interest frequency band (Sanders et al., 2005). NE is related to spectral complexity; a higher value of NE implies that spectral activity is focused on DF, i.e. the activity is lesser complex.