ArticlePDF Available

Musculoskeletal model-based inverse dynamic analysis under ambulatory conditions using inertial motion capture

Authors:

Abstract and Figures

Inverse dynamic analysis using musculoskeletal modeling is a powerful tool, which is utilized in a range of applications to estimate forces in ligaments, muscles, and joints, non-invasively. To date, the conventional input used in this analysis is derived from optical motion capture (OMC) and force plate (FP) systems, which restrict the application of musculoskeletal models to gait laboratories. To address this problem, we propose the use of inertial motion capture to perform musculoskeletal model-based inverse dynamics by utilizing a universally applicable ground reaction force and moment (GRF&M) prediction method. Validation against a conventional laboratory-based method showed excellent Pearson correlations for sagittal plane joint angles of ankle, knee, and hip (ρ=0.95, 0.99, and 0.99, respectively) and root-mean-squared-differences (RMSD) of 4.1 ± 1.3° 4.4 ± 2.0° and 5.7 ± 2.1° respectively. The GRF&M predicted using IMC input were found to have excellent correlations for three components (vertical: ρ=0.97, RMSD = 9.3 ± 3.0 %BW, anteroposterior: ρ=0.91, RMSD = 5.5 ± 1.2 %BW, sagittal: ρ=0.91, RMSD = 1.6 ± 0.6 %BW*BH), and strong correlations for mediolateral (ρ=0.80, RMSD = 2.1 ± 0.6 %BW) and transverse (ρ=0.82, RMSD = 0.2 ± 0.1 %BW*BH). The proposed IMC-based method removes the complexity and space restrictions of OMC and FP systems and could enable applications of musculoskeletal models in either monitoring patients during their daily lives or in wider clinical practice.
Content may be subject to copyright.
Aalborg Universitet
Musculoskeletal model-based inverse dynamic analysis under ambulatory conditions
using inertial motion capture
Karatsidis, Angelos; Jung, Moonki; Schepers, Martin; Bellusci, Giovanni; de Zee, Mark;
Veltink, Peter H.; Andersen, Michael Skipper
Published in:
Medical Engineering & Physics
Publication date:
2019
Document Version
Accepted author manuscript, peer reviewed version
Link to publication from Aalborg University
Citation for published version (APA):
Karatsidis, A., Jung, M., Schepers, M., Bellusci, G., de Zee, M., Veltink, P. H., & Andersen, M. S. (Accepted/In
press). Musculoskeletal model-based inverse dynamic analysis under ambulatory conditions using inertial
motion capture. Medical Engineering & Physics.
General rights
Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners
and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights.
? Users may download and print one copy of any publication from the public portal for the purpose of private study or research.
? You may not further distribute the material or use it for any profit-making activity or commercial gain
? You may freely distribute the URL identifying the publication in the public portal ?
Take down policy
If you believe that this document breaches copyright please contact us at vbn@aub.aau.dk providing details, and we will remove access to
the work immediately and investigate your claim.
