https://doi.org/10.31449/inf.v45i2.3192 Informatica 45 (2021) 213–222 213
Method for Estimating Tensiomyography Parameters from Motion Capture
Data
Dino Vlahek, Tadej Stoši´ c and Tamara Golob
Institute of Computer Science
Faculty of Electrical Engineering and Computer Science at the University of Maribor, Maribor, Slovenia
E-mail: dino.vlahek1@um.si, tadej.stosic1@um.si, t.golob@um.si
Miloš Kalc, Teja Liˇ cen and Matjaž V ogrin
Institute of Sports Medicine, Faculty of Medicine at the University of Maribor, Maribor, Slovenia
E-mail: milos.kalc@ism-mb.si, teja.licen@um.si, matjaz.vogrin@um.si
Domen Mongus
Institute of Computer Science
Faculty of Electrical Engineering and Computer Science at the University of Maribor, Maribor, Slovenia
E-mail: domen.mongus@um.si
Keywords: tensiomyography, marker-based motion capture, 3D points, geometric transformation
Received: June 05, 2020
Tensiomyography is a muscle performance assessment technique that measures its mechanical responses.
In this study, we explored the possibility of replacing traditional tensiomyography measurement systems
with motion capture. The proposed method allows the measurement of multiple muscle points simultane-
ously while achieving measurements during a patient’s movements. The results showed that approximately
5 mm error was achieved when estimating maximal muscle displacement, while time delay in muscle con-
traction and contraction time was assessed with up to 20 ms error. As confirmed by physicians, the intro-
duced errors are within the acceptable margin and, thus, the obtained results are medically valid.
Povzetek: V ˇ clanku predstavimo novo metodo, ki omogoˇ ca veˇ ctoˇ ckovno merjenje tensiomiografije.
Metoda temelji na snemanju mišiˇ cne kontrakcije s sistemom za zajem gibanja. Rezultati metode in pri-
padajoˇ ce napake so ovrednoteni s strani zdravnikov. Le-ti ocenjujejo, ali so napake še znotraj sprejemljive
meje, da so rezultati medicinsko veljavni.
1 Introduction
Tensiomyography (TMG) is a non-invasive mechanomyo-
graphy method that measures mechanical muscle response
based on radial muscle belly displacement induced by the
electrical stimulus. The measurement unit usually includes
an electrical stimulator, a data acquisition subunit, a dig-
ital sensor, and muscle electrodes [28]. TMG output is a
displacement-time curve evaluated with the following pa-
rameters: Delay time (T
d
) is a time difference between the
electrical impulse and 10% of the contraction, contraction
time (T
c
) is a time difference between 10% and 90% of
the contraction, sustain time (T
s
) is a time difference be-
tween 50% of the contraction and 50% of the relaxation,
and relaxation time (T
r
) is a time difference between 90%
and 50% of the relaxation and maximal displacement of the
muscle contraction (D
m
).
TMG’s resulting parameters are usually used for the
evaluation of an individual’s maximal speed, explosive-
ness, endurance, and flexibility [16]. They are also applied
in the training optimization process in order to prevent neg-
ative effects of muscle asymmetry and asynchrony on an in-
dividual’s performance [19]. Additionally, after an injury,
muscle functional capacity can be assessed using TMG, so
that the most effective rehabilitation treatment is adminis-
tered [21], while its usage in medical research includes es-
timation of muscle composition [24], evaluation of muscle
atrophy [10], measuring adaptation to different pathologies
[12], and for determination of muscle fiber type composi-
tion [6].
However, TMG has significant drawbacks, as it is a
fixed, static tool that can perform single-point measure-
ments [28, 10]. Additionally, the reliability of measure-
ment highly depends on the experiences of the measurer,
since placements of sensors and electrodes could affect the
reliability of the results [24], while measurements are gen-
erally performed in a static and relaxed position [28].
In order to address the above-mentioned drawbacks, we
propose a method that generates output similar to TMG
using marker-based motion capture. The proposed ap-
proach allows for measuring multiple points simultane-
ously, thus reducing the effort required in order to measure
muscle contractions. In addition, the measurements can be
achieved not only in the relaxed positions but also while
moving, as control markers are used in order to stabilize

214 Informatica 45 (2021) 213–222 D. Vlahek et al.
natural limb movement in markers. Accordingly, related
work in motion capture is described in Section 2. Section 3
introduces a new method that estimates TMG output from
motion capture. The proposed method validation results
are presented and discussed in Section 4, while section 5
concludes the paper.
2 Related work
Motion capture allows for recording the movement of ob-
jects or people. Various motion capturing systems were in-
troduced recently, including acoustical, mechanical, mag-
netic, and optical ones. The most widely used systems are
optical. They use a camera for recording the motions of
markers attached to an object [18]. Two types of markers
are used for this purpose, namely, passive and active ones.
Passive markers reflect light generated by a near camera
lens, while the active ones use their own light source. In
any case, 3D positions of markers over time can be recon-
structed using optical triangulation, and the estimated tra-
jectories can be used for pinpointing positions of displace-
ments for analysis, visualization, and simulation purposes
[11]. Both motion capture systems have been used in the
entertainment industry for years as well, where its success-
ful implementation ranges from famous films like Avatar
and Lord of the rings [1] to the gaming industry [20]. Opti-
cal motion tracking usage, with the support of virtual real-
ity, was also demonstrated for tracking and reconstruction
of hand movements for sign language interpretation and
dance coaching [26]. Furthermore, optical motion capture
technology is today an emerging technology in sports and
medicine. For instance, its usage was examined for the pur-
poses of facial performance acquisition [13], animation of
the natural bending, bulging, and jiggling [4], reconstruc-
tion of three-dimensional rotations of human joints [7], and
gait analysis [3]. Within this context, the efficiency com-
parison of marker-less and marker-based motion capture
for gait analysis was conducted, where the authors con-
cluded that maker-based motion capture is more suitable
for clinical use. A more recent study, however, has shown
that motion capture, in general, can introduce errors due to
linear scaling and technology imperfection [14]. Here, the
musculoskeletal models of different centers of joints, ob-
tained from marker-based motion capture, were scaled and
compared with measurements obtained from MRI images
that are today believed to be the gold standard.
Nevertheless, optical motion capture is still widely used
in sport gesture analysis that ranges from repetitive stresses
and movements on the shoulder [23] to underwater body
motions [2]. Moreover, efficient utilization of motion cap-
ture technology for medical uses was proposed in [22, 25].
