Global Decision Making for Wavelet Based ECG Segmentation

In this work, we propose an improvement of an established single lead electrocardiogram (ECG) beat segmentation algorithm based on the wavelet transform. First, for a particular recording a reference beat is determined by averaging over a certain amount of beats. Subsequently, this beat is used to obtain recording specific thresholds and search windows needed for the segmentation of the whole recording. Since noise and artifacts significantly influence the segmentation process, we show that using the information provided by the reference beat positively impacts the results. Specifically, using this global information of the reference beat, the algorithm becomes more robust against transient noise and signal abnormalities. Consequently, the proposed approach leads to an ECG beat segmentation algorithm specifically suited for detecting subtle relative changes of characteristic time intervals and amplitude levels.