A New Algorithm for Innervation Zone Estimation Using Surface

Electromyography: A Simulation Study Based on a Simulator for

Continuous sEMGs

Malte Mechtenberg

1,2 a

, Nils Grimmelsmann

1,2 b

and Axel Schneider

1,2 c

Biomechatronics and Embedded Systems Group, University of Applied Sciences and Arts, Bielefeld, NRW, Germany

Institute of System Dynamics and Mechatronics, University of Applied Sciences and Arts, Bielefeld, NRW, Germany

Keywords:

Innervation Zone, Muscle, Innervation Point, EMG, EMG Simulation, Electromyography, Motor Unit, Firing

Pattern, EMG Array.

Abstract:

In this work, a novel algorithm for the estimation of the innervation zone location within a muscle head is

presented. The algorithm is able to identify innervation zone clusters within continuous surface electromyog-

raphy (sEMG) recordings based on linear electrode arrays. The presented algorithm is tested in a simulation

environment, which is capable of simulating EMG signals based on a common drive signal (activation). The

simulator was used to generate sEMGs of six virtual muscle based on six different conﬁgurations for the re-

spective muscle ﬁbre distributions. The virtual muscles were each activated with a trapezoidal signal (common

drive). The new algorithm was able to identify the location of the innervation zone centers with a mean ab-

solute error of 3.8% of the inter electrode distance. In the best case, the absolute error was below 1 % of the

inter electrode distance.

1 INTRODUCTION

This work introduces a new algorithm for the esti-

mation of innervation zones in an electromyography

recording. It is based on a previously published al-

gorithm (Mechtenberg and Schneider, 2023) that was

only able to ﬁnd one innervation zone location per

EMG recording, thus requiring manual labor when

multiple innervations zone locations have to be identi-

ﬁed in a recoding. The novel extension presented here

allows an automatic detection of multiple innervation

zone locations within one recording. Thus being com-

parable to other methods presented in the litrature,

that are for example (Mesin et al., 2009; Beck et al.,

2012; Marateb et al., 2016; Huang et al., 2023). The

method presented here has a small parameter set and

an implementation that is available as open source

software (Apache 2.0 license) (Mechtenberg, 2023b).

A Brief Introduction to Innervation Zones. Mus-

cle ﬁbres of skeletal muscles are controlled by so-

called motor neurons which are situated in the pe-

https://orcid.org/0000-0002-8958-0931

https://orcid.org/0000-0002-4864-4978

https://orcid.org/0000-0002-6632-3473

ripheral nervous system within the spine. A motor

neuron connects to several muscle ﬁbres. The combi-

nation of muscle ﬁbres and a motor neuron is called

a motor unit. An axon of a motor neuron connects to

the respective muscle ﬁbre at the so-called innervation

points (IPs) via the motor end plates. The innervation

points of a motor unit are distributed over the muscle

body. This distribution of IPs is called the innervation

zone (IZ) of the motor unit. The IZ locations vary

considerably between subjects (Guzm

an et al., 2011).

Moreover, during contraction, the length and position

changes of the muscle also change the relative posi-

tion of the IZs (Piitulainen et al., 2009; Martin and

MacIsaac, 2006).

In summary, the IZ location is speciﬁc for a sub-

ject and experiment condition. This means it is hard

to follow heuristics, based for example on anatomi-

cal measures to localize the IZ. An attempt was made

with an atlas of innervation zones (Barbero et al.,

2012). This atlas is considered as a general reference

for the distribution of IZs between multiple subjects.

It is, however, lacking when precise knowledge of the

inervation zone is needed, as for example in the fol-

lowing scenarios.

Mechtenberg, M., Grimmelsmann, N. and Schneider, A.

A New Algorithm for Innervation Zone Estimation Using Surface Electromyography: A Simulation Study Based on a Simulator for Continuous sEMGs.

DOI: 10.5220/0012375100003657

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 17th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2024) - Volume 1, pages 629-636

ISBN: 978-989-758-688-0; ISSN: 2184-4305

629

intra

muscular

EMG

sensor

activation

{

∆t

CD(t)

I (t)

Figure 1: Representation of a motor unit pool consisting of several motor units (green and blue) that are triggered by a

common drive (CD) signal (trapezoidal, top left). The times between individual ﬁring events (∆t) are drawn from normal

distributions with the corresponding mean frequency and standard deviations as described in eqs. (1) and (2). The simulator

determines the measurable potentials on the skin surface (EMG sensor, top right) from the temporal course of all ﬁring events

(respectively the corresponding spread of the motor unit action potentials on the muscle ﬁbres).

