Robust Multiresolution Wavelet Analysis and Window Search Based Approach for Electrocardiogram Features Delineation

摘要

Delineation of Electrocardiogram (ECG) features is a deep-seated algorithmic module for computer-aided diagnosis of cardiac disorders. The amplitude and intervals of different peaks of characteristic waves of ECG depict the functioning of the heart in normal and abnormal conditions. The changes in morphological features are very subtle and difficult to correlate with the abnormalities and demand a lot of clinical acumen. Hence, an accurate delineation of ECG's characteristic waves is principally required for computerized electrocardiography. Firstly, the original ECG signal is de-noised based on discrete wavelet transform (DWT) technique by discarding the coefficients using soft and hard thresholding corresponding to the artifact frequency component for baseline wandering, power line interference (PLI), muscle tremors and high frequency noise. Then window search for local minima/maxima pairs along with adaptive thresholding is performed for the detection of R-peaks followed by P, Q, S and T peaks. Additionally, the onset and offset points of these peaks have also been identified. The various interval features corresponding to the characteristic waves have been deliberated and validated using European ST-T database (EDB) records. The proposed method is robust to artifacts and delineation process. The result shows 99.94% sensitivity (S-E), 99.98% positive predictivity (+P) and percentage root mean square difference (PRD) of 0.132%, 0.122%, and 0.242% for baseline wandering, PLI and muscle and high frequency and muscle tremors elimination respectively, which also addresses the issues of conventional methods reported in literature.