Study On Methods Of Ventricular Late Potentials Recognization

Ventricular Late Potential (VLP), which is high-frequency, low-amplitude and irregular abnormal ECG signal, appears in the terminal portion of the QRS complex and extends to ST segment. Researches show that VLP is closely associated with ventricular tachycardia (VT) and the detection of VLP has important value in precaution of sudden death after acute myocardial infraction, diagnosing asphyxia of unknown reasons, and the evaluation of medical surgery care of VT. At present, the VLP is measured using the time-and frequency-domain methods. However, the accuracy of time-domain method is affected by fixation of QRS's end which is severely disturbed by noise, and as persistence of VLP is so short that the frequency-domain method's frequency resolution is not high. To improve traditional methods, the work has been completed in this paper. Firstly, fixation of R wave by means of singularity detection with wavelet transform and confirmation the period of time of analysis. The R wave is fixed by detection of the max. of wavelet transform module plot in characteristic scales. The even symmetrical Marr wavelet overcomes limitation of detecting the zeros-crossing point of the odd symmetry wavelet transform which is apt to be disturbed by noise. Secondly, analyzing qualitatively the characters of VLP of High-Resolution ECG by means of wavelet transform. It is found that there is observable difference between VLP negative and VLP positive in some high frequency subsignals which are in decomposed or reconstructed plot. The patient with VLP exhibits greater fluctuation or irregularity than one without VLP in these subsignals. The analysis provides a reliable experiment basis for using wavelet analysis to extract the VLP characters of HRECG. Thirdly, the eigenvalues of subsignals in every subsignal has been extracted by approximate entropy (ApEn), and the ApEn of every subsignal is regarded as the eigenvalue vector of HRECG, which are input into BP networks, and completing the recognizing of VLP. Approximate entropy has been recently proposed to quantify the irregularity in physiological signal, so it can be eigenvalue vector of HRECG. The network output is 0 or 1 which indicate VLP appearance or disappearance respectively. To improve the networks'classification effect and train speed, the additive momentum and self-adaptive–study-rate adjustment method are adopted further to improve traditional BP algorithm. The quite good result of the identification is obtained by the experiments through 25 examples of X leads HRECG. The accuracy of identifying late potentials is above 85%. The detection of VLP from HRECG with satisfied accuracy is achieved by means of the combination of wavelet transform, approximate entropy and artificial neural networks. The advantage of this method is that it doesn't require locating accurately the end point of QRS wave of HRECG and overcomes low resolution of frequency-domain analysis.