An algorithm for EMG noise detection in large ECG data

Large collections of electrocardiogram recordings (ECG) are valuable for researchers. However, some sections of the recorded ECG may be corrupted by electromyogram (EMG) noise from muscle. Therefore, EMG noise needs to be detected and filtered before performing data processing. In this study, an automated algorithm for detecting EMG noise in large ECG data is presented. The algorithm extracts EMG artifact from the ECG by using a morphological filter. EMG is identified by setting a threshold for the moving variance of extracted EMG. The algorithm achieved 100% detection rate on the training data. The algorithm was tested on 150 test signals from three sets of test signals (50 signals in each set). Set 1 was created by adding EMG noise to EMG-free ECG signals, set 2 was manually selected ECG sections which contain EMG noise, and set 3 contained randomly selected ECG signals. Sensitivity was 100%, 94%, and 100% on sets 1, 2, and 3, respectively. All sets had 100% specificity. The algorithm has computational complexity of O(N).