Rectification of surface EMG impairs cortico-muscular coherence estimation

Cortico-muscular coherence (CMC) is an established method for the analysis of interactions between cortical and muscular activity. CMC can be studied noninvasively with millisecond precision using multi-channel EEG/MEG recordings. Previous CMC studies have been primarily based on bipolar and Laplacian transformations of EEG -approaches which do not necessarily guarantee an optimal detection of coherence.

Here, we propose a novel method, which provides an optimal solution for finding spatial filters. The core idea is to use multiple regression where narrowly filtered EEG signals serve as predictors and the electromyogram (EMG) serves as the dependent variable. This approach, which we call Regression-CMC (R-CMC), allows an optimal detection of linear relationships between cortical and muscular activities. The regression coefficients serve as spatial filters and can be used to calculate the topography of cortical sources contributing to CMC.

Using this R-CMC approach, we investigated the effects of EMG rectification on CMC, which remains a highly debated issue. A majority of previous studies advocated the use of EMG rectification. However, recent EMG modeling studies (Stegeman et al., 2010) showed that rectification is an unnecessary procedure which in fact can even obscure the true relationship between cortical and muscle activities. Here, we investigated CMC during isometric contraction of abductor pollicis brevis muscle using multi-channel EEG recordings. CMC was estimated based on either single-channel activity (C3/C4 vs. linked earlobes), Laplacian transform or our novel R-CMC approach. For all three types of analysis, we detected CMC in the β frequency range above the contralateral sensorimotor areas. No changes in the strength of CMC on the basis of rectified and unrectified EMG were detected when EEG was estimated from the single-channel – in agreement with findings of Yao et al. (2007). Importantly, EMG rectification resulted in smaller CMC values in case of both, the Laplacian and R-CMC approach. The reason for CMC not being different for rectified/unrectified EMG in case of single-channel estimation is probably due to the excessive superposition of β oscillations originating from neighboring cortical areas. On the other hand, Laplacian transformation and R-CMC approaches are spatial high-pass filters and thus are capable of extracting primarily local β oscillations. This in turn allows a more refined estimation of CMC with a possibility to detect eventual effects of rectification.

The results of our study confirm predictions made from previous modeling EMG studies, namely the fact that CMC during the isometric contraction is based rather on a linear relationship between cortical and muscle activity. Moreover, we also show that EMG rectification is an unnecessary processing step and even might have a detrimental effect on the estimation of CMC.

References:

[1] Stegeman D.F., Van De Ven W.J.M., Van Elswijk G.A., et al (2010). The α-motoneuron pool as transmitter of rhythmicities in cortical motor drive. Clin Neurophysiol, 121(10), 1633–1642.

[2] Yao B., Salenius S., Yue G.H., et al (2007). Effects of surface EMG rectification on power and coherence analyses: an EEG and MEG study. J Neurosci Methods, 159(2), 215–23.

Keywords: Cortico-muscular Coherence, Rectification, Multiple Linear Regression