Improving ventricular late potentials detection effectiveness

Ventricular late potentials (VLPs) are low-amplitude, wideband-frequency waveforms that appear in the high-resolution electrocardiogram of patients with some cardiac diseases. VLP detection and analysis represents a prominent non-invasive marker for cardiac and cardiac related diseases.; In a VLP diagnostic system, it is necessary to implement some noise-diminishing strategies to improve the SNR. Later, some methods to obtain relevant information on VLPs have to be applied. Time domain and frequency domain analyses have been the methods traditionally used. More recently, great interest has been focused on time-frequency analysis techniques.; The global objective of this investigation was to improve VLP detection, while obtaining computationally affordable algorithms that could be implemented in computer-based HRECG (high-resolution electrocardiogram) analysers. This research proposes novel pre-processing and processing schemes to improve VLP detection in the noisy ECG environment.; A HRECG database was created and rigorous simulations of the HRECG records, VLP waveforms, and noise and interference were incorporated to test every algorithm.; A novel powerline interference canceller, based on an isoelectric interval detector, proved to diminish the interference better than traditional cancellers, without appreciable distortion of the ECG signal. A QRS detector, based on the double-level algorithm and cross-correlation adjustment, used a new composite channel to obtain better precision. To complete the pre-processing, an FIR high-pass filter combining an all-pass and a binomial low-pass filter outperformed other previous FIR designs for VLP detection.; The time-domain analysis improved robustness and repeatability by using a new modified signal-averaging (MSA) scheme, which is a combination of mean and median filtering. A novel adaptive enhancer with MSA and maximum absolute value “averaging” was designed to obtain certain beat-to-beat information and to emphasise the boundaries of the QRS complex. In addition, a novel 2-D VLP detection scheme was implemented, including stationary wavelet transform (SWT) denoising, Canny edge detection, and moment-based feature extraction.; The number of heartbeats needed for the processing algorithms was reduced 5 times approximately. The new algorithms can handle certain non-stationary environments and provide beat-to-beat information. The results show improvement in the sensitivity and specificity. All this will have, consequently, a direct impact on future research in this area.