Beat-to-beat ECG ventricular late potentials variance detection by filter bank and wavelet transform as beat-sequence filter

This paper presents a novel method that employs a wavelet transform and filter bank to detect ventricular late potentials (VLPs) from beat to beat in order to keep its variance. Conventionally, three time-domain features, which are highly related to the QRS complex endpoint, are generally accepted as criteria for classifying VLPs. Signal averaging is a general and effective de-noising method in electroencephalogram late potentials detection, but it may also eliminate the beat-to-beat variance. Other types of filter applied to the time sequence may destroy the late potentials as well when trying to filter out the noise. To preserve the variance from beat to beat as well as late potentials as much as possible, the concept of a beat-sequence filter will be introduced and the wavelet transform can be directly applied to the beat sequence, as will be demonstrated in this paper. After de-noising, instead of applying the voltage comparison on the de-noised signal to determine the QRS complex endpoint, the signal will be processed by a filter bank, and the QRS complex endpoint will be determined by consideration of the correlation between two beats. Both simulation and clinical experimental results will be presented to illustrate the effectiveness of this method.