In addition, motion capture technology was successfully
used for the rehabilitation of patients with spastic hemi-
plegic cerebral palsy [15] and Duchenne muscular dys-
trophy [9]. Thus, as marker-based motion capture is
frequently used for gait and skeleton analysis in sports
medicine and animating 3D objects in the entertainment in-
dustry, it provides a solid technological foundation for our
study.
3 Method
In this section, a method for estimating TMG parameters
from 3D motion capture data is presented. The proposed
method uses a set of markers in order to trace muscle con-
traction using motion capture, while TMG parameters are
estimated during the following steps:
– Point stabilization is achieved first in order to com-
pensate for natural limb movements and preserve only
those movements that result from muscle contractions.
– Construction of displacement-time curves is
achieved next by estimating displacement distances
from stabilized 3D marker positions.
– Extraction of TMG parameters is finally achieved
based on the estimated displacement-time curve.
Following the description of the mathematical framework,
these steps are described in detail.
3.1 Theoretical background
The implementation of the proposed mathematical frame-
work is given in the homogeneous coordinate system. This
allows for implementing all the used geometric transforma-
tions, including translation, by matrix multiplication and,
thus, enables efficient utilization of a graphic processing
unit [8].
Let a set of markers M = f
t
m
i
g, where
t
m
i
=
[
t
x
i
;
t
y
i
;
t
z
i
; 1], whilei is a markers index andt is the time
t of its capture. A vector between points
t
m
i
and
t
m
j
is de-
noted as
t
~ v
i;j
=
t
m
i
 t
m
j
, while its projections toXY andXZ planes are denoted as
t
u
i;j
= (
t
x
i;j
;
t
y
i;j
; 1) and
t
w
i
= (
t
x
i;j
;
t
z
i;j
; 1), respectively. A translation for an ar-
bitrary vector
t
~ v
i;j
= (
t
x
i;j
;
t
y
i;j
;
t
z
i;j
) is then given by a
translation matrixM
T
, defined as
M
T
(~ v
T
) =
2
6
6
4
1 0 0
t
x
i;j
0 1 0
t
y
i;j
0 0 1
t
z
i;j
0 0 0 1
3
7
7
5
: (1)
In addition, rotation matricesM
Ry
(
y
) andM
Rz
(
z
) that
define rotation aroundY andZ axis for given angles  y
and  z
, respectively, are denoted by
M
Ry
(
y
) =
2
6
6
4
cos y
0 sin y
0
0 1 0 0
 sin y
0 cos y
0
0 0 0 1
3
7
7
5
;
M
Rz
(
z
) =
2
6
6
4
cos z
 sin z
0 0
sin z
cos z
0 0
0 0 1 0
0 0 0 1
3
7
7
5
(2)

Method for Estimating Tensiomyography Parameters from. . . Informatica 45 (2021) 213–222 215
[27].
3.2 Point stabilization
In order to account for the natural movement of a limb,
two control markers were placed on the limb joints in such
a way that they were not affected by muscle contractions.
They are denoted by the indices i = 1 and i = 2, while
the corresponding vector
t
~ v
1;2
was used to stabilize the set
of markers M (see Figure 1). In order to achieve stabi-
lization, the origin of coordinate system was shifted to the
control markeri = 1, while theX axis was aligned with
t
~ v
1;2
. Note that the latter only requires rotation aroundY andZ axis, while the rotation aroundX axis can be ne-
glected due to the nature of measurement that limits such
limb movements. Thus, rotations aroundY andZ axis
were denoted by rotation angles  y
and  z
, defined as
angles between projected vectors
t
~ u
1;2
and
t
~ w
1;2
and the
X axis, respectively [27]. This stabilization, denoted as
M
S
, is formally defined as
M
S
=M
T
(
t
m
1
) M
Ry
(
y
) M
Rz
(
z
) =
2
6
6
4
cos
t
 y
cos
t
 z
 sin
t
 z
cos
t
 y
sin
t
 y
t
x
1
sin
t
 z
0 0
t
y
1
 sin
t
 y
cos
t
 z
sin
t
 y
sin
t
 z
cos
t
 y
t
z
1
0 0 0 1
3
7
7
5
;
(3)
where M
T
(
t
m
1
) denotes translation of the origin of co-
ordinate system to control markeri = 1, whileM
Ry
(
y
)
andM
Rz
(
z
) rotations by  y
and  z
, respectively. More-
over, stabilized set of markers S = f
t
s
i
g, where
t
s
i
=
(
t
x
0
i
;
t
y
0
i
;
t
z
0
i
) is, thus, given as:
t
s
i
=M
S
 t
m
i
: (4)
3.3 Construction of displacement-time
curves and TMG parameters extraction
This step aims to construct a displacement time curves
fromS and extract the required TMG parameters. As mus-
cle contraction is captured by the movement of stabilized
markers, a displacement curve for each marker
t
s
i
2 S
is generated by measuring its distance
t
d
i
in time t > 0
from its starting point, given at t = 0. Formally, a dis-
placement curve is given by a discrete mapping function
D : (t;i)!R defined by:
D(t;i) =
p
(
0
s
i
 t
s
i
)
2
: (5)
As Eq. 5 cannot produce negative values, it is critical
that the initial measurement given at time t = 0 is mea-
sured in the relaxing (non-contracted) state of the mus-
cle. D(t;i), thus, provides a set of control points based
on which a polynomial interpolation is achieved in order
to increase the precision of the estimated TMG parame-
ters. As polynomial interpolation is a well-know problem,
it is not further discussed here. Its efficient implementa-
tion is described in [5]. Moreover, as explained in Section
1, five parameters can be extracted from a displacement
curve, where most of the medically relevant information is
contained in maximal contractionD
m
, delay timeT
d
, and
contraction time T
c
. Given an interpolated displacement
curved
i
(t), definitions are as follows:
D
m
(i) = max
t
d
i
(t);
T
d
(i) = arg min
t
(t;d
i
(t) 0:1 D
m
(i));
T
c
(i) =T
d
(i) arg min
t
(t;d
i
(t) 0:9 D
m
(i)):
(6)
4 Results and discussion
The proposed method’s implementation was done using
C++, and experiments were conducted on a workstation
with Intel
®
Core
TM
i5 CPU and 16 GB of main mem-
ory. Experimental data about twelve different participants
were collected using a 4 5 matrix of reflective markers
that were placed on the quadriceps femoris of participants’
left leg, while two control markers were placed over the
trochanter head and lateral condyle (see Fig. 1). The par-
ticipants were instructed to lie supine on a therapeutic ta-
ble where each placed its left leg on a triangular cushion
that provided approximately 30
 knee angle support. Then,
Rectus Femoris (RF - the upper central part of the thigh)
and Vastus Medialis (VM - lower internal part of the thigh)
muscles were stimulated with a single electrical impulse
provided by a high voltage constant current electrical stim-
ulator, while control measurements were obtained using a
traditional TMG sensor (TMG-BMC Ltd, Ljubljana, Slove-
nia). One series of these measurements consisted of five
consecutive muscle stimulation with a 5 s interstimulus in-
tervals in order to prevent post-activation potentiating. For
each muscle, six different sets of stimulations were admin-
istrated, starting with the stimulation intensity of 30 mA,
increasing the power in each measurement by 10 mA, un-
til a maximum of 80 mA was reached. Thus, a total of 30
stimuli for each muscle were measured. At the same time,
the same muscle contractions were captured from reflective
markers with a Smart-D, BTS s.p.a. motion capture sys-
tem. The system consisted of eight infrared cameras with
800 600 spatial and 60 Hz temporal resolution, while their
position at the therapeutic table is shown in Figure 2.
