1. Contact Info: simone.toma@asu.edu
[1] Waller AD. “The Sense of Effort,” Brain. 1981.
[2] Gandevia SC, McCloskey DI. “Change in motor commands as shown by changes in perceived heaviness during partial curarization and peripheral anaesthesia in man,” J Physiology, 1977.
[3] Sanes JN, Shadmehr R. “Sense of muscular effort and somesthetic afferent information in humans,” Can J Phys. Pharmacol, 1995.
[4] Cheung VCK, D’Avella A., Tresch M.C. and Bizzi E., “Central and sensory contribution to the activation and organization of muscle synergies during natural motor behaviors,” J. Neurosci., 2005.
[5] Moscatelli A., Mezzetti M. and Lacquaniti F., “Modeling psychophysical data at the population level: the generalized linear mixed model,” J. Vision, 2012.
2016-S-14738
EMG Regression as a Measure of Contribution of Central Signals to Force Perception
Neural Control of
Movement Laboratory
Simone Toma¹, Francesco Lacquaniti²
¹School of Biological and Health Systems Engineering, Arizona State University
² IRCCS Fondazione Santa Lucia, Rome, Italy
• 14 subjects; 8 elbow and shoulder muscles
• PSE : Point of Subjective Equality (p(response) = 0.5)
• β0ps, β1ps and x: force intensity, intercept and slope
defining psycho-metric curves
• MAVi: mean absolute value of EMG for muscle i
• normMAVi: MAVi / maxMAVi
Data Collection & Data Analysis
• 𝑤𝑖 : normMAVi at 30N - normMAVi at 0N
• 𝑃𝑜𝑜𝑙𝐸𝑀𝐺𝑡𝑟: weighted sum of normMAV values across
muscles per trial
• Crit: mean of PoolEmg distribution
• β0emg, β1emg and x: force intensity, intercept and slope
defining muscle-metric curves
It has been largely established that force estimation mainly relies on motor commands underlying force
production, i.e., the so-called ‘sense of effort’ 1
. Previous psychophysical studies showed a perceptual
sensitivity reduction when efferent signals were manipulated, supporting the idea that force perception
is mainly mediated by central signals 2
. However very few attempts have been made to quantify the
correlation between muscles activity and subjects’ perceived muscular effort 3
. Although these seminal
studies provided a measure of dis/concordance between muscular activity (EMG) and participants’ sense
of effort, a method to quantify the role of central signals to force perception is lacking. The present study
aimed to fill this gap by correlating muscular activity of eight arm muscles with force estimation during a
quasi-isometric force detection task
If descending motor commands are relevant signals for force perception, regression of muscular activity
(i.e., EMG) against subjects perceptual answers can be interpreted as reliable quantification of the
contribution of descending signals to force perception
Muscle metric Curve Parameters Overlap Psychometrics
3 Term Muscular Model Accounts for 60% of Perceptual Variance
• We propose a reliable method to quantify the contribution of
descending commands to force perception. Indeed by regressing
muscular activity with respect to force stimuli we were able to build
a metric (‘muscle-metric curve’) directly comparable to subjects
perceptual behavior.
• Curve parameters extracted from logistic regression of both weighted
sum of muscular activity and psychophysical answers were not
statistically different for 11 out of 14 of our subjects.
• Overall the whole muscle-metric curves explained up to 60% of
subjects perceptual answers variability. Moreover muscular
thresholds extracted by our method accounted for the wide inter-
individual differences observed among subjects perceptual
thresholds.
F
x
z
y
Cw
10 30 50 70 90
0
20
10
30
UpwardForce(N)
n. of Trials
descendant
ascendant
𝑃𝑜𝑜𝑙𝐸𝑀𝐺𝑡𝑟 = 𝑛𝑜𝑟𝑚𝑀𝐴𝑉𝑖 ∗ 𝑤𝑖
8
𝑖=1
. . . .