Title:
Musculoskeletal model-based inverse dynamic analysis under ambulatory con-
ditions using inertial motion capture
Authors:
Angelos Karatsidis1,2, Moonki Jung3, H. Martin Schepers1, Giovanni Bellusci1,
Mark de Zee4, Peter H. Veltink2, Michael Skipper Andersen5
Affiliations:
1Xsens Technologies B.V., Enschede 7521 PR, The Netherlands
2Department of Biomedical Signals and Systems, Faculty of Electrical Engineer-
ing, Mathematics and Computer Science, University of Twente, Enschede 7500
AE, The Netherlands
3AnyBody Technology A/S, Aalborg 9220, Denmark
4Department of Health Science and Technology, Aalborg University, Aalborg
9220, Denmark
5Department of Materials and Production, Aalborg University, Aalborg 9220,
Denmark
Corresponding author:
Angelos Karatsidis, Address: Xsens Technologies B.V., Pantheon 6-8, Enschede
7521 PR, The Netherlands, E-mail: angelos.karatsidis@xsens.com, Tel.: +31 88
97367 36, Fax: +31 88 97367 01
Keywords:
musculoskeletal modeling, inertial motion capture, inverse dynamics, ground
reaction forces and moments, gait analysis
Wordcount:
Abstract: 197 words
Main text: 3582 words
1
Accepted manuscript
Abstract1
Inverse dynamic analysis using musculoskeletal modeling is a powerful tool,2
which is utilized in a range of applications to estimate forces in ligaments, mus-3
cles, and joints, non-invasively. To date, the conventional input used in this4
analysis is derived from optical motion capture (OMC) and force plate (FP)5
systems, which restrict the application of musculoskeletal models to gait labo-6
ratories. To address this problem, we propose the use of inertial motion cap-7
ture to perform musculoskeletal model-based inverse dynamics by utilizing a8
universally applicable ground reaction force and moment (GRF&M) prediction9
method. Validation against a conventional laboratory-based method showed10
excellent Pearson correlations for sagittal plane joint angles of ankle, knee, and11
hip (ρ= 0.95, 0.99, and 0.99, respectively) and root-mean-squared-differences12
(RMSD) of 4.1±1.3, 4.4±2.0, and 5.7±2.1, respectively. The GRF&M pre-13
dicted using IMC input were found to have excellent correlations for three com-14
ponents (vertical: ρ= 0.97, RMSD=9.3±3.0 %BW, anteroposterior: ρ= 0.91,15
RMSD=5.5±1.2 %BW, sagittal: ρ= 0.91, RMSD=1.6±0.6 %BW*BH), and16
strong correlations for mediolateral (ρ= 0.80, RMSD=2.1±0.6 %BW ) and17
transverse (ρ= 0.82, RMSD=0.2±0.1 %BW*BH). The proposed IMC-based18
method removes the complexity and space-restrictions of OMC and FP systems19
and could enable applications of musculoskeletal models in either monitoring20
patients during their daily lives or in wider clinical practice.21
2
Accepted manuscript
1. Introduction22
Assessment of muscle, joint, and ligament forces is important to understand23
the mechanical and physiological mechanisms of human movement. To date,24
the measurement of such in-vivo forces is a challenging task. For this reason,25
computer-based musculoskeletal models have been widely used to estimate the26
variables of interest non-invasively [1, 2].27
The most common approach used in musculoskeletal modeling is the method28
of the inverse dynamics [3]. This analysis utilizes the equations of motion with29
input from human body kinematics in conjunction with kinetics obtained from30
external forces [4], to estimate joint reaction and muscle forces, as well as net31
joint moments using muscle recruitment methods [5]. Measurements of the32
external forces are typically required and measured using force plates (FPs),33
however, the use of FPs has several limitations. First, subjects tend to alter34
their natural gait patterns in order to hit the small and fixed measurement area35
of a plate [6]. In addition, this static and limited measurement area, impedes36
the assessment of several consecutive steps, when only a couple of FPs are37
available. Finally, the combined use of FP with motion input introduces a38
dynamic inconsistency, which results to residual forces and moments in the39
inverse dynamics. [7, 8].40
Several studies have proposed replacing the FP input with wearable de-41
vices such as shoes with three-dimensional force and torque sensors beneath42
the sole [9, 10, 11]. In a similar fashion, pressure insoles were proposed to re-43
construct the complete ground reaction forces and moments (GRF&M) from44
