Study of frequency and time domain parameters extracted by means of wavelet transform applied to ECG to distinguish between VF and other arrhythmias

Implementation of algorithms to detect VF and other malignant arrhythmias has been the subject of many publications, using methods based on frequency and time-domain methods. In order to take profit of information obtained from both domains we have implemented a method for single ECG-lead signal processing and arrhythmia discrimination, based on Wavelet Transform (WT) and advanced statistical decision methods. The model has been tested for discrimination among malignant ventricular arrhythmias (VF and VT, amenable to be cardioverted-defibrillated) and other arrhythmias, according to the AHA expert committee recommendations. Single-lead 4096 ms-long episodes of different rhythms and from different patients were selected from international (AHA. VF-DB&CU-DB from MITBIH) and custom (CCU monitor and EP Laboratory) databases. All the signals were 250 Hz digitized and transformed into time-frequency domain by WT. First 5 scales were computed: temporal parameters were calculated from the second and third scales, obtaining more than 40 parameters. After statistical parameter reduction discriminating functions were obtained through logistic regression and validated using intensive scanning techniques. Morphological analysis of single ECG-lead by means of WT and statistically constructed discriminant functions can accurately diagnose malignant ventricular arrhythmias in a fast and low computational cost algorithm, according to the AHA recommendations