EMG profiles
Muscles Pooling
Participants performed a quasi-isometric perceptual task providing their judgments about force
verbally. Applied upward force intensity (F) was changed at each trial following a double staircase
adaptive procedure by either increasing or decreasing balancer weight (Cw). Eight shoulder and
forearm muscles were recorded (sEMG): brachio-radialis, biceps long, triceps long, deltoid anterior,
deltoid posterior, trapezius middle, trapezius upper, latissimus dorsi.
PerceptualThresholds
Muscular ThresholdsIntercept Slope Pse
0
1.5
-1.5
-3.0
3.0
Model reduction does not affect Muscle–Perception concordance
Synergy Extraction & GLM Fitting Leave one out GLMM nested Model Selection Synergy Contribution to Perception
Muscle Metric Curves account for Variability of Perceptual Answer
PerceptualminusMuscular
Participants PSEs ± C.I.
p(Poolemg > crit)
p(“Yes”)
Muscle-metric curve
Psycho-metric curve
R²p : muscle-metric curve fitting
answer variability (i.e., p(yes))
Psycho-metric Curve
Force Distribution
p(’yes’)
Force Level
p(’Yes’) =1./(1+exp(-(β0ps + β1ps.* x)))
GLM Logistic Fit
Point of subjective
equality
p(poolEmg > crit) =1./(1+exp(-(β0emg+ β1emg.* x)))
GLM Logistic Fit
Muscle-metric Curve
p(PoolEmg>crit)
Force Level
R²PoolEmg
Nested Models
.5
1
0
.5
1
0
NormDeltaBIC
Exp 7 6 5 4 Best 2 Least
2
1.5
1
0.5
0
PSEratioSloperatio
3
2
1
0
Perceptual / Muscular
Exp Best Least
CoefficientofVariation
0.5
0.4
0.3
0.2
0.1
p(R²p|nestedmodel)
76542
.8 .6 .4 .2 .0
R²p
1
0.5
1
0
Least2BestExp 7 6 5 4
Number of model terms
likh (nested model | R²p = .6)
likh (nested model | R²p = .7)
likh (nested model | R²p = .8)*
5th
synergy Syn. Modules Syn. Coeffs p(syn. coeffs > crit)
1st
synergy
2st
synergy
3rd
synergy
4th
synergy
BrBcTrTmTuLdDaDp Upward Forces (N)
0
1
0
1
0
1
0
1
0
1
4 fixed effect terms
3 fixed effect terms
2 fixed effect terms
DeltaAIC
Nested Models
0
60
20
40
Full
Mdl
3rdSyn
out
2ndSyn
out
1stSyn
out
Mdl w/o
4th syn
Mdl
w/o 5th
syn
• Nested Models extracted iteratively reduce multiple
regression terms backward elimination procedure
• Exp : full regression model with 8 regression terms
• Best : nested model with best trade-off between model
simplicity (delta BIC) and variance explained (R²)
• Least : nested model with only one regression term
Upward Forces
300 10 20 300 10 20
.5
0
1
Subject 1 Subject 4 Subject 7
300 10 20
R²p 0.75 R²p 0.69 R²p 0.15
Probability
Upward Forces (N)
0
.5
1
300 15 300 15
p(‘yes’)
Fit p(Coeff > crit) 2nd Synergy
Fit p(Coeff > crit) 3rd Synergy Fit p(Coeff > crit) merged Synergy
p(‘yes’)
Probability
Hypothesis
Background
Materials and Methods
Discussion
• Predictive power of muscles activity on psychophysical answers
about force remains relatively high even when muscle-metric curves
were extracted from a small number of muscles. In particular,
parameters curves ratio close to unity and the variance accounted for
were maximize with a model with only three muscles.
• Analysis on the pattern of muscular activity confirms our results by
showing that probabilities of perceptual answers can be relatively
well reconstructed only by specific muscles synergy.
• We suggest that the introduction of powerful methods to extract
muscle synergy (i.e., NNMF 4
) as well as by taking into account the
variance of cluster data (i.e., GLMM 5
) will provide new insights on to
what extent muscles coordination may underpin force perception.
Results