pressure distributions [12, 13, 14]. Although these wearable devices are suitable45
for the assessment of external forces, the increased height and weight of the46
shoes equipped with force/torque sensors [15, 16], as well as the repeatability47
of the pressure sensors [17] are considered important limitations.48
Recent research has suggested the replacement of the force input with predic-49
tions derived solely from motion input [18, 19, 20, 21, 22, 23]. In these studies,50
human body kinematics are combined with the inertial properties of the body51
3
Accepted manuscript
segments, from which Newton-Euler equations are utilized to compute the exter-52
nal forces and moments. Since the system of equations becomes indeterminate53
during the double stance of gait, each of the aforementioned studies focused on54
methods to solve this issue. Ren et al. [19] suggested a gait event-based func-55
tion which is only applicable in gait, while Oh et al. [20] and Choi et al. [21]56
suggested methods based on a machine learning that require a training database57
and thus are not applicable for movements not included in that database. A58
last approach enables the universal application of these methods using a muscle59
recruitment approach has shown promising performance for various activities of60
daily living [22] and sports [23].61
The majority of the existing research which studied the prediction of GRF&M,62
used conventional optical motion capture (OMC) input. Despite the high ac-63
curacy of this method in tracking marker trajectories, its dependence on lab-64
oratory equipment restricts possible applications during daily life activities or65
in wider clinical practice. In the previous decade, ambulatory motion tracking66
systems based on inertial measurement units (IMUs), have been proposed as a67
suitable alternative for estimating 3D segment kinematics [24, 25, 26, 27]. A68
key benefit of such systems is that they can be applied in virtually any environ-69
ment without depending on external infrastructure, such as cameras. Driven70
by these advances in inertial motion capture (IMC), recent work of the authors71
demonstrated its ability to estimate three-dimensional GRF&M [28], which were72
distributed between the feet using a smooth transition assumption concept [19].73
However, limitations of that approach is that it is only valid for gait and has no74
muscle, bone or ligament force estimate capabilities.75
To date, the use of detailed musculoskeletal modeling with kinematic inputs76
from IMUs has only received limited attention. Koning et al. [29] previously77
demonstrated the feasibility of kinematically driving a musculoskeletal model78
using only orientations from IMUs. However, that study only compared the79
kinematics of the musculoskeletal model, without any inverse dynamic calcula-80
tions.81
The aim of this study was to develop a workflow to perform musculoskeletal82
4
Accepted manuscript
model-based inverse dynamics using exclusively IMC input, applicable in am-83
bulatory environments and validate it against a conventional laboratory-based84
approach.85
2. Methods86
2.1. Subjects87
The experimental data was collected at the Human Performance Labora-88
tory, at the Department of Health Science and Technology, Aalborg University,89
Aalborg, Denmark following the ethical guidelines of The Scientific Ethical Com-90
mittee for the Region of North Jutland (Den Videnskabsetiske Komit for Region91
Nordjylland). Eleven healthy male individuals with no present musculoskeletal92
or neuromuscular disorders volunteered for the study (age: 31.0 ±7.2 years;93
height: 1.81 ±0.06 m; weight: 77.3 ±9.2 kg; body mass index (BMI): 23.6 ±94
2.4 kg/m2). All participants provided written informed consent, prior to data95
collection.96
2.2. Instrumentation97
Full-body IMC data were collected using the Xsens MVN Link (Xsens Tech-98
nologies B.V., Enschede, the Netherlands), in which 17 IMUs were mounted on99
the head, sternum, pelvis, upper legs, lower legs, feet, shoulders, upper arms,100
forearms and hands using the dedicated clothing. The exact location of each101
sensor on the respective segment followed the manufacturer guidelines described102
in the manual of Xsens MVN [30]. The affiliated software Xsens MVN Studio103
4.2.4 was used to track the IMU orientations with respect to an earth-based104
coordinate frame [24, 25]. Segment orientations were obtained by applying the105