At each marker, the measured motion capture data was
used in order to reconstruct the displacement-time curves,
while their agreement with the control TMG curve was es-
timated in terms of Pearson correlation coefficient [17].
The obtained results are shown in Table 1. Obviously,
the displacement-time curves showed a different agreement
level with the control TMG measurements, depending on
the markers’ proximities to the TMG sensor. On average,
VM measurements displayed lower correlations with con-
trol ones than those performed on RF due to the dilated
oscillations of the muscular surface, while those markers

216 Informatica 45 (2021) 213–222 D. Vlahek et al.
Figure 1: The placement of twenty markers and two control markers on the subject’s leg. The Violet area represents the
placement of the TMG sensor during measurement, while red circles indicate control markers.
Figure 2: Position of cameras where measurements were performed.
placed near the TMG sensor displayed better correlation
in both cases. As follows from Table 1, in the case of
VM, the highest correlations with control TMG were mea-
sured atm5 andm9, whilem15 displayed best results in
case of RF stimulation. In addition, the results obtained
atm11,m16,m19, andm20 were also statistically signifi-
cant in both cases. Such results are expected as these mark-
ers were located in anatomical regions of measured mus-
cles. Displacement-time curves from markers that show the
highest agreement with corresponding control TMG curves
are further presented in Figure 3, while TMG parameters
were extracted from these particular markers and further
examined.
In order to assess the accuracy of the extracted TMG pa-
rameter, their values were estimated from displacement-
time curves generated from markers. Moreover, param-
eters error rates represent differences between their val-
ues and the parameters’ values from corresponding control
TMG. The results are shown in the appendix (Tables 2 7). When consideringT
c
andT
d
for VM, the lowest error
rates were observed in case of m5 at 50 mA and m20 at
50 mA with 0:2% and 0%, respectively, while error rates
between 1:1 25:3% in case ofT
c
and 3 30:4% in case
T
d
were observed in other cases. On the other hand, no er-
ror was observed forD
m
atm20 at 60 mA, while the error
rates in other cases ranged between 1:7 61:7%. In the
case of RF,T
c
error rates were in the range of 0:9 33:7%,
with the smallest related tom20 at 30 mA. However,T
d
in-
troduced inconsistent error rates, from 1:8% in case ofm16
at 30 mA, up to 75:7% in case ofm11 at 60 mA.D
m
error
rates were between 6:4 33:7%, where the lowest one is
associated withm19 at 80 mA.
According to the evaluation provided by the medical ex-
perts, the obtained error rates were within the acceptable
ranges and can be considered as medically irrelevant. The
error of D
m
can be explained by the fact that the TMG
sensor is slightly pressed into the soft tissue, resulting in a
small depression at a baseline level, causing a higher value
ofD
m
when a traditional TMG is measured. As expected,
there were high errors in theT
d
parameter since the signals
from motion capture and TMG were not properly synchro-
nized. Additionally, obtained errors could be explained by

Method for Estimating Tensiomyography Parameters from. . . Informatica 45 (2021) 213–222 217
Table 1: Pearson correlation coefficients between the displacement-time curves of markers and control TMG measure-
ments together with average result according to each marker.
30mA 40mA 50mA 60mA 70mA 80mA Average
VM RF VM RF VM RF VM RF VM RF VM RF VM RF
m3 0.652 0.774 0.527 0.803 0.497 0.81 0.514 0.716 0.607 0.701 0.657 0.628 0.575 0.738
m4 0.696 0.745 0.543 0.721 0.504 0.672 0.529 0.557 0.604 0.583 0.698 0.512 0.595 0.632
m5 0.74 0.837 0.606 0.842 0.581 0.839 0.569 0.75 0.637 0.747 0.729 0.672 0.644 0.781
m6 0.604 0.77 0.508 0.76 0.584 0.785 0.538 0.699 0.647 0.732 0.681 0.638 0.594 0.731
m7 0.575 0.735 0.393 0.779 0.352 0.767 0.386 0.677 0.478 0.756 0.515 0.664 0.45 0.73
m8 0.745 0.766 0.578 0.725 0.559 0.686 0.529 0.625 0.63 0.691 0.584 0.644 0.604 0.689
m9 0.73 0.818 0.578 0.825 0.629 0.809 0.621 0.699 0.628 0.723 0.681 0.666 0.644 0.757
m10 0.734 0.732 0.596 0.789 0.578 0.761 0.555 0.767 0.646 0.801 0.718 0.697 0.638 0.757
m11 0.746 0.761 0.573 0.839 0.607 0.845 0.659 0.743 0.652 0.829 0.666 0.644 0.65 0.777
m12 0.531 0.748 0.374 0.796 0.35 0.825 0.365 0.724 0.425 0.79 0.453 0.65 0.417 0.755
m13 0.612 0.738 0.482 0.759 0.431 0.717 0.423 0.656 0.567 0.613 0.55 0.593 0.511 0.679
m14 0.647 0.732 0.525 0.727 0.516 0.668 0.494 0.626 0.605 0.635 0.607 0.654 0.566 0.674
m15 0.729 0.878 0.562 0.846 0.537 0.858 0.6 0.758 0.665 0.81 0.638 0.718 0.622 0.811
m16 0.731 0.789 0.574 0.797 0.589 0.826 0.632 0.724 0.631 0.758 0.669 0.679 0.637 0.762
m17 0.528 0.816 0.45 0.813 0.422 0.801 0.412 0.758 0.408 0.776 0.43 0.693 0.441 0.776
m18 0.567 0.558 0.427 0.603 0.381 0.639 0.406 0.558 0.536 0.535 0.523 0.496 0.473 0.565
m19 0.712 0.795 0.661 0.761 0.62 0.817 0.53 0.714 0.668 0.785 0.647 0.703 0.639 0.763
m20 0.61 0.789 0.632 0.837 0.613 0.819 0.561 0.788 0.698 0.71 0.665 0.691 0.63 0.772
m21 0.641 0.824 0.527 0.796 0.53 0.795 0.484 0.727 0.609 0.761 0.617 0.589 0.563 0.749
m22 0.483 0.76 0.439 0.801 0.542 0.774 0.504 0.688 0.617 0.801 0.576 0.67 0.527 0.749
0 0:2 0:4 0:6 0:8 1
0
2
4
6
8
10
12
s
mm
m5 at40 mA
m9 at30 mA
m11 at30 mA
TMG at30 mA
TMG at40 mA
(a)
0 0:2 0:4 0:6 0:8 1
0
2
4
6
8
s
mm
m15 at30 mA
m16 at50 mA
m19 at30 mA
m20 at40 mA
TMG at30 mA
TMG at40 mA
TMG at50 mA
(b)
Figure 3: Displacement-time curves from traditional TMG (dashed lines) and corresponding markers (solid lines) that
produced the highest level of agreement with traditional TMG for muscle a) VM and b) RF. On thex axis, there is time
in s, while they axis represents displacement in mm.
