Detection of Ventricular Late Potentials in High-Resolution ECG Signals by a Method Based on the Continuous Wavelet Transform and Artificial Neural Networks

-Ventricular Late Potentials (VLPs) are low-amplitude, high-frequency signals which appear at the end part of the QRS complex of a High-Resolution ECG (HRECG) record including three orthogonal (XYZ) leads. VLPs are clinically useful in identifying post-MI (Myocardial Infarction) patients prone to Ventricular Tachycardia (VT) and Sudden Cardiac Death (SCD). The Continuous Wavelet Transform (CWT), Principal Component Analysis (PCA), and Artificial Neural Networks (ANNs) are used to detect VLPs in this work. The terminal part of the QRS complex in the Vector Magnitude (VM) waveform is processed with the CWT to extract a feature vector. In this way, the resulted time-scale representation is subdivided into several regions, and the sum of the squared decomposition coefficients is computed in each region. Then, the resulted feature vector is processed by PCA to reduce its dimensionality. Finally, a supervised feedforward ANN, trained by an appropriate set of these feature vectors, is applied in the analysis of HRECG signals in order to identify VLPs. A set of different HRECG records, which includes real ECG records without VLPs and semi-simulated ECG signals with VLPs, was used to evaluate this method. The results reveal good improvements in sensitivity and specificity comparing to the conventional time-domain method, developed by Simson. Key-Words:-Ventricular Late Potentials, High-Resolution ECG, Wavelet Transform, Neural Networks