IMU-to-segment alignment, found using a known upright pose (N-pose) per-106
formed by the subject at a known moment in time, while taking specific care107
to minimize the effect of magnetic disturbances. In addition, this information108
is fused with updates regarding the joints and external contacts to limit the109
position drift [26].110
5
Accepted manuscript
For validation purposes, an OMC system utilizing 8 infrared high speed111
cameras (Oqus 300 series, Qualisys AB, Gothenburg, Sweden) and the software112
Qualisys Track Manager 2.12 (QTM) were used to track the trajectories of 53113
reflective markers mounted on the human body, as described in the Appendix of114
[28]. In addition, three FP systems (AMTI OR6-7-1000, Advanced Mechanical115
Technology, Inc.,Watertown, MA, USA) embedded in the floor of the laboratory,116
were utilized using QTM to record the GRF&Ms. Both IMC and OMC systems117
sampled data at a frequency of 240 Hz, while the FP system sampled data at118
2400 Hz and subsequently downsampled to 240 Hz to match the IMC and OMC119
sampling rate.. A second-order forward-backward low-pass Butterworth filter120
was applied to the reflective marker trajectories and measured GRF&M, with121
cut-off frequencies of 6 Hz and 15 Hz, respectively.122
2.3. Experimental protocol123
For each participant, the body dimensions were extracted using a conven-124
tional tape following the guidelines of Xsens. During the data collection, the125
subjects were instructed to walk barefoot in three different walking speeds (com-126
fortable; CW, fast; FW, and slow; SW). The walking speeds performed experi-127
mentally were quantified as 1.28±0.14 m/s (mean ±standard deviation) for CW,128
1.58±0.09 m/s for FW (CW + 23%) and 0.86±0.11 m/s for SW (CW33%).129
For every walking speed, five successful trials were assessed. A successful trial130
was obtained when a single foot hit one of the FPs entirely, followed by an entire131
hit of the other foot on the successive FP.132
2.4. Overall description of the components in the musculoskeletal models133
Three musculoskeletal models have been constructed in AnyBodyTM Mod-134
eling System (AMS) v.6.0.7 (AnyBodyTM Technology A/S, Aalborg, Denmark)135
[1]:136
a model in which the kinematics are driven by IMC and the GRF&M are137
predicted from the kinematics (IMC-PGRF).138
6
Accepted manuscript
a model in which the kinematics are driven by OMC and the GRF&M are139
predicted from the kinematics (OMC-PGRF).140
a model in which the kinematics are driven by OMC and the GRF&M are141
measured from FPs (OMC-MGRF).142
In the IMC-PGRF model, a Biovision Hierarchy (BVH) file is exported from143
Xsens MVN Studio and imported in AMS, in which a stick figure model is ini-144
tially reconstructed. The BVH file contains a hierarchy part with a description145
of the linked segment model in a static pose, as well as a motion part that146
contains, for each time frame, the absolute position and orientation of the root147
pelvis segment, and the joint angles between the segments described in the hier-148
archy. The generated stick figure model contains 72 degrees-of-freedom (DOF).149
In order to match the stick figure model with the musculoskeletal model, we150
utilize a concept of virtual markers (VMs) demonstrated in a previous Kinect-151
based study [31]. The VMs are mapped to particular points of each model that152
are well defined in both models, such as joint centers and segment end points.153
The VM placement is illustrated in Figure 1 and described in more detail in154
the supplementary material. Following this step, the VMs are treated as actual155
experimental markers, as if they were derived from an OMC system and they156
are assigned weights in three directions in the segmentframe. Contrary to OMC,157
no filtering was applied to the VM trajectories.158
In all models, the GaitFullBody template of the AnyBodyTM Managed159
Model Repository (AMMR) 1.6.2 was used to reconstruct the musculoskele-160
tal models in AMS. The lumbar spine model was derived from the study of161
de Zee et al. [32], the lower limb model was derived from the Twente Lower162
Extremity Model Klein-Horsman et al. [33], and the shoulder and upper limb163
models were based on the model of the Delft Shoulder Group [34, 35, 36]. The164
full-body kinematic model contained 39 DOF in total. Specifically, a pelvis165
