Detection of late potentials in electrocardiogram signals in both time and frequency domains using artificial neural networks

Electrocardiogram (ECG) signals of patients who suffered damage in their myocardium may contain high-frequency low amplitude signals called ventricular late potentials (LPs), which occur at the end of the QRS complex. Although LPs alone can not be used as a predictor of arrhythmic events and sudden cardiac death, there is a 95% probability that there will be no more complications for patients who do not have LPs in their ECG signals. Last 40 msecs of the QRS segment is fed to an artificial neural network (ANN) along with the three time domain parameters, which are used as a standard of predicting LPs. Fourier transform of last 80 msecs of the QRST segment is fed to another ANN along with these three criteria in order to overcome the problem of locating QRS endpoint, and performances of these two networks are compared in the case of misdetection of QRS end points