Wavelet transform and neural-network-based adaptive filtering for QRS detection

This paper presents an adaptive neural-network-based ECG signal filtering, and a wavelet transform based QRS detection method. An adaptive whitening filter is modeling the lower frequencies of the ECG, which are inherently nonlinear and nonstationary. In this way the estimation error will consist of the QRS wave. The wavelet-transform-based QRS detection method will determine the position of these complexes and will separate the normal and abnormal beats. The structure of the algorithm allows us to modify in real time the basic QRS template, in this way it can be customized to an individual subject. From the correctly estimated QRS waveforms, a parametrical model will determine the optimal filtering parameters for the wavelet-based detector and calculate the "optimal" input pattern for the whitening filter. For an adequate comparison with other processing algorithms, tests have been made for the commonly used MIT-BIH database. The correct QRS detection rate was above 99,9%.