segment with three rotational and three translational DOF, two spherical hip166
joints, two revolute knee joints, two universal ankle joints, a spherical pelvic-167
lumbar joint, two glenohumeral joints with five DOF each, two universal elbow168
7
Accepted manuscript
joints, and two universal wrist joints. The motion of the neck joint was locked169
to a neutral position.170
2.5. Scaling and kinematics analysis of the musculoskeletal models171
For each subject, a standing reference trial with an anatomical pose was172
utilized to identify the parameters of segment lengths and the (virtual) marker173
positions, using a least-square minimization between the model and input (vir-174
tual or skin-mounted) marker positions [37]. In the IMC-PGRF musculoskeletal175
model, the lengths of the shanks, thighs, head, upper arm and forearms were176
derived directly from the stick figure, as generated from Xsens MVN studio us-177
ing the measured body dimensions. In contrast, the pelvis width, foot length,178
and trunk height were optimized based on the above-mentioned least-square179
minimization method. The estimated segment lengths were used in all subse-180
quent dynamic trials to perform the kinematic analysis based on the method of181
Andersen et al. [38].182
2.6. Inertial and geometric scaling of the musculoskeletal models183
The mass of each segment was linearly scaled based on the total body mass184
and the segment mass ratio values reported by Winter [4]. The inertial pa-185
rameters were calculated by considering the segments as cylinders with uniform186
density. In addition, geometric scaling of each segment, where the longitudinal187
axis was defined as the second entry, was achieved using the following matrix:188
S=
qms
ls0 0
0ls0
0 0 qms
ls
(1)
where Sis the scaling matrix, lsis the ratio between the unscaled and scaled189
lengths of the segment, msis the mass ratio of the segment.190
8
Accepted manuscript
2.7. Muscle recruitment191
The muscle recruitment problem was solved by defining an optimization192
problem where a system of equations minimizes the cost function, subject to193
the dynamic equilibrium equations and non-negativity constraints, so that each194
muscle can only pull, but not push, while its force remains below its strength195
[1, 31, 39].196
The strengths of the muscles were derived from previous studies which de-197
scribed the models of the body parts, and were considered constant for different198
lengths and contraction velocities [32, 33, 34, 35, 36]. To scale the muscle199
strengths, fat percentage was used as in Veeger et al. [35], calculated from the200
body mass index [40]. The model of the lower body contained 110 muscles,201
distributed into 318 individual muscle paths. In contrast, in the upper body202
model, ideal joint torque generators were utilized. Actuators for residual forces203
and moments with capacity up to 10 N and Nm, respectively, were placed at the204
origin of the pelvis and included in the muscle recruitment problem previously205
described.206
2.8. Ground reaction force and moment prediction207
The GRF&M were predicted by adjusting a method of Skals et al. [23]. A208
set of eighteen dynamic contact points were overlaid 1 mm beneath the inferior209
surface of each foot. Each dynamic contact point consisted of five unilateral210
force actuators, which could generate a positive vertical force perpendicular to211
the ground, and static friction forces in the anterior, posterior, medial, and212
lateral directions using a friction coefficient of 0.5. In addition, the height and213
velocity activation thresholds were set to 0.03 m and 1.2 m/s, respectively.214
2.9. Data Analysis215
Lower limb joint angles calculated in the IMC-PGRF model were compared216
to the OMC-PGRF/OMC-MGRF. In addition, GRF&M and JRF&M of the217
IMC-PGRF and OMC-PGRF were compared to OMC-MGRF.218
9
Accepted manuscript
Forces were normalized to body weight (BW) and moments to body weight219
times body height (BW*BH). The time axis of the curves was normalized to220
100% of the gait cycle for the kinematics (time between two consecutive heel-221
strike events of the analyzed limb) and 100% of the stance phase (time between222
heel-strike and toe-off events of the analyzed limb) for the kinetics. Measured223
and estimated GRF&M were expressed on the right handed coordinate frame224