Possible Applications of Precision IZ Location

Estimation. The precise localization of innervation

zones is of interest for example in:

• medical treatments, when drugs have to be deliv-

ered close to the motor end plates (Zhang et al.,

2016).

• exoskeletal control the observed IZ movement

could be used as a measure for the amount of

muscle shortening (Mechtenberg and Schneider,

2023).

• case of sEMG acquisitions where a consistent sig-

nal quality is required over a wide range of move-

ments and subjects. As the sEMG signal char-

acteristics are different when the recording elec-

trodes are placed close to innervation zones com-

pared to locations above the muscle ﬁber with no

IZ or muscle ﬁber end close by.

In general, an accessible and reliable algorithm

to identify the IZ in situ could help to improve

experimental setups involving sEMG recordings.

Over the years, different methods to estimate the

innervation zone location based on sEMGs emerged.

An experienced experimenter is able to detect the

IZs by visual sEMG inspection. There were multi-

ple attempts to automate this process with signal pro-

cessing algorithms (Mesin et al., 2009; Beck et al.,

2012; Marateb et al., 2016). However, these algo-

rithms require the selection of appropriate parame-

ter sets. Tuning them appropriately can be challeng-

ing. Mechtenberg and Schneider proposed a new al-

gorithm for the innervation zone detection which uses

only two parameters. In a simulation study, an opti-

mization of these two parameters was performed suc-

cessfully under varying noise conditions (Mechten-

berg and Schneider, 2023). In that study, the algo-

rithm operated only on a single motor unit poten-

tial, i.e. the EMG signal generated by the sum of

all action potentials that travel along the muscle ﬁ-

bres of their respective motor unit. An sEMG record-

ing usually consists of several motor unit potentials

generated from different motor units. In order to

use the proposed algorithm to estimate the innerva-

tion centre in more realistic scenarios (with multi-

ple motor units and activation events), the algorithm

needs to be extended. To test this extension, Mechten-

berg and Schneider’s simulator (Mechtenberg, 2023a)

must also be modiﬁed to simulate the activation of

many motor units, resulting in a continuous sEMG.

Both, the extension of the algorithm and of the simu-

lation framework are subject of this work.

The simulator extension is described ﬁrst in sec-

tion 2.1. The experimental setup is then described

in section 2.2 before the extension of the innervation

zone estimation algorithm is described in section 2.3.

The presented algorithm achieved a mean absolute

error of 0.19mm (SD(AE) = 0.19 mm), that is 3.8%

of the inter electrode distance.

BIOSIGNALS 2024 - 17th International Conference on Bio-inspired Systems and Signal Processing

630

2 METHODS

2.1 Extension of the EMG Simulator

Using the Motor Unit Pool Model

As a basis for the EMG simulation, the approach de-

scribed in (Mechtenberg and Schneider, 2023) was

used. That simulator was only capable of simulating

single motor unit potentials. This means, that several

motor units were only triggered once in the simula-

tion. It was therefore expanded to include the capabil-

ity of generating a continuous EMG signal based on

a continuous muscle activation signal (the common

drive). The concept is depicted in ﬁg. 1. It is sim-

ilar to the approach that was described by (Petersen

and Rostalski, 2019). Here, it was assumed that mo-

tor units are organized in a pool. Such a motor unit

pool is assumed to receive a common activation sig-

nal the common drive (CD). The common drive signal

is always between zero and one, where zero corre-

sponds to no activation and one is the full activation

of a motor unit pool. Each motor unit within a pool is

assumed to start ﬁring from a distinct common drive

level CDS

that is unique for each motor unit i

The value of CDS

follows the size principle as de-

scribed later in this section (see also eq. (3)).