the fact that the TMG measurement unit provides more pre-
cise measurements because of its 1000 Hz temporal reso-
lution when comparing it with 60 Hz of the motion capture
system. On the other hand, markers m19 and m11 reg-
istered significant movements, even though they were not
placed in the anatomical regions, where contraction of RF
and VM was expected. Such an outcome might have dif-
ferent explanations:
– strong electrical stimulation can cause the propagation
of the electrical stimuli in deeper tissues, causing mus-
cle contraction of adjacent muscles,
– the passive mass, represented by inactivated muscles
and adipose tissue near the stimulated region, can vi-
brate, causing errors in measurements.

218 Informatica 45 (2021) 213–222 D. Vlahek et al.
5 Conclusion
A new method for estimating TMG parameters from 3D
motion capture, proposed in this paper, allows for mea-
surement of TMG parameters at multiple points simulta-
neously, while measurements can be obtained during the
patient’s movement. With the error rates of 5 mm when
estimating maximal muscle displacement and up to 20 ms
when estimating delay time and contraction time, the pro-
vided results proved to be medically relevant. Nevertheless,
selection and proper placement of markers are required.
One of the future tasks is a synchronization of the TMG
and motion capture signals that would allow us to obtain
the exact starting time of muscle contraction and, thus, fur-
ther improved contraction and delay time assessment. In
addition, improved point stabilization with compensating
for rotations along theX-axis will be considered. Finally,
as the described study provides validation of the proposed
method from the engineering point of view, the extended
medical one is required to prove its real value.
Acknowledgement
This work was supported by the Slovenian Research
Agency under Grants J2-8176 and P2-0041, and the
project "Wearable Integrated Smart Brace for Reha-
bilitation Monitoring and Diagnostic of Disorders in
Muscular Functions - WIBRANT" that is co-financed
by the Republic of Slovenia, Ministry of Education,
Science and Sport, and the European Union under the
European Regional Development Fund. More info:
www.eu-skladi.si/?set_language=en.
References
[1] Rufino R. Ansara. and Chris Joslin. Adding cartoon-
like motion to realistic animations. In Proceedings of
the 12th International Joint Conference on Computer
Vision, Imaging and Computer Graphics Theory
and Applications - Volume 1: GRAPP , (VISIGRAPP
2017), pages 137–147. INSTICC, SciTePress, 2017.
https://doi.org/10.5220/0006174001370147.
[2] Gustavo R. D. Bernardina, Pietro Cerveri, Ricardo
M. L. Barros, João C. B. Marins, and Amanda P. Sil-
vatti. Action sport cameras as an instrument to per-
form a 3d underwater motion analysis. PLOS ONE,
11:1–14, 08 2016. https://doi.org/10.1371/journal.
pone.0160490.
[3] Elena Ceseracciu, Zimi Sawacha, and Claudio Co-
belli. Comparison of markerless and marker-based
motion capture technologies through simultaneous
data collection during gait: Proof of concept. PLOS
ONE, 9(3):1–7, 03 2014. https://doi.org/10.1371/
journal.pone.0087640.
[4] Caecilia Charbonnier, Frank C. Kolo, Victoria B.
Duthon, Nadia Magnenat-Thalmann, Christoph D.
Becker, Pierre Hoffmeyer, and Jacques Menetrey.
Assessment of congruence and impingement of the
hip joint in professional ballet dancers: A motion
capture study. The American Journal of Sports
Medicine, 39(3):557–566, 2011. https://doi.org/10.
1177/0363546510386002.
[5] M. F. I. Chowdhury, C. Jeannerod, V . Neiger,
E. Schost, and G. Villard. Faster algorithms for mul-
tivariate interpolation with multiplicities and simul-
taneous polynomial approximations. IEEE Transac-
tions on Information Theory, 61(5):2370–2387, May
2015. https://doi.org/10.1109/TIT.2015.2416068.
[6] Raja Dahmane, Srdjan Djordjevic, Bostjan Simu-
nic, and V ojko Valencic. Spatial fiber type distri-
bution in normal human muscle Histochemical and
tensiomyographical evaluation. Journal of biome-
chanics, 38(12):2451–2459, 2005. https://doi.org/10.
1016/j.jbiomech.2004.10.020.
[7] R. Degeorges, J. Parasie, D. Mitton, N. Imbert, J.-
N. Goubier, and F. Lavaste. Three-dimensional rota-
tions of human three-joint fingers: an optoelectronic
measurement. preliminary results. Surgical and Ra-
diologic Anatomy, 27(1):43–50, Mar 2005. https:
//doi.org/10.1007/s00276-004-0277-4.
[8] Peng Du, Rick Weber, Piotr Luszczek, Stanimire To-
mov, Gregory Peterson, and Jack Dongarra. From
CUDA to OpenCL: Towards a performance-portable
solution for multi-platform GPU programming. Par-
allel Computing, 38(8):391 – 407, 2012. https://doi.
org/10.1016/j.parco.2011.10.002.