defined by the walking direction within the trial (given that the subjects walked225
straight) and the vertical axis equal to the vertical axis of the respective mo-226
tion capture system used. On the other hand, JRF&M were expressed on the227
coordinate frame of the segment distal to the body in both IMC and OMC228
methods.229
The above-mentioned comparisons of kinematic and kinetic variables to their230
respective references were performed using absolute and relative root-mean-231
square-differences (RMSD and rRMSD, respectively)as described by Ren et al.232
[19]. In addition, for every curve, the magnitude (M) and phase (P) differ-233
ence metrics [41] have been utilized. Pearson correlation coefficient (ρ) were234
calculated, averaged using Fisher’s z transformation method [42], and cate-235
gorized similarly to Taylor et al. [43], as ”weak” (ρ0.35), ”moderate”236
(0.35 < ρ 0.67), ”strong” (0.67 < ρ 0.90), and ”excellent” (ρ > 0.90).237
3. Results238
3.1. Estimated kinematics of the musculoskeletal model239
Table 1 presents the results for the accuracy analysis for the joint angles240
of the IMC-driven model versus the OMC-driven model. Similarly, Figure 2241
illustrates the curves for the joint angles of the lower extremities averaged across242
all gait cycles performed by the eleven subjects. Excellent Pearson correlation243
coefficients have been found in all sagittal plane angles for ankle, knee, and hip244
(0.95,0.99, and 0.99, respectively). For the same variables, the RMSDs across a245
gait cycle were found as 4.1±1.3, 4.4±2.0and 5.7±2.1, respectively (mean246
±standard deviation). Hip flexion angles were overall underestimated (M=247
10
Accepted manuscript
4.0±13.9%), whereas knee and ankle magnitude differences showed an average248
overestimation (0.7±6.2% and 8.6±16.4%). The hip abduction showed excellent249
correlations (ρ= 0.91) with an RMSD of 4.1±2.0and a mean underestimation250
with a magnitude difference M=12.2±34.7%. Strong correlation values (ρ=251
0.68) were observed in the hip internal-external rotation angle with an RMSD252
of 6.5±2.8and an underestimation of magnitude difference M= 5.5±39.0%.253
Finally, the subtalar eversion angle showed strong correlation (ρ= 0.82), RMSD254
of 9.66 ±3.07and M= 24.0±34.7%.255
3.2. Predicted kinetics using inertial and optical motion capture256
The results of the accuracy analysis for GRF&M and JRF&M are presented257
in Table 2 and 3, for IMC-PGRF and OMC-PGRF, respectively. The mean258
values and standard deviations of the curves from IMC-PGRF, OMC-PGRF,259
and OMC-MGRF models, are illustrated in Figures 3 and 4, for the forces and260
moments, respectively.261
The Pearson correlation coefficients of the IMC-PGRF model were excellent262
for vertical (ρ= 0.97) and anteroposterior GRF&M (ρ= 0.91) and strong for263
mediolateral GRF&M (ρ= 0.80). For the same components, RMSD values264
observed were of 9.3±3.0, 5.5±1.2 and 2.1±0.6 %BW, respectively (mean265
±standard deviation). The OMC-PGRF model performed better in the an-266
teroposterior GRF&M components (ρ= 0.96, RMSD = 3.7±1.1 %BW), and267
similarly to IMC-PGRF for the other two GRF&M components (mediolateral:268
ρ= 0.79, RMSD = 1.9±0.5 BW, vertical: ρ= 0.99, RMSD = 5.9±1.9 BW).269
Concerning GRM, the sagittal plane was predicted with similar excellent270
correlations in both IMC-PGRF (ρ= 0.91) and OMC-PGRF (ρ= 0.94) driven271
models. The correlation coefficients for frontal and transverse GRM components272
found in the IMC-PGRF model were ρ= 0.64, ρ= 0,82, respectively, whereas273
in the OMC-PGRF model (ρ= 0.66, ρ= 0,81, respectively). The RMSDs274
found in the IMC-PGRF approach were 0.9±0.6, 1.6±0.6 , and 0.2±0.001275
%BW*BH for frontal, sagittal and transverse GR&M, respectively, which were276
either slightly higher or similar to the RMSDs of the OMC-PGRF approach277
11
Accepted manuscript
(0.7±0.2, 1.2±0.4, and 0.2±0.1 %BW*BH, respectively).278
4. Discussion279
We have presented a method to perform musculoskeletal model-based in-280
verse dynamics using exclusively IMC input (IMC-PGRF). First, we compared281
the kinematic joint angle estimates of the lower limbs against those assessed282
through a conventional, laboratory-based OMC input. In addition, we tested283
the performance of the approach in calculating the JRF&M, while predicting284
the GRF&M from the kinematics, against a similarly constructed model (OMC-285