When the i

motor unit is active, it generates ac-

tion potentials with a mean ﬁring frequency

(CD)

that is dependent on the common drive level. The re-

lationship of the mean ﬁring frequency to the com-

mon drive is modeled utilizing eq. (1) as described by

(Petersen and Rostalski, 2019):

(CD) =











−[c

−CD]

·c

·CDS

·CD +c

−[c

−c

·CD]

·e

−

CD−CDS

CDS

≤ CD

CD ≤ 1

undeﬁned otherwise

(1)

The actual ﬁring events are generated according to al-

gorithm 1. Once the common drive CD reaches the

starting common drive level CDS

of a motor unit,

the time difference of the next ﬁring instance to the

current instance of this motor unit is drawn from a

normal distribution. The mean of that distribution is

deﬁned by

(CD) (see eq. (1)). The standard devi-

ation of the normal distribution according to (Petersen

and Rostalski, 2019) is

f ,i

(CD) =

10 + 20 ·exp



−

CD−CDS

2.5



100 ·

(CD)

. (2)

Result: motor unit activation pattern

t := time span.start;

while t ≤ time span.end do

CD := CD(t);

if CD ≥ CDS

then

∆t := N



(CD)

,σ

f ,i

(CD)



;

add activation event at t + ∆t

else

∆t := 10

−5

;

end

t+ = ∆t ;

end

Algorithm 1: Activation pattern generation of the

motor unit.

Employment of the Motor Unit Size Principle.

All motor unit parameters, which are not related to

the muscle geometry are assumed to be depended on

the motor unit size. Here, the motor units are ordered

by size, i.e. the smallest motor unit has an index of

= 0 and the largest of i

= N

. N

is the

amount of all motor units in the simulated muscle.

This assumption is commonly known as motor unit

size principle. The motor unit parameters that depend

on the motor unit size are the mean conduction veloc-

ity, number of muscle ﬁbres within a motor unit and

the common drive threshold CDS

. The relation of

the ﬁrst two motor unit parameters are described in

(Mechtenberg and Schneider, 2023). The latter is pa-

rameterized by the description of (Petersen and Ros-

talski, 2019), which is an adoption of the motor unit

pool model described by (Fuglevand et al., 1993).

CDS

= exp



log(100 ·CDMax

)

·i



(3)

Where CDMax

= 1 was chosen as the the maximal

common drive for a given motor unit.

After the generation of the motor unit ﬁring events

the motor unit potentials are simulated once per mo-

tor unit. Afterwards a motor unit potential train is

generated by shifting the motor unit potentials to the

generated ﬁring events. This shift is done in the dis-

crete time domain. Therefore, the motor unit poten-

tials can not be shifted to the exact time point of each

ﬁring event. Instead, the potentials are shifted to the

discrete point in time that is closest to the respective

ﬁring event. The time resolution is set to the time step

size T = 0.05 ms used during the motor unit potential

simulation.

In a last step, all instances of motor unit potentials

are summed up at each electrode location and down-

sampled to f = 2 kHz.

A New Algorithm for Innervation Zone Estimation Using Surface Electromyography: A Simulation Study Based on a Simulator for

Continuous sEMGs

631

x-axis positions

of the electrode array

T L

T R

(A)

IZ,x

x-axis

−R

IZ,z

z-axis

(B)

−R

IZ,y

y-axis

−R

IZ,z

z-axis

(C)

Figure 2: (A) Schematic depiction of the muscle shape deﬁning parameters in the x-axis direction. W

T L

and W

T R

are the width

of the tendon placement regions on the left and right side of the muscle. L

and L

are the distances between the center of

the innervation zone and the centers of the two tendon regions. The parameter W

deﬁnes the width of the innervation zone

of the muscle. The location of the motor end plate (innervation point IP) as well as the left and right myotendinous junctions

in the x-axis are drawn from a uniform distribution in these regions. The relative position in the x-axis of the electrodes is

also shown as black dots in the top part. (B, C) depict 50 muscle ﬁbres generated from the setup in (A), where the z- and y-

coordinates of the innervation points are drawn from a uniform circular shaped distribution with the radius R around the center

of the innervation zone. The innervation points are marked with a black dot. The muscle ﬁbres are drawn with a transparency.

2.2 Setup of the EMG Simulation

The simulator in this study can in principle gener-

ate virtual muscle ﬁbres of motor units with differ-

ent lengths (L

), sizes of the innervation zones

) and sizes of tendon regions (W

T L

T R

, loca-

tion of the junction from muscle ﬁbre to tendon) as

shown in ﬁg. 2 (A). These ﬁve parameters are called

shape deﬁning parameters. The muscle ﬁbres can be

distributed across a spatial volume with the help of

their position parameters (R, P

IZ,y

IZ,z

, see ﬁg. 2 (B,

C)). For details of the shape deﬁning and position pa-

rameters of the muscle ﬁbres see (Mechtenberg and

Schneider, 2023).