[9] Raluca Ganea, Pierre-Yves Jeannet, A. Paraschiv-
Ionescu, Nathalie Goemans, Christine Piot, Mar-
leen Hauwe, and Kamiar Aminian. Gait assessment
in children with duchenne muscular dystrophy dur-
ing long-distance walking. Journal of child neu-
rology, 27:30–8, 07 2011. https://doi.org/10.1177/
0883073811413581.
[10] Oscar Garcia-Garcia, Alba Cuba-Dorado, Tania
Alvarez-Yates, Javier Carballo-Lopez, and Mario
Iglesias-Caamano. Clinical utility of tensiomyogra-
phy for muscle function analysis in athletes. Open
Access Journal of Sports Medicine, 10:49–69, 2019.
https://doi.org/10.2147/OAJSM.S161485.
[11] Michael Gleicher. Animation from observation: Mo-
tion capture and motion editing. SIGGRAPH Com-
put. Graph., 33(4):51–54, November 1999. https:
//doi.org/10.1145/345370.345409.
[12] K. Grabljevec, H. Burger, K. Kersevan, V . Valencic,
and C. Marincek. Strength and endurance of knee
extensors in subjects after paralytic poliomyelitis.

Method for Estimating Tensiomyography Parameters from. . . Informatica 45 (2021) 213–222 219
Disability and Rehabilitation, 27(14):791–799, July
2005. https://doi.org/10.1080/09638280400020623.
[13] Haoda Huang, Jinxiang Chai, Xin Tong, and Hsiang-
Tao Wu. Leveraging motion capture and 3D scanning
for high-fidelity facial performance acquisition. ACM
Trans. Graph., 30(4):74:1–74:10, July 2011. https:
//doi.org/10.1145/2010324.1964969.
[14] Hans Kainz, Hoa Hoang, Chris Stockton, Roslyn
Boyd, David Lloyd, and Christopher Carty. Accuracy
and reliability of marker based approaches to scale the
pelvis, thigh and shank segments in musculoskeletal
models. Journal of Applied Biomechanics, 33:1–21,
03 2017. https://doi.org/10.1123/jab.2016-0282.
[15] M.C.M. Klotz, L. Kost, F. Braatz, V . Ewerbeck,
D. Heitzmann, S. Gantz, T. Dreher, and S.I. Wolf.
Motion capture of the upper extremity during activi-
ties of daily living in patients with spastic hemiplegic
cerebral palsy. Gait & Posture, 38(1):148–152, 2013.
https://doi.org/10.1016/j.gaitpost.2012.11.005.
[16] Irineu Loturco, Saulo Gil, Cristiano Laurino, Hamil-
ton Roschel, Ronaldo Kobal, Cesar Abad, and Fabio
Nakamura. Differences in muscle mechanical prop-
erties between elite power and endurance athletes:
A comparative study. The Journal of Strength and
Conditioning Research, 29(6):1723–1728, 12 2014.
https://doi.org/10.1519/JSC.0000000000000803.
[17] J. Neyman and E. S. Pearson. On the problem of the
most efficient tests of statistical hypotheses. Philo-
sophical Transactions of the Royal Society of Lon-
don. Series A, Containing Papers of a Mathematical
or Physical Character, 231:289–337, 1933. https:
//doi.org/10.1098/rsta.1933.0009.
[18] Haldun M. Ozaktas and Levent Onural. Three-
Dimensional Television: Capture, Transmission, Dis-
play. Springer Science & Business Media, 2007.
https://doi.org/10.1007/978-3-540-72532-9.
[19] Ezequiel Rey, Carlos Lago-Penas, and Joaquin Lago-
Ballesteros. Tensiomyography of selected lower-limb
muscles in professional soccer players. Journal of
Electromyography and Kinesiology, 22(6):866 – 872,
2012. https://doi.org/10.1016/j.jelekin.2012.06.003.
[20] A. L. Rincon, H. Yamasaki, and S. Shimoda. Design
of a video game for rehabilitation using motion cap-
ture, emg analysis and virtual reality. In 2016 Inter-
national Conference on Electronics, Communications
and Computers (CONIELECOMP), pages 198–204,
Feb 2016. https://doi.org/10.1109/CONIELECOMP.
2016.7438575.
[21] Pedro S Dias, Joan S Fort, Daniel A Marinho, Al-
bano Santos, and MÃ ˛ ario Marques. Tensiomyogra-
phy in physical rehabilitation of high level athletes.
The Open Sports Sciences Journal, 3:47–48, 03 2014.
https://doi.org/10.2174/1875399X010030100047.
[22] L. A. Schwarz, A. Mkhitaryan, D. Mateus, and
N. Navab. Estimating human 3D pose from time-of-
flight images based on geodesic distances and opti-
cal flow. In Face and Gesture 2011, pages 700–706,
2011. https://doi.org/10.1109/FG.2011.5771333.
[23] Elena Seminati, Alessandra Marzari, Oreste Va-
condio, and Alberto Minetti. Shoulder 3D range
of motion and humerus rotation in two volleyball
spike techniques: injury prevention and performance.
Sports Biomechanics, 14(2):216–231, 07 2015. https:
//doi.org/10.1080/14763141.2015.1052747.
[24] Bostjan Simunic. Between-day reliability of a
method for non-invasive estimation of muscle com-
position. Journal of electromyography and kinesiol-
ogy, 22:527–30, 04 2012. https://doi.org/10.1016/j.
jelekin.2012.04.003.
[25] Ivo Stancic, Tamara Grujic Supuk, and Ante Pan-
jkota. Design, development and evaluation of op-
tical motion-tracking system based on active white
light markers. IET Science, Measurement & Tech-
nology, 7:206–214(8), July 2013. https://doi.org/10.
1049/iet-smt.2012.0157.
[26] Jeff Tang, J.C.P. Chan, and H. Leung. Interactive
dancing game with real-time recognition of continu-
ous dance moves from 3D human motion capture. 5th
International Conference on Ubiquitous Information
Management and Communication, pages 50–58, 01
2011. https://doi.org/10.1145/1968613.1968674.
[27] John Vince. Mathematics for Computer Graph-
ics. Springer-Verlag, Berlin, Heidelberg, 5th edition,
2014. https://doi.org/10.1007/978-1-4471-7336-6.
[28] T. Zagar and D. Krizaj. Validation of an accelerome-
ter for determination of muscle belly radial displace-
ment. Medical and Biological Engineering and Com-
puting, 43(1):78–84, February 2005. https://doi.org/
10.1007/BF02345126.

220 Informatica 45 (2021) 213–222 D. Vlahek et al.
6 Appendix
Table 2: The table shows average values and associated standard deviation for parameters extracted from markers at a
stimulation intensity of 30 mA. Associated errors were placed below the results. The lowest errors are highlighted.
