Coherence of the Surface EMG and Common Synaptic Input to Motor Neurons

Jakob Lund Dideriksen, Francesco Negro, Deborah Falla, Signe Rom Kristensen, Natalie Mrachacz‐Kersting, Dario FarinaFrontiers in Human Neuroscience 2018 · DOI

TL;DR

This study investigates how well surface EMG coherence can estimate the common synaptic input to motor neurons, a key signal for motor control. They found that rectified surface EMG can partially capture high-frequency (>5 Hz) common input during low-level contractions, but this relationship weakens with increased force and fails to detect low-frequency input.

What Problem Does This Paper Address?

Understanding how motor neurons receive input is crucial for deciphering motor control. A common method to infer this "common synaptic input" is by analyzing the coherence between surface electromyography (EMG) signals recorded from muscles. The assumption is that synchronized activity in motor neurons, driven by common input, will manifest as synchronized electrical activity detected by surface EMG. However, the exact relationship between surface EMG coherence and the underlying neural input is not fully understood. This is partly because the motor unit action potential waveforms themselves can act as high-pass filters, potentially distorting the information about lower frequency synaptic inputs. Furthermore, the non-linear process of rectifying EMG signals, often used to improve coherence estimates, can introduce its own distortions. This paper aims to systematically clarify how different EMG processing methods (raw, rectified, and high-pass filtered rectified) relate to a more direct measure of common synaptic input derived from individual motor unit activity.

Methods

The researchers recorded both surface and intramuscular EMG from the tibialis anterior muscle of 10 healthy participants during steady ankle dorsiflexion at 5% and 10% of their maximum voluntary contraction (MVC). The intramuscular EMG signals were then decomposed to identify individual motor unit spike trains.

To establish a "ground truth" for common synaptic input, they created "cumulative spike trains" (CSTs) by summing the spike trains of multiple motor units. Coherence was then calculated between pairs of these CSTs. This CST-CST coherence served as their reference measure.

They then compared this reference measure to the coherence calculated from surface EMG signals processed in three ways:

  1. Raw EMG: The unprocessed signal.
  2. Rectified EMG: The signal after rectification (absolute value taken).
  3. High-pass filtered rectified EMG: The signal first high-pass filtered at 250 Hz (a common practice) and then rectified.

Linear regression was used to assess the correlation between the surface EMG coherence and the CST-CST coherence across different frequency bands.

Datasets

Not discussed in the paper. The study used data collected from human participants during a controlled experimental procedure.

Key Insights and Ideas

  • Rectification is generally better, but not a silver bullet: While rectified surface EMG showed a stronger correlation with motor unit coherence than raw EMG, it didn't perfectly capture the underlying neural input.
  • Frequency matters: Surface EMG coherence is more effective at reflecting common synaptic input in higher frequency bands (above 5 Hz). Lower frequency inputs (<5 Hz) are largely missed by surface EMG coherence analysis.
  • Contraction intensity is a confounder: As muscle contraction force increases (from 5% to 10% MVC), the ability of surface EMG coherence to reflect common synaptic input diminishes. This is likely due to amplitude cancellation effects becoming more pronounced at higher forces.
  • High-pass filtering doesn't add much: Pre-filtering the EMG signal at 250 Hz before rectification did not significantly improve its ability to predict motor unit coherence compared to simple rectification.
  • There's still noise in the system: Even when significant correlations were found, the error in the linear fits was substantial, indicating that surface EMG coherence is only a partial estimate of motor unit coherence.

Key Results

  • At 5% MVC, rectified EMG coherence showed a strong correlation with CST-CST coherence for frequencies > 5 Hz (r² > 0.8, p < 0.01).
  • At 10% MVC, this correlation was only significant for frequencies > 15 Hz (r² > 0.8, p < 0.01).
  • Using raw EMG, the only significant correlation with motor unit coherence was observed in the 5-15 Hz band (r² > 0.68, p = 0.04).
  • In all cases, there was no association between motor unit and EMG coherence for frequencies < 5 Hz (r² ≤ 0.2, p ≥ 0.51).
  • The correlation between rectified EMG and CST-CST coherence weakened significantly as contraction intensity increased.

Limitations and Future Work

  • Limited number of participants: The study included only 10 participants, and the analysis was further limited to 7 and 8 participants for the 5% and 10% MVC conditions, respectively, due to insufficient motor unit identification. This limits the generalizability of the findings.
  • Motor unit representativeness: The identified motor units accounted for only about 17% of the surface EMG power, suggesting they might not be fully representative of the entire motor unit pool contributing to the signal.
  • Cross-talk: While efforts were made to minimize it, some level of cross-talk between EMG recording sites is inherent and could potentially bias coherence values.
  • Muscle specificity: The findings are specific to the tibialis anterior muscle and may differ in other muscles with different anatomical properties or motor unit structures.

Future work could explore these relationships in different muscles, at higher contraction intensities, and with more advanced EMG decomposition techniques to identify a larger number of motor units. Investigating the impact of different filtering strategies and their interaction with contraction intensity would also be valuable.

Summary updated

Abstract

Coherence between electromyographic (EMG) signals is often used to infer the common synaptic input to populations of motor neurons. This analysis, however, may be limited due to the filtering effect of the motor unit action potential waveforms. This study investigated the ability of surface EMG-EMG coherence to predict common synaptic input to motor neurons. Surface and intramuscular EMG were recorded from two locations of the tibialis anterior muscle during steady ankle dorsiflexions at 5 and 10% of the maximal force in 10 healthy individuals. The intramuscular EMG signals were decomposed to identify single motor unit spike trains. For each trial, the strength of the common input in different frequency bands was estimated from the coherence between two cumulative spike trains, generated from sets of single motor unit spike trains (reference measure). These coherence values were compared with those obtained from the coherence between the surface EMG signals (raw, rectified, and high-passed filtered at 250 Hz before rectification) using linear regression. Overall, the high-pass filtering of the EMG prior to rectification did not substantially change the results with respect to rectification only. For both signals, the correlation of EMG coherence with motor unit coherence was strong at 5% MVC (<i>r</i><sup>2</sup> > 0.8; <i>p</i> < 0.01), but only for frequencies > 5 Hz. At 10% MVC, the correlation between EMG and motor unit coherence was only significant for frequencies > 15 Hz (<i>r</i><sup>2</sup> > 0.8; <i>p</i> < 0.01). However, when using raw EMG for coherence analysis, the only significant relation with motor unit coherence was observed for the bandwidth 5-15 Hz (<i>r</i><sup>2</sup> > 0.68; <i>p</i> = 0.04). In all cases, there was no association between motor unit and EMG coherence for frequencies < 5 Hz (<i>r</i><sup>2</sup> ≤ 0.2; <i>p</i> ≥ 0.51). In addition, a substantial error in the best linear fit between motor unit and EMG coherence was always present. In conclusion, high-frequency (>5 Hz) common synaptic inputs to motor neurons can partly be estimated from the rectified surface EMG at low-level steady contractions. The results, however, suggest that this association is weakened with increasing contraction intensity and that input at lower frequencies during steady isometric contractions cannot be detected accurately by surface EMG coherence.
BibTeX
@inproceedings{dideriksen2018coherence,
  author = {Jakob Lund Dideriksen and Francesco Negro and Deborah Falla and Signe Rom Kristensen and Natalie Mrachacz‐Kersting and Dario Farina},
  title = {{Coherence of the Surface EMG and Common Synaptic Input to Motor Neurons}},
  booktitle = {Frontiers in Human Neuroscience},
  year = {2018},
  doi = {10.3389/fnhum.2018.00207},
}
Figure