In this study, two scenarios were simulated. Sce-

nario I is displayed in ﬁg. 3 (A) and scenario II in

ﬁg. 3 (B). The shape deﬁning parameters and the mo-

tor unit pool recruitment related parameters are set to

the same values for both scenarios. All constant pa-

rameters are listed in table 1 in the appendix.

In Scenario I, the muscle shape related parameters

T L

T R

are kept constant. But the loca-

tion of the whole muscle was shifted along the x-axis

relative to the electrode array (see ﬁg. 3 (A)).

IZ,x

∈ {5.75 cm, 4 cm, 1.75 cm} (4)

This way, different innervation zone locations can be

tested with a varying inﬂuence of the end effect, that

occurs at the myotendinous junction.

In Scenario II, the muscle shape related parame-

ters L

and the innervation zone center along the

x-axis were varied. The remaining parameters were

kept constant. L

and L

were chosen such that the

myotendinous junctions remain at the same position

left and right of the electrode array, as displayed in

ﬁg. 3 (B), while the innervation zone locations were

set to eq. (4).

The virtual electrode array was always placed at

the same position for all scenarios. The electrode

array consists of 16 monopolar electrodes with a

5mm spacing. All electrodes had the same y- and z-

coordinates (table 1). The ﬁrst electrode in the array

is placed at x = 0 cm.

The result of the simulation is a monopolar record-

ing from a continuous simulated EMG for each mus-

cle based on a trapezoid common drive signal. The

common drive signal that was used is shown in ﬁg. 4.

For the later use of the EMG signal in the al-

gorithm for innervation zone center estimation, the

monopolar recording was converted into a double dif-

BIOSIGNALS 2024 - 17th International Conference on Bio-inspired Systems and Signal Processing

632

scenario 1 scenario 2

conf. 4

conf. 5

conf. 6

conf. 1

conf. 2

conf. 3

Figure 3: Depiction of the innervation zone and myotendinous junction centers of each simulated virtual muscle relative to

the x-axis position of the electrode array. Two types of innervation zone position shifts were simulated. In (A) the innervation

zone location relative to the virtual muscle stays the same, but the virtual muscle as a whole is shifted along the x-axis relative

to the electrode array. In (B) the myotendinous junction centers remain at the same position on the x-axis, but the innervation

zone centers are shifted relative to the electrode array. In total, six muscle conﬁgurations were simulated, all with the same

muscle belly radius R and with the same y-, and z-axis coordinates for the innervation zone center.

ferential recording as described in (Mechtenberg and

Schneider, 2023), resulting in 14 double differential

(DD) EMG signals.

2.3 Innervation Zone Center Estimation

As a basis for the innervation zone center estima-

tion, the algorithm introduced in (Mechtenberg and

Schneider, 2023) was used. That algorithm operates

on a time window of the EMG signal and is able to

ﬁnd one estimation of the innervation zone center per

time window. The algorithm basically consists of two

steps.

Step 1. Per double differential electrode potential

(one of the above-mentioned 14 DD-signals), one mo-

tor unit potential (MUP) is identiﬁed using a wavelet

correlation. This step is parameterized with the

wavelet width λ. As wavelet, the third Hermite-

Rodriguez series expansion was selected, as this was

proposed by (Farina et al., 2000) for the identiﬁca-

tion of double differential MUPs. After identiﬁca-

tion of the motor unit potentials in each of the DD-

signals, the algorithm uses pairs of any combination

of two of those potentials to set up a linear extrapola-

tion (line) that represents the location of the MUPs at

earlier points in time. Since the MUPs travel in both

directions away from the innervation zone, the inter-

section of those extrapolated lines of MUPs that travel

in one direction and those of MUPs that travel in the

other direction represent the innervation zone.