Vastus medialis Rectus femoris
Tc(ms) T
d
(ms) Dm(mm) Tc(ms) T
d
(ms) Dm(mm)
m5 39 ( 15.95) 27.2 ( 11.43) 6.5 ( 3.58) 47.2 ( 7.14) 36.7 ( 8.21) 3.4 ( 1.16)
m9 41.8 ( 17.95) 19.4 ( 3.23) 9 ( 5.08) 55.1 ( 7.56) 28.3 ( 9.84) 3 ( 1.43)
m11 43.5 ( 19.18) 21.6 ( 5.48) 5.3 ( 3.31) 59.2 ( 15.13) 27.6 ( 11.02) 2.9 ( 1.8)
m15 44.8 ( 20.08) 21.9 ( 4.64) 7.4 ( 4.74) 49.4 ( 6.01) 33.8 ( 6.75) 4.9 ( 1.8)
m16 45.3 ( 17.85) 22 ( 4.47) 5.2 ( 3.28) 50.5 ( 10.46) 26.1 ( 10.68) 3.5 ( 1.11)
m19 45.7 ( 18.22) 25.4 ( 9.35) 5 ( 3.62) 50.4 ( 6.9) 34.8 ( 11.04) 2.8 ( 1.35)
m20 41.9 ( 17.6) 29.1 ( 10.43) 4.8 ( 3.42) 51 ( 4.5) 29.8 ( 7.58) 3.3 ( 1.33)
TMG 36.5 ( 14.11) 24.2 ( 2.93) 4.9 ( 1.31) 51.5 ( 20.1) 25.6 ( 3.37) 4.3 ( 1.4)
Error(Tc) Error(T
d
) Error(Dm) Error(Tc) Error(T
d
) Error(Dm)
m5 2.5 (6.9 %) 3.0 ( 12.4 %) 1.6 ( 33.7%) 4.3 ( 8.4%) 11.1 ( 43.2%) 0.9 (20.4%)
m9 5.4 (14.7 %) 4.8 ( 19.8 %) 4.1 ( 83.7%) 3.6 ( 6.9%) 2.7 ( 10.4%) 1.3 (30.9%)
m11 7.1 (19.4 %) 2.6 ( 10.8 %) 0.4 ( 8.7%) 7.7 ( 14.9%) 2.0 ( 8.0%) 1.5 (33.8%)
m15 8.3 (22.9 %) 2.3 ( 9.3 %) 2.5 ( 51.1%) 2.1 ( 4.1%) 8.2 ( 32.2%) 0.5 (12.5%)
m16 8.8 (24.2 %) 2.2 ( 9.0 %) 0.3 ( 6.8%) 1.0 ( 1.9%) 0.5 ( 1.8%) 0.8 (18.8%)
m19 9.2 (25.3 %) 1.2 ( 5.0 %) 0.2 ( 3.2%) 1.1 ( 2.2%) 9.2 ( 36.0%) 1.5 (33.8%)
m20 5.4 (14.9 %) 4.9 ( 20.4 %) 0.1 ( 1.9%) 0.5 ( 0.9%) 4.2 ( 16.4%) 1.0 (22.9%)
Table 3: The table shows average values and associated standard deviation for parameters extracted from markers at a
stimulation intensity of 40 mA. Associated errors were placed below the results. The lowest errors are highlighted.
Vastus medialis Rectus femoris
Tc(ms) T
d
(ms) Dm(mm) Tc(ms) T
d
(ms) Dm(mm)
m5 30.9 ( 11.9) 29.1 ( 16.9) 6.9 ( 3.7) 47 ( 10.8) 35.4 ( 13.7) 4 ( 1)
m9 33.1 ( 14.5) 24 ( 14.1) 6.9 ( 5.6) 50.9 ( 7.8) 33.7 ( 10.6) 3.8 ( 1.7)
m11 35.3 ( 16) 24.8 ( 12.7) 5.5 ( 2.6) 55.3 ( 14.5) 30.3 ( 9.9) 3.8 ( 2.1)
m15 33.6 ( 15.1) 26.2 ( 13.7) 7.8 ( 4.7) 50.2 ( 7.3) 33.9 ( 7.2) 6.2 ( 1.7)
m16 36.2 ( 14) 25.9 ( 13.8) 5.3 ( 3) 54.3 ( 10) 29.1 ( 10.1) 4.3 ( 2)
m19 35.3 ( 13.4) 28.7 ( 15.5) 5.4 ( 4.1) 49.3 ( 6.1) 34.9 ( 10.8) 3.2 ( 0.8)
m20 35.6 ( 12.5) 28 ( 13.8) 5.1 ( 3.8) 49.9 ( 6.8) 32 ( 11) 4.3 ( 1.4)
TMG 31.3 ( 10.9) 23.1 ( 2.5) 5.7 ( 1.8) 45.1 ( 18.3) 24.8 ( 3.1) 4.9 ( 1.6)
Error(Tc) Error(T
d
) Error(Dm) Error(Tc) Error(T
d
) Error(Dm)
m5 0.3 (1.1 %) 6 ( 25.8 %) 1.2 ( 20.3%) 1.9 ( 4.2%) 10.7 ( 43%) 0.9 (18.2%)
m9 1.8 (5.7 %) 0.9 ( 3.8 %) 1.2 ( 21.0%) 5.8 ( 12.9%) 8.9 ( 36%) 1 (21.4%)
m11 4.1 (13 %) 1.6 ( 7.1 %) 0.2 ( 3.5%) 10.2 ( 22.7%) 5.5 ( 22.3%) 1.1 (21.7%)
m15 2.3 (7.4 %) 3.1 ( 13.2 %) 2.1 ( 36.3%) 5.1 ( 11.3%) 9.1 ( 36.9%) 1.3 (26.9%)
m16 4.9 (15.8 %) 2.8 ( 12.1 %) 0.4 ( 6.2%) 9.2 ( 20.5%) 4.3 ( 17.6%) 0.6 (11.8%)
m19 4 (12.8 %) 5.6 ( 24.1 %) 0.4 ( 6.2%) 4.2 ( 9.4%) 10.1 ( 40.8%) 1.6 (33.9%)
m20 4.3 (13.7 %) 4.8 ( 20.9 %) 0.6 ( 10.8%) 4.9 ( 10.8%) 7.2 ( 29.3%) 0.6 (11.9%)

Method for Estimating Tensiomyography Parameters from. . . Informatica 45 (2021) 213–222 221
Table 4: The table shows average values and associated standard deviation for parameters extracted from markers at a
stimulation intensity of 50 mA. Associated errors were placed below the results. The lowest errors are highlighted.
