A novel method for beat-to-beat detection of ventricular late potentials (V
LP) from high-resolution electrocardiograms (ECGs) is presented. ECG signal
s from the X lead are first filtered using a bandpass filter, and then a ti
me-sequence adaptive filter, to improve its signal-to-noise ratio. Eight fe
atures are extracted using wavelet transform, from the VLP time-frequency d
istribution of the filtered ECG signals, and used as inputs of specially de
signed artificial neural network for VLP recognition. The artificial neural
network was trained and tested using clinical data, respectively. The resu
lts show that the presented method can detect beat-to-beat-based VLP with s
ensitivity of 80% and specificity of 77%, and the detection accuracy is 78%
.