Step 2. The intersection points are clustered using

the density based DBSCAN algorithm (Pedregosa

et al., 2011). The cluster that contains the largest

number of intersections is used to calculate the

innervation zone center location estimate, i.e. the

center of that cluster. This step is parameterized with

0 2 4

time in s

250

500

750

motor unit ID

0.0

0.5

1.0

common drive

Figure 4: The activation patterns for every 50th motor unit

within a virtual muscle are shown, with the corresponding

common drive signal that was used to generate the activa-

tion patterns (trapezoidal shape). The color of the ﬁring

events encodes the motor unit size, which also becomes big-

ger with increasing motor unit ID i

the density parameter ε of the DBSCAN algorithm.

The result of these two steps is a position rela-

tive to the x-axis of the electrode array and a point in

time. In the previous study (Mechtenberg and Schnei-

der, 2023), a parameter variation was performed in or-

der to ﬁnd a suitable parameter set, which is tolerant

to noise but still accurate. Based on that work, the pa-

rameters λ = −0.00391667 and ε = 1.10083333 were

selected. For details on the internal paramters please

refer to the publication (Mechtenberg and Schneider,

2023) or the open source implementation of the algo-

rithm (Mechtenberg, 2023b).

Extension to Estimate Innervation Zone Centers

in Continuous EMG Recordings. In this study, the

algorithm described above was extended to operate on

a continuous EMG signal. This extension contains

two aspects.

First, the continuous EMG had to be segmented

into multiple time windows. For that, a 40 ms Hann

window was used. The Hann window reduces the

A New Algorithm for Innervation Zone Estimation Using Surface Electromyography: A Simulation Study Based on a Simulator for

Continuous sEMGs

633

0 1 2 3 4

time in s

DD 0

DD 1

DD 2

DD 3

DD 4

DD 5

DD 6

DD 7

DD 8

DD 9

DD10

DD11

DD12

DD13

(A)

-25 0 25 50 75 100

x coordinate in mm

100

Count

DD 0 DD 13

AE= 0.02mm

(B)

Figure 5: In (A) all 14 simulated double differential EMG signals for one of the six virtual muscles is displayed. Blue

dots mark the innervation zone location estimates of an innervation zone center cluster. The black dots are innervation zone

estimates, which are not part of a cluster. In (B) the distribution of innervation zone estimates is shown. The blue asterisk

marks the position of the innervation zone cluster center. In the lower part of the plot, the actual innervation zone region and

the myotendinous junction regions are displayed in the same style as in ﬁg. 3. The absolute error from the cluster center (blue

asterisk) and the actual innervation zone center is shown at the top. The absolute error in this case is below 1 % of the inter

electrode distance.

chance that motor unit potentials, which are only

partly within the window, are tracked during Step 1 of

the innervation zone center estimation. The window

was shifted over the EMG signals in 20 ms steps, re-

sulting in 20 ms overlaps. The window width as well

as the window overlap was chosen after inspection of

the motor unit potentials.

For each window, an innervation zone center was

estimated as described above. Second, under the as-

sumption of an isometric contraction, these estimated

positions are clustered in the position domain (axis of

the electrode array) using DBSCAN (Pedregosa et al.,

2011). This leads to an estimate of possible multiple

innervation zone locations per simulation experiment.

The innervation zone center estimation was parame-

terized with eps = 0.1 and min samples = 10.

Per simulation setup presented in ﬁg. 3 the innerva-

tion zone clusters were identiﬁed and compared to the

ground truth, i.e. the center of all innervation point lo-

cations.

3 RESULTS

As described, two scenarios with three parameter sets

each were simulated, resulting in six simulated EMGs

(virtual muscles). The input for each simulation was

a common drive (CD) signal in a trapezoidal shape as

displayed in ﬁg. 4 and as depicted as inset in ﬁg. 1.

For all six simulated EMG signals, the extended ver-

sion of the innervation zone estimation algorithm was

applied as described in section 2.3. For one arbitrar-

ily selected of the six virtual muscles a detailed view

on the estimated innervation zones is shown in ﬁg. 5.

In ﬁg. 5 (A) the whole simulated EMG signal is dis-

played with the estimated innervation zone centers for

all time windows. The innervation zone estimates that

are part of the accepted cluster are displayed as blue

dots. The rejected innervation zone estimates are dis-

played in black. In ﬁg. 5 (B) the distribution of in-

nervation zone estimates (compared with (A)) over

the double differential electrode positions is shown,

as well as the center of the accepted cluster (blue as-

terisk). There is a noticeable accumulation of inner-

vation zone centers in the middle of the array. The ac-

cepted innervation zone center is near the maximum

in the histogram and close to the actual innervation

zone center of the simulated muscle (AE= 0.02 mm).