Vastus medialis Rectus femoris
Tc(ms) T
d
(ms) Dm(mm) Tc(ms) T
d
(ms) Dm(mm)
m5 28.93 ( 11.01) 23.89 ( 8.60) 7.35 ( 3.76) 48.14 ( 4.01) 38.70 ( 13.90) 12.86 ( 1.29)
m9 31.41 ( 12.99) 19.56 ( 4.33) 10.07 ( 5.65) 49.42 ( 4.20) 39.29 ( 16.25) 6.19 ( 2.01)
m11 33.45 ( 15.45) 19.91 ( 3.23) 6.62 ( 2.65) 46.98 ( 4.67) 39.21 ( 17.36) 9.83 ( 1.75)
m15 30.44 ( 14.53) 20.99 ( 5.26) 9.08 ( 4.76) 50.57 ( 6.98) 36.57 ( 9.64) 8.00 ( 1.73)
m16 33.12 ( 14.47) 21.45 ( 6.30) 6.50 ( 2.92) 48.79 ( 4.89) 39.78 ( 13.43) 4.56 ( 1.72)
m19 31.90 ( 13.95) 23.77 ( 8.49) 6.14 ( 4.32) 49.07 ( 3.34) 37.13 ( 11.97) 4.31 ( 0.78)
m20 31.93 ( 14.84) 23.11 ( 7.21) 6.09 ( 3.83) 53.01 ( 4.79) 29.9 ( 11.49) 9.15 ( 1.21)
TMG 28.86 ( 9.14) 23.08 ( 2.24) 6.31 ( 1.88) 42.04 ( 5.22) 24.44 ( 2.98) 18.97 ( 1.66)
Error(Tc) Error(T
d
) Error(Dm) Error(Tc) Error(T
d
) Error(Dm)
m5 0.1 (0.2 %) 0.8 ( 3.5 %) 1.0 ( 16.5%) 6.1 ( 14.5%) 14.3 ( 58.3%) 1.2 (23.1%)
m9 2.5 (8.8 %) 3.5 ( 15.3 %) 3.8 ( 59.5%) 7.4 ( 17.6%) 14.8 ( 60.5%) 1.0 (19.5%)
m11 4.6 (15.9 %) 3.2 ( 13.7 %) 0.3 ( 4.8%) 4.9 ( 11.8%) 14.8 ( 60.5%) 0.6 (10.6%)
m15 1.6 (5.5 %) 2.1 ( 9.0 %) 2.8 ( 43.9%) 8.5 ( 20.3%) 12.1 ( 49.6%) 1.8 (33.6%)
m16 4.3 (14.7 %) 1.6 ( 7.0 %) 0.2 ( 3.0%) 6.7 ( 16.0%) 15.3 ( 62.7%) 0.3 (6.4%)
m19 3.0 (10.5 %) 0.7 ( 3.0 %) 0.2 ( 3.0%) 7.0 ( 16.7%) 12.7 ( 51.9%) 1.9 (36.0%)
m20 3.1 (10.6 %) 0.0 ( 0.0 %) 0.2 ( 3.0%) 11.0 ( 26.1%) 5.5 ( 22.5%) 0.4 (8.3%)
Table 5: The table shows average values and associated standard deviation for parameters extracted from markers at a
stimulation intensity of 60 mA. Associated errors were placed below the results. The lowest errors are highlighted.
Vastus medialis Rectus femoris
Tc(ms) T
d
(ms) Dm(mm) Tc(ms) T
d
(ms) Dm(mm)
m5 29.5 ( 9) 25.4 ( 10.5) 7.6 ( 3.8) 47.3 ( 13.2) 35.2 ( 10.9) 4 ( 1.7)
m9 30.2 ( 12.2) 21.6 ( 8.9) 10.8 ( 5.6) 45.7 ( 4.3) 39 ( 19.2) 4.5 ( 2.4)
m11 32.8 ( 12.3) 20.7 ( 5.3) 7.6 ( 2.9) 40.9 ( 10.4) 43 ( 24.3) 5 ( 2.4)
m15 31.1 ( 14.1) 22 ( 7.1) 10.3 ( 4.8) 51.1 ( 11.7) 32 ( 6.4) 7.3 ( 3)
m16 33.6 ( 15.6) 22.8 ( 8.6) 7.5 ( 3) 45.1 ( 3) 38.8 ( 17.5) 5.1 ( 2.4)
m19 32.8 ( 13.4) 25 ( 10) 6.8 ( 4.3) 46.7 ( 4.9) 38.1 ( 18) 3.4 ( 1.3)
m20 32.8 ( 13.2) 25 ( 9.5) 7 ( 3.8) 51 ( 9) 31.9 ( 10.6) 4.7 ( 1.7)
TMG 28.2 ( 9.2) 23 ( 2.1) 6.9 ( 1.7) 41.5 ( 19.4) 24.5 ( 2.8) 5.7 ( 1.9)
Error(Tc) Error(T
d
) Error(Dm) Error(Tc) Error(T
d
) Error(Dm)
m5 1.3 (4.7 %) 2.4 ( 10.4 %) 0.6 ( 9.1%) 5.8 ( 13.9%) 10.8 ( 44.1%) 1.7 (29.3%)
m9 2.1 (7.4 %) 1.4 ( 6.2 %) 3.9 ( 55.8%) 4.1 ( 10%) 14.6 ( 59.5%) 1.2 (21.4%)
m11 4.6 (16.5 %) 2.3 ( 9.8 %) 0.7 ( 10%) 0.6 ( 1.4%) 18.5 ( 75.7%) 0.7 (11.8%)
m15 3 (10.5 %) 1 ( 4.5 %) 3.3 ( 48.2%) 9.6 ( 23%) 7.5 ( 30.7%) 1.6 (27.7%)
m16 5.4 (19.2 %) 0.2 ( 0.9 %) 0.6 ( 8.2%) 3.6 ( 8.6%) 14.4 ( 58.8%) 0.6 (10.8%)
m19 4.7 (16.6 %) 2 ( 8.8 %) 0.1 ( 1.7%) 5.2 ( 12.4%) 13.6 ( 55.8%) 2.3 (41%)
m20 4.7 (16.6 %) 1.9 ( 8.4 %) 0 ( 0.0%) 9.5 ( 22.8%) 7.5 ( 30.6%) 1 (17.3%)

222 Informatica 45 (2021) 213–222 D. Vlahek et al.
Table 6: The table shows average values and associated standard deviation for parameters extracted from markers at a
stimulation intensity of 70 mA. Associated errors were placed below the results. The lowest errors are highlighted.