MGRF) which uses input from both FP and OMC. Finally, we performed a sim-286
ilar comparison to evaluate the predicted kinetics of a model driven exclusively287
by OMC input (OMC-PGRF).288
Regarding the IMC-based joint angles in the musculoskeletal model, all three289
sagittal plane angles provided excellent correlations (range: 0.95-0.99) and aver-290
age RMSD values remained below 6. Slightly lower correlations were observed291
in the frontal and transverse plane angles, which can be explained due to the292
smaller range of motion within these planes. For instance, even though the293
hip abduction and external rotation joint angles present absolute RMSD values294
similar to the flexion component, their rRMSDs which take into account the295
range of motion are two and three times higher, respectively.296
Both GRF&M and JRF&M of the vertical axis presented higher correlations297
and lower RMSDs than the ones in the anteroposterior and mediolateral axes.298
Similarly, sagittal plane moments were found in most cases to be more accurate299
than frontal and transverse plane moments. By visual inspection of the curves,300
we observe that the magnitude of the IMC-PGRF anteroposterior GRF&M301
seems to be underestimated both in the negative early stance and positive late302
stance peak, which can be confirmed by the magnitude difference for that curve303
(M=28.3%). However, this behaviour is not observed in the OMC-PGRF,304
nor during the single stance of the IMC-PGRF curve. Despite the higher rRMSD305
found in the non-sagittal joint angles, the performance of the IMC-PGRF in306
12
Accepted manuscript
the mediolateral, frontal and transverse plane GRF&M components matched307
closely the OMC-PGRF approach. This observation reveals that OMC-based308
kinematics suffer from errors of similar size, when capturing the typically small309
movements of the frontal and transverse planes, given the fact that both IMC-310
PGRF and OMC-PGRF had the same model characteristics. Therefore, OMC-311
MGRF should also be used with caution, when comparing either kinematic or312
JRF&M quantities of the non-sagittal planes.313
A number of error sources contribute to discrepancies in the OMC kinemat-314
ics. First, soft tissue artefacts can create a relative movement of the marker315
with respect to the bone [44, 45]. In addition, mismatches between the experi-316
mental and modelled marker positions can lead to errors in segment orientations317
calculated during inverse kinematics. Both error sources would have a relatively318
larger impact on the kinematics of the frontal and transverse plane, than on the319
sagittal plane. Finally, the JRF&M of the OMC-PGRF were compared against320
a non-independent OMC-MGRF reference, which could have contributed to un-321
derestimation of the actual errors.322
The IMC-PGRF approach has a number of possible sources of errors which323
would influence the performance. Similarly to OMC models, soft-tissue ar-324
tifacts may compromise the kinematic estimates. Further errors in segment325
kinematics may stem due to the N-pose calibration assumptions. In particular,326
mismatches between the practised and modelled N-pose could result in offsets327
in the estimated positions. Other common error sources in IMC include manual328
measurements of segment lengths as well as IMU inaccuracies. In addition, the329
stick figure model, which was utilized to recreate the VMs, has a higher number330
of DOF, compared to the musculoskeletal model used.331
A possible source of error in all inverse dynamic approaches concerns the332
inertial parameters (masses and moments of inertia), as well as the center of333
mass (CoM) locations of each human body segment, which were calculated334
based on anthropometric tables found in the literature.335
This study focused on presenting and evaluating a general workflow for mus-336
culoskeletal model-based inverse dynamic simulations using ambulatory IMC337
13
Accepted manuscript
systems. The presentation of results in this study was performed on the level of338
ground and joint reaction forces and moments. These measures are calculated339
from muscle force estimates derived from a muscle recruitment optimization340
technique. Given the high number of muscles in the model (110) and without341
a clear medical research question, it is challenging to choose which muscles are342
