Evaluation of multi-component Electrocardiogram beat detection algorithms: Implications of three different noise artifacts

Abstract

Motion artifacts, caused by changes in the electrode-skin impedance, electromyographic (EMG) interference, caused by muscle contractions, and possible baseline drifts are three of the most common sources of noise present in ECG recordings. The present study investigates the effects of these noise sources on the performance of ECG beat detection algorithms. Four different beat detection methods were used to evaluate the influence of noise sources with varying signal to noise ratios (SNRs). A database consisting of recordings from approximately 100 subjects consisting of approximately 3000 cardiac cycles was used for evaluation. Hence, 1200 records were subsequently tested by the detectors after adding three different noise sources with four different SNRs of 24 dB, 12 dB, 6 dB and -6 dB to the original 100 records. The four classifiers achieved beat detection results from 98% down to 68% for correctly detected QRS-complexes at SNRs between 24 dB and 6 dB.