For all simulated virtual muscles, the innervation

zone distribution over the electrodes and the accepted

cluster center are shown in ﬁg. 6. In general, all the

accepted clusters of innervation zone estimates are

close to the actual innervation zone locations. For all

clusters, the difference between their mean estimate

and the actual innervation zone center is below 1 mm.

In case of those virtual muscles with their innervation

zone center close to the middle of the array (2. col-

umn in ﬁg. 6) the error becomes minimal (0.05mm

and 0.08 mm) for both scenarios. The worst value for

the estimation accuracy is 0.56mm.

In case of the simulations where the innervation

zone is close to the end of the array, there seem to be

clusters of false estimates in the middle of the elec-

trode array.

In general, the mean absolute error is MAE =

0.19mm (SD(AE) = 0.19mm).

BIOSIGNALS 2024 - 17th International Conference on Bio-inspired Systems and Signal Processing

634

100

Count

AE= 0.02mm

conf. 1

AE= 0.05mm

conf. 2

AE= 0.28mm

conf. 3

-25 0 25 50 75 100

x coordinate in mm

100

Count

AE= 0.16mm

conf. 4

-25 0 25 50 75 100

x coordinate in mm

AE= 0.08mm

conf. 5

-25 0 25 50 75 100

x coordinate in mm

AE= 0.56mm

conf. 6

Figure 6: Depiction of the innervation zone location estimate distributions for all six simulated virtual muscles are shown in

the same style as in ﬁg. 5 (B). The blue asterisk marks the center of the accepted innervation zone cluster. All absolute errors

between the cluster centers and the actual innervation zone centers are below 1 mm.

4 DISCUSSION AND

CONCLUSION

This study shows that the newly proposed algorithm

for the innervation zone center estimation can be used

for continuous EMG signals by segmenting the EMG

in overlapping windows. Even with this straight for-

ward approach of segmenting the EMG data, a con-

siderably good result was achieved, since the absolute

prediction error was below 0.57 mm for all tested con-

ﬁgurations (MAE = 0.19 mm, SD(AE) = 0.19 mm).

The mean absolute error is 3.8% of the inter electrode

distance 5 mm. This is a result that is close to an-

other recently published innervation zone estimation

algorithm (Huang et al., 2023). Huang et al. report

for their principal component based algorithm a mean

difference of 3 % (up to 8 % depending on the exper-

iment conditions) of the inter electrode distance, that

is 8mm in their setup. For en extensive comparison,

the two algorithmes would have to be evaluated with

the same dataset.

A further result of the investigation was that the

prediction error of the algorithm varies with the elec-

trode array location relative to the innervation zone.

When the electrode array was placed above the inner-

vation zone, a prediction error of AE ≤ 0.05 mm was

achieved. Next steps will include the investigation of

the dependency between the prediction error and dif-

ferent electrode conﬁgurations as well as the deriva-

tion of a model which is also able to follow the move-

ment of the innervation zone under different contrac-

tion scenarios of a muscle (e.g. isometric, isotonic

and auxotonic conditions).

ACKNOWLEDGEMENTS

This work has been supported by the research train-

ing group “Dataninja” (Trustworthy AI for Seamless

Problem Solving: Next Generation Intelligence Joins

Robust Data Analysis) and by the “TransCareTech”

project (Transformation in Care & Technology), both

funded by the German federal state of North Rhine-

Westfalia. It was also supported by a Career@Bi

grant of the University of Applied Sciences and Arts,

Bielefeld, Germany.

REFERENCES

Barbero, M., Merletti, R., and Rainoldi, A. (2012). Atlas

of muscle innervation zones: understanding surface

electromyography and its applications. Springer Sci-

ence & Business Media.

A New Algorithm for Innervation Zone Estimation Using Surface Electromyography: A Simulation Study Based on a Simulator for

Continuous sEMGs

635

Beck, T. W., DeFreitas, J. M., and Stock, M. S. (2012). Ac-

curacy of three different techniques for automatically

estimating innervation zone location. Computer meth-

ods and programs in biomedicine, 105(1):13–21.