more important to present and analyze. Future studies could examine specific343
applications and pathologies in order to identify the most important muscles344
and evaluate their respective force estimates.345
A limitation of this study is that, even though the method has been pre-346
viously shown to be universally applicable in OMC-based studies [22, 23], we347
only evaluated its performance in gait of three different speeds. In addition,348
our experiments included only young healthy male subjects, but the underlying349
methods to predict kinetics from kinematics have been recently shown to be350
applicable in Parkinson’s patients [46]. Future studies could investigate the ap-351
plication of IMC systems combined with musculoskeletal modeling in groups of352
larger sample size than the current study, including patients, as well as female353
subjects.354
5. Conclusion355
In this study, we have demonstrated a workflow to perform musculoskeletal356
model-based inverse dynamics using input from a commercially available IMC357
system. Our validation findings indicate that the prediction of GRF&M as well358
as JRF&M using musculoskeletal model-based inverse dynamics based on only359
IMC data provides comparable performance to both OMC-PGRF and OMC-360
MGRF methods. The proposed method allows assessment of kinetic variables361
outside the laboratory.362
Ethical approval363
The ethical guidelines of The Scientific Ethical Committee for the Region of364
North Jutland (Den Videnskabsetiske Komit for Region Nordjylland) were fol-365
14
Accepted manuscript
lowed and all volunteers signed written informed consent after receiving detailed366
information prior to data collection.367
Conflict of interest statement368
Three of the authors are employees of Xsens Technologies B.V. that manu-369
factures and sells the Xsens MVN. One of the authors is employee of AnyBody370
Technology A/S that owns and sells the AnyBody Modeling System.371
Acknowledgements372
This study was performed in the context of KNEEMO Initial Training Net-373
work, funded by the European Unions Seventh Framework Programme for re-374
search, technological development, and demonstration under Grant Agreement375
No. 607510 (www.kneemo.eu). This work was also supported by the Danish376
Council for Independent Research under grant no. DFF-4184-00018 to M. S.377
Andersen. Finally, this research received funding in part from the European378
Unions Horizon 2020 research and innovation programme under grant agree-379
ment No. 680754 (The MovAiD project, www.movaid.eu).380
15
Accepted manuscript
References381
References382
[1] M. Damsgaard, J. Rasmussen, S. T. Christensen, E. Surma, M. de Zee,383
Analysis of musculoskeletal systems in the anybody modeling system, Sim-384
ulation Modelling Practice and Theory 14 (2006) 1100–1111.385
[2] S. L. Delp, F. C. Anderson, A. S. Arnold, P. Loan, A. Habib, C. T. John,386
E. Guendelman, D. G. Thelen, Opensim: Open-source software to cre-387
ate and analyze dynamic simulations of movement, IEEE Transactions388
on Biomedical Engineering 54 (2007) 1940–1950. doi:10.1109/TBME.2007.389
901024.390
[3] A. Erdemir, S. McLean, W. Herzog, A. J. van den Bogert, Model-based391
estimation of muscle forces exerted during movements, Clinical Biomechan-392
ics 22 (2007) 131 – 154. doi:https://doi.org/10.1016/j.clinbiomech.393
2006.09.005.394
[4] D. A. Winter, Biomechanics and Motor Control of Human Movement395
(1990).396
[5] J. Rasmussen, M. Damsgaard, M. Voigt, Muscle recruitment by the397
min/max criterion - a comparative numerical study, Journal of Biome-398
chanics 34 (2001) 409–415. doi:10.1016/S0021-9290(00)00191-3.399
[6] J. H. Challis, The variability in running gait caused by force plate targeting,400
Journal of Applied Biomechanics 17 (2001) 77–83.401
[7] R. Riemer, E. T. Hsiao-Wecksler, X. Zhang, Uncertainties in inverse dy-402
namics solutions: A comprehensive analysis and an application to gait, Gait403
and Posture 27 (2008) 578–588. doi:10.1016/j.gaitpost.2007.07.012.404
[8] H. Hatze, The fundamental problem of myoskeletal inverse dynamics and405
its implications, Journal of Biomechanics 35 (2002) 109–115. doi:10.1016/406
S0021-9290(01)00158-0.407
16
Accepted manuscript
[9] P. H. Veltink, C. Liedtke, E. Droog, H. Van Der Kooij, Ambulatory mea-408
surement of ground reaction forces, IEEE Transactions on Neural Systems409
and Rehabilitation Engineering 13 (2005) 423–