Automated Pacing Artifact Removal in Electrocardiograms

摘要

Electrocardiogram (ECG) is a basic and ubiquitous clinical tool to evaluate the electrical activity of the heart. Research and commercial software make automated calculations and interpretations from ECGs that have clinical value. An artificial pacemaker can however change the ECG and invalidate routine interpretation. Automated evaluation of paced ECGs is further hampered by electrical pacing artifacts that distort the physiological electrical signal. Sophisticated pacing systems like cardiac resynchronization therapy further complicate the problem by introducing pacing artifacts that are not only preceding but are also within the relevant ECG signal where they simply cannot be ignored. The pacing spike generates outliers that skews results and hinders both regression analysis and principal component analysis of the physiological signal. This is the first paper to show effective elimination of pacing spike outliers in ECGs. In order to eliminate pacing spikes, this paper proposes a novel filter and compares to prior techniques used in alternate fields. This filter uses modified Z-scores calculated from detrended data to locate outliers and replaces the spikes with a hyperbolic cosine function that connects the gap created from removing the spikes. The filtering improves the QRS area measurement by over 46% compared to median filtering and 65.2 % compared to unfiltered ECGs. The filter is fast (7.53 ms) and inexpensive.