Farina, D., Fortunato, E., and Merletti, R. (2000). Non-

invasive estimation of motor unit conduction velocity

distribution using linear electrode arrays. IEEE Trans-

actions on Biomedical Engineering, 47(3):380–388.

Fuglevand, A. J., Winter, D. A., and Patla, A. E. (1993).

Models of recruitment and rate coding organization

in motor-unit pools. Journal of neurophysiology,

70(6):2470–2488.

Guzm

an, R. A., Silvestre, R. A., Arriagada, D. A.,

GUZM

AN, R., SILVESTRE, R., and ARRIAGADA,

D. (2011). Biceps brachii muscle innervation zone lo-

cation in healthy subjects using high-density surface

electromyography. int J Morphol, 29(2):347–52.

Huang, C., Chen, M., Zhang, Y., Li, S., Klein, C. S., and

Zhou, P. (2023). A novel muscle innervation zone esti-

mation method using monopolar high density surface

electromyography. IEEE Transactions on Neural Sys-

tems and Rehabilitation Engineering, 31:22–30.

Marateb, H. R., Farahi, M., Rojas, M., Ma

nanas, M. A.,

and Farina, D. (2016). Detection of multiple innerva-

tion zones from multi-channel surface emg recordings

with low signal-to-noise ratio using graph-cut seg-

mentation. PLoS One, 11(12):e0167954.

Martin, S. and MacIsaac, D. (2006). Innervation zone

shift with changes in joint angle in the brachial bi-

ceps. Journal of Electromyography and Kinesiology,

16(2):144–148.

Mechtenberg, M. (2023a). UAS-Embedded-Systems-

Biomechatronics/EMG-concentrated-current-

sources: v0.2.2.

Mechtenberg, M. (2023b). UAS-Embedded-Systems-

Biomechatronics/sEMG-innervation-zone-

estimation.

Mechtenberg, M. and Schneider, A. (2023). A method for

the estimation of a motor unit innervation zone center

position evaluated with a computational semg model.

Frontiers in Neurorobotics, 17.

Mesin, L., Gazzoni, M., and Merletti, R. (2009). Automatic

localisation of innervation zones: a simulation study

of the external anal sphincter. Journal of Electromyo-

graphy and Kinesiology, 19(6):e413–e421.

Pedregosa, F., Varoquaux, G., Gramfort, A., Michel, V.,

Thirion, B., Grisel, O., Blondel, M., Prettenhofer,

P., Weiss, R., Dubourg, V., Vanderplas, J., Passos,

A., Cournapeau, D., Brucher, M., Perrot, M., and

Duchesnay, E. (2011). Scikit-learn: Machine learning

in Python. Journal of Machine Learning Research,

12:2825–2830.

Petersen, E. and Rostalski, P. (2019). A comprehen-

sive mathematical model of motor unit pool organiza-

tion, surface electromyography, and force generation.

Frontiers in physiology, 10:176.

Piitulainen, H., Rantalainen, T., Linnamo, V., Komi, P., and

Avela, J. (2009). Innervation zone shift at different

levels of isometric contraction in the biceps brachii

muscle. Journal of electromyography and kinesiology,

19(4):667–675.

Zhang, C., Peng, Y., Li, S., Zhou, P., Munoz, A., Tang,

D., and Zhang, Y. (2016). Spatial characterization of

innervation zones under electrically elicited m-wave.

In 2016 38th Annual International Conference of the

IEEE Engineering in Medicine and Biology Society

(EMBC), pages 121–124. IEEE.

APPENDIX

Table 1: List of constant parameters for the EMG simula-

tion experiments.

Parameter Value Reference

electrode y 0cm estimated

electrode z 2cm estimated

1cm estimated

T L

0.5cm estimated

T R

0.5cm estimated

10cm

estimated

IZ,y

0cm estimated

IZ,z

0cm estimated

774 (Mechtenberg and

Schneider, 2023)

20 (Petersen and Rostalski,

2019)

1.5 (Petersen and Rostalski,

2019)

30 (Petersen and Rostalski,

2019)

13 (Petersen and Rostalski,

2019)

8 (Petersen and Rostalski,

2019)

8 (Petersen and Rostalski,

2019)

0.05 (Petersen and Rostalski,

2019)

BIOSIGNALS 2024 - 17th International Conference on Bio-inspired Systems and Signal Processing

636