ISEK 2024 - Day 1

Disclaimer: those are rough notes from the talks that are still to be cleaned and re-organized.

Keynote speech: Andrea d’Avella

Decomposition of EMG and patterns of muscle synergies

• One of the biggest challenges is to understand how the motor system orchestrate complex tasks
• There is a clear gap between how humans learn a motor task and the state at which the field of machine learning is currently at.
• The brain is able to perform complex tasks with a very small number of parameters. This is what we call muscle synergies.
• The question: is how does the brain perform these complex tasks so efficiently?
  • To answer this question, we can go to Marr’s 3 levels of analysis:
    1. Computational theory
    2. Repersentataion models
    3. Implementation
• He puts forward the muscle synergy hypothesis. Okay, how can we understand this hypothesis and at which level of Marr’s analysis does it pay off?
  • He started with a toy arm model
  • We know that the muscular system is very redundant; we have many muscles; some of which get activated together when doing a task.
  • Looking at Marr’s levels:
    1. At the very simple level of Marr’s analysis; at a computational level: there must be an optimization going on. model
    2. at the representaiton level: we look into how a representation of goal is mapped into motor commmand? there must be a functional transformation. An optimal solution would be to compute an inverse function? but how this inverse of an internal model is implmented? how is this computation done?
• Key point: it is at this algorithmic/repesenatation level that we can formualte the muscle synergies hypothesis
  • Why this simplification: because we only need a small number of parameters to describe large repertoire of cmotor commands

• Illustration on the toy model: planar arm

    - 3rd implementaiton level: Neural networks are doing that. This synergy at the implementation lebel, must be a modular  network
        - what is the evidence for that? 
            - Early on there has been attemots to look for those synergies at the individual finger 
            - Hypothesis: EMG decomposition as a combination of muscle synergies (TResch et al Nat. Neuroscience 1999) 
                 - NMF to identigy this syneriges with the non-negativity constratins
        - Over the years there have been many forumation of mucle synergies models
  

• Achievements:
  • EMG decomposition in many sspecies and conditions
    • Synerfises in degensive reflexes (spatio-temporal synergies - D’avella Nat. Neuroscience 2003? ) . Thought: can we create and train a RNN to reveal those muscle synergies (top-down). In other words can synergies emerge in a modular network Thorugh very simple representaiton, brain is able to implement very complex and large umber of muscle combinations
    • Upper limb control in humans
    • human locomotor
      • using NMF, they could extract temporal component (ivanenko, J physio, 2004?)
  • Neural basis: can we provide evidence at the neural level
    • Experiment with stimulation. Saltiel J neurophysio, 1998, 2001. Simulation os spinal cord reveals muscle synergies
    • Optogeneic stimulation revels “motor synergy encoding” interneurons
    • Intra cortical miscrostimulation (OVerduin et al Neuron 2012) Summary: stimulation exp. and neural recording support modular organizaiton of motor control at the implementation lvel
  • Application to clinical field
    • How upper-limb synergues after stroke (cheung et al PNAS 2012)
    • Summary: muscle synery decmp. provide a novel characterization of motor imparirent after neurological lesions + changes in spatiotemporal synergies after cerebellar damage sugges ttheir corg by cortico-cerevellar loops
• Open issues:
  • Are synergies of neural origin or do they derive from task or biomechancical constraints?
  • Testing synergies as a causal model: virtual synergies. Can we predict?
  • Selection of the number of synergies
  • Rlelation to task space. How are synergies related to task space
• Issue 1:
  • Interlude: minimum spatial synergeis to describe a 3 D isometric force space is 3+1 =4 synergies?

• Issue 2: tendon- transfer syngery that makes a synerfy inefficeitve. BErger et al JNeurosicence 2013. On how subjects adapt to incompatible surgeries?

• Issue 3: how to determine the number of synergies

• Issue 4: how to characterise rile of muscle synergies in task space?

• From muscle synergies to Motor neuron syngerfies:
  • Are muscle synergies implementeed at the spinal level as motor neuron synergies. Hug et al, J Physiokogy 2023
  • One possible each muscle synergy may drive motor e=beyribs if nultiple muscle.
    • Laine Dario’s lab
    • Simone tanzerella
    • Birzelli et al J, neurophysioofy 2024 (co-activateon of biceps and triped for force generation v sjoint stiffeniing)
    • Onlne control of synergiestivc MU (Rossato et al, J, Neurosi in press)
    • SynergyAnalyzer MAtlab toolbox
    • Berg et al RSS 2023 (on the Myosuite end-to end RF, synergy based learning)
• Advertisement: Open postdoc positions on modelling :D

Oral session 2

• Impact of electrode placement on the identified muscle synergies.
  • Takeaway: same representation of muscle synergies regardless of placement even though the EMG envelope showed different muscle activiation patterns. This is during walking.
• Talk 4: Christopher how to verify that what we observe is true?
  • One approach would be to shuffle the data
  • What does it mean to change the center of pressure? looking at events…

Talk 5: - EMG-EMG coherence analysis: in the alpha and beta bands between pairs of muscles.

Session 2:

• Simon’s talk
  • Successive tracking of MUs across MVCs

Keynote speaker 2:

• Most of time when we breath: they are the voluntary breath. Motor cirtical and cortical output. There are also the reflex that are active all the time.
• Best way to measure EMG is using intramuscular in the awake wuietly breathing people also from valve muscles
• Why measure MU activity: diaphram mu discharge reglects the discharge of hrenic motoneurons.
• In quiet breathing we are at roughly 10% so we are using only the low-threshold units.
  • Instantenous discharge frequency aligned with the airflow activity: a roughtly of 20 Hz
• Part 2: Inspiratory motorneurons in quiet breathing
  • Lost of muscles that are acitve in every breath
  • Diaphram is one of the biggest muscles expanding the rip cage.
  • Early it was beleived that MU are activated buphasocally during breathing: like they are active at teh begining of the breath and another time towards the end of the breathing. (Butler 1999 J Physiology).
    • Cofired the Henneman’s size principle (5 MUs recordied with the electrode)
    • TAFPLOT (time frequency plot: mean range of firing frequency is 12.5 Hz)