The Difficulty Of EMG Burst Identification

Abstract

Surface electromyography (EMG) is commonly used to add physiologic context to observed patterns of movement. The onset of the EMG is often used to determine when a person responds to stimuli or to identify maladaptive muscle coordination strategies. Our research group has previously shown that Bayesian Changepoint Analysis (BCP) is superior to the standard linear envelope methodologies when determining a single EMG onset in a data series. PURPOSE: Examine the effectiveness of visual inspection and the novel BCP algorithm to detect EMG bursts in multiple muscles during complex movements. METHODS: Muscle activity from 10 healthy subjects was collected from the gastrocnemius, biceps femoris, and vastus lateralis muscles using surface EMG electrodes (4kHz sampling rate). Subjects completed 4 minutes of exercise on three modalities: treadmill running, ergometer cycling and stair climbing. All exercises were performed at self-selected low-to-moderate intensities. Six to ten seconds of EMG was collected at 90 and 210 seconds into the exercise. Three researchers visually identified the number of bursts in a trial twice (randomized, double-blind methodology). The instances where all six identifications (three reviewers, twice) agreed on the number of bursts, were compared with the identification results from the BCP algorithm. RESULTS: While the within rater reliability (ICC: 0.85) and between rater reliability (ICC: 0.83) were good, the visual review only resulted in 111 trials (out of 180) where all raters agreed on the number of bursts (61.7% total agreement). The correlation between the number of bursts raters identified and the BCP algorithm was moderate (Pearson’s R: 0.52). Furthermore, across all trials, there was a difference of 711 EMG bursts between the two methods. CONCLUSIONS: While visual assessment of the EMG is the “gold standard” for burst detection, its reproducibility is generally poor in dynamic tasks. Despite initial success with the BCP algorithm in determining EMG onset, the current iteration of the algorithm is insufficient for EMG burst identification in complex waveforms. This work highlights the need for a standardized algorithm of EMG burst detection, but also indicates that further work is necessary to make algorithms of sufficient accuracy and reliability.