[1/27]

I would like to talk (a bit) about high-density surface EMG-based motor unit identification

#ISEK2020 starting soon!

2 years ago I felt excited & overwhelmed at #ISEK2018 in Dublin with all the cool stuff people were doing with HD-EMG. And I was about to start!

[2/27]
Full disclosure: I am not an expert. I have just been learning a few things during my PhD which I would like to share and maybe generate some discussion. This thread might be more interesting/useful for those of you who are not very familiarised with this technique (yet)

[3/27]
Shall we start with a throwback?

Full room for an interesting workshop by Dario Farina and Ales Holobar at #ISEK2018

"Extraction of information from high-density EMG: recent developments and perspectives."

Excited to hear more about recent progresses at #ISEK2020

[4/27]
#MotorUnits are the key functional units of human muscle contractions.

Motor Unit

Spinal motor neurone
+
Muscle fibres it innervates

[5/27]
Electromyographic (EMG) signals are the electrical activity of the motor units of a muscle.

Can we take a close look at these signals and identify the firing of individual motor units?

[6/27]
Yes! EMG signals are the summation of the action potentials of active motor units (MUs), and different MUs have different waveforms which can be identified.

Via invasive intramuscular EMG

[7/27]
Via high-density surface EMG: matrices/arrays with many small electrodes

Non-invasive
High(er) number of MUs identified

@AlecsDelVecchio et al
Tutorial: Analysis of motor unit discharge characteristics from high-density surface EMG signals
https://www.sciencedirect.com/science/article/pii/S1050641120300419

[8/27]
Which matrix (or matrices) to use?
No simple answer (?) Depends on the size and shape of your muscle(s) of interest. Also consider number of electrodes, inter-electrode distance, max number of channels of your system

Pilot testing w/ different ones? Probably a good idea

[9/27]
Location of the matrix
In my case, for gastrocnemius medialis, I use palpation to find the orientation and shape of the muscle and place the matrix in the location suggested by this paper
(which was actually the best one in my pilot testing)

https://www.researchgate.net/publication/221762414_Preferred_Sensor_Sites_for_Surface_EMG_Signal_Decomposition

[10/27]
Location of the matrix

Some researchers have identified the motor point and aligned it between columns of the grid.

This might improve the decomposition but honestly, I am not sure.

[11/27]
Matrix orientation
It shouldn't matter much in theory (?)

Maybe better results if it is oriented in relation to the propagation of action potentials? If we do this, we can also estimate conduction velocity
Figure below: @AlecsDelVecchio et al

https://www.sciencedirect.com/science/article/pii/S1050641120300419

[12/27]
Conductive paste? Cream? Or gel?

I did some pilot testing and...
Paste (ac cream) and cream (EC2) gave me a similar number of MUs
It was harder to remove the cream (EC2) from the matrices
Gel was not so good to identify MUs

[13/27]
During automatic decomposition, the shapes of the motor units are identified from the background noise.

To increase the ratio of the "energy" in the EMG signals to the "energy" in the noise signal, the noise signal should be minimised.

A few things I have noticed

[14/27]
Tracking MUs between trials is possible due to the high spatial resolution of motor unit action potential shapes in HD-EMG signals.

Note: During a fatiguing task, the MUAP shapes will likely change and it will be harder to track MUs.

Some data from my PhD

[15/27]
And MUs have been successfully tracked across days... Even weeks!

Methodological papers:
http://tiny.cc/8pyasz 
http://tiny.cc/jqyasz 

[16/27]
#ManualEditingNeeded
Manual editing of spike trains is required. Automatic decomposition procedures are not perfect: they will include firings of similar MUs & not include firings of the MU of interest.

[17/27]
#ManualEditingNeeded
Pulse to noise ratio of the firings might be lower at MU recruitment as the MUAP shape slightly changes during recruitment.

Higher likelihood of automatic decomposition procedures missing firings at recruitment?

[18/27]
#ManualEditingNeeded
Nicely demonstrated by @AlecsDelVecchio et al. (2020)

https://www.sciencedirect.com/science/article/pii/S1050641120300419

[19/27]
#LessCanBeMore
Excluding 5% of the channels may improve the decomposition procedures. This will exclude a couple of channels with a lower pulse to noise ratio. Below is a layout of the DEMUSE tool and the exclusion of 2 channels is highlighted

[20/27]
Factors influencing number of identified MUs

Contraction intensity
It becomes harder to discriminate different MUs at high intensities.

Below some of my pilot data at different %MVC from 2 (good) participants (gastroc medialis). Each dot is a different trial

[21/27]
(...)
Contraction intensity
"For example, in the tibialis anterior muscle, which is a reliable muscle for decomposition, we observed a 30% reduction in the number of motor units that can be identified when the target force increases from 35% to 70% of maximum force."

[22/27]
Factors influencing number of identified MUs

Subcutaneous layer
Thicker subcutaneous layer greater distance between the electrodes and the muscle

It can act as a biological “filter”: shapes from different MUs will look + similar
Example of possible MU merging

[23/27]
Factors influencing number of identified MUs

Effect of sex?
When investigating this effect we need to be careful & take into consideration confounding factors (e.g. subcutaneous fat)

In fact, my participant with the highest number of MUs is a lean female

[24/27]
Factors influencing number of identified MUs

Orientation of muscle fibres
Possibly less MUs identified in muscles with parallel fibres, due to superimposition of muscle fibre action potentials (?)

[25/27]
MUs during dynamic contractions? Challenging. Change of
Distance between electrodes & muscle
Latency at which APs arrive at muscle-tendon junction

Glaser & Holobar used overlapping short epochs to consider the change in MUAP shapes
https://ieeexplore.ieee.org/document/8579211

[26/27]
#RealTimeNeuralInterfacing
In fact, a real-time decomposition method for the identification of MU discharges from HD-EMG signals in dynamic contractions has recently been proposed

https://ieeexplore.ieee.org/document/9075399?denied=

[27/27]
Final thought

We can't identify a high number of reliable MUs from everyone

Shall we check during data collection the signals are decomposable? If not, adjust the matrix and/or exclude the participant? Of course the latter needs transparency and could bias our sample.

Post-conference thought 1/3

Another interesting #ISEK2020 came to an end. Nice to see so many developments in EMG decomposition. I must say that we should not forget that it is exciting to have more people using this technique, but there is also a higher risk for misuse...

Post-conference thought 2/3

E.g. making strong claims when the data is not accurate

Some of the things that should NOT happen IMO
Not carefully editing the spike trains after decomp
Including inaccurate MUs
Including spike trains with merged MUs
Including repeated MUs

Post-conference thought 3/3

We now have the tools to avoid these And we must use them.