Vastus medialis Rectus femoris
Tc(ms) T
d
(ms) Dm(mm) Tc(ms) T
d
(ms) Dm(mm)
m5 31.3 ( 11.7) 26.5 ( 24.9) 8 ( 3.7) 50.3 ( 19.2) 23.2 ( 7.1) 4.5 ( 1.4)
m9 37.6 ( 23.8) 21.4 ( 6.6) 12 ( 5.2) 51.3 ( 10.1) 25.9 ( 6.5) 5.3 ( 1.8)
m11 38.1 ( 23.7) 22 ( 7.5) 8.4 ( 2.7) 45.2 ( 5.6) 33.8 ( 12.8) 5.6 ( 1.9)
m15 33.3 ( 16.8) 26.1 ( 17.3) 11.4 ( 4.4) 46.1 ( 5.4) 31.8 ( 9.1) 8.4 ( 2)
m16 39.8 ( 24) 22.6 ( 8.5) 8.3 ( 2.8) 45.5 ( 3.1) 32.8 ( 13.1) 5.9 ( 1.7)
m19 34.8 ( 14.1) 30 ( 23.2) 7.6 ( 4.2) 53 ( 8.3) 23.3 ( 3.3) 3.9 ( 0.9)
m20 34.5 ( 12) 30.4 ( 25.3) 7.6 ( 3.6) 46.8 ( 3.4) 27.2 ( 8.5) 5.9 ( 1)
TMG 32 ( 19.5) 23.3 ( 2.7) 7.4 ( 2) 48.2 ( 24.4) 24.7 ( 3.1) 6.4 ( 2.3)
Error(Tc) Error(T
d
) Error(Dm) Error(Tc) Error(T
d
) Error(Dm)
m5 0.7 (2.2 %) 3.2 ( 13.7 %) 0.6 ( 7.7%) 2.1 ( 4.4%) 1.5 ( 6.1%) 1.9 (30.2%)
m9 5.6 (17.5 %) 2 ( 8.5 %) 4.6 ( 61.7%) 3.1 ( 6.4%) 1.2 ( 4.7%) 1.1 (16.5%)
m11 6.1 (19.1 %) 1.3 ( 5.8 %) 1 ( 13.6%) 3 ( 6.1%) 9 ( 36.5%) 0.8 (12.7%)
m15 1.3 (3.9 %) 2.8 ( 11.9 %) 4 ( 53.3%) 2.1 ( 4.3%) 7 ( 28.4%) 2 (30.9%)
m16 7.8 (24.4 %) 0.8 ( 3.2 %) 0.9 ( 11.5%) 2.8 ( 5.7%) 8.1 ( 32.6%) 0.5 (7.4%)
m19 2.8 (8.6 %) 6.7 ( 28.6 %) 0.1 ( 1.7%) 4.8 ( 9.9%) 1.4 ( 5.8%) 2.5 (39.2%)
m20 2.5 (7.7 %) 7.1 ( 30.4 %) 0.2 ( 2.6%) 1.4 ( 3%) 2.5 ( 10%) 0.5 (7.5%)
Table 7: The table shows average values and associated standard deviation for parameters extracted from markers at a
stimulation intensity of 80 mA. Associated errors were placed below the results. The lowest errors are highlighted.
Vastus medialis Rectus femoris
Tc(ms) T
d
(ms) Dm(mm) Tc(ms) T
d
(ms) Dm(mm)
m5 26.8 ( 7.6) 21.1 ( 13.3) 8.5 ( 3.4) 39.4 ( 12.7) 20.9 ( 10.0) 4.1 ( 1.3)
m9 29.7 ( 15.3) 15.2 ( 1.7) 13.1 ( 4.6) 36.8 ( 9.4) 25.2 ( 12.7) 5.7 ( 1.4)
m11 32.8 ( 17.2) 15.8 ( 2.3) 9.3 ( 2.3) 35.2 ( 9.1) 24.4 ( 13.2) 4.8 ( 1.8)
m15 29.0 ( 14.0) 17.6 ( 6.2) 12.5 ( 3.8) 31.3 ( 9.4) 31.4 ( 11.8) 7.1 ( 2.4)
m16 32.2 ( 18.4) 16.2 ( 2.9) 9.2 ( 2.5) 31.5 ( 4.5) 31.5 ( 13.5) 4.9 ( 1.3)
m19 30.3 ( 10.7) 20.1 ( 10.9) 8.1 ( 4.0) 35.2 ( 7.7) 25.1 ( 14.6) 3.4 ( 1.3)
m20 28.5 ( 10.9) 20.5 ( 11.7) 8.3 ( 3.3) 33.9 ( 5.8) 26.1 ( 13.5) 5.2 ( 1.1)
TMG 31.4 ( 19.5) 23.8 ( 3.6) 7.6 ( 2.3) 47.6 ( 24.3) 24.8 ( 3.3) 6.7 ( 2.7)
Error(Tc) Error(T
d
) Error(Dm) Error(Tc) Error(T
d
) Error(Dm)
m5 4.6 (14.6 %) 2.7 ( 11.3 %) 0.9 ( 11.7%) 8.2 ( 17.1%) 3.9 ( 15.6%) 2.6 (39.1%)
m9 1.7 (5.4 %) 8.6 ( 36 %) 5.4 ( 71%) 10.7 ( 22.6%) 0.4 ( 1.6%) 1 (15.0%)
m11 1.4 (4.5 %) 8 ( 33.8 %) 1.7 ( 22%) 12.4 ( 26.1%) 0.4 ( 1.6%) 1.9 (28.8%)
m15 2.4 (7.7 %) 6.2 ( 26.1 %) 4.9 ( 64.1%) 16.2 ( 34.1%) 6.6 ( 26.8%) 0.5 (7.1%)
m16 0.8 (2.4 %) 7.6 ( 32.1 %) 1.6 ( 20.9%) 16 ( 33.7%) 6.7 ( 27.1%) 1.8 (26.7%)
m19 1.1 (3.5 %) 3.7 ( 15.5 %) 0.5 ( 6.4%) 2.4 ( 26%) 0.3 ( 1.4%) 3.3 (48.8%)
m20 2.9 (9.2 %) 3.3 ( 14 %) 0.6 ( 8.2%) 13.6 ( 28.6%) 1.3 ( 5.2%) 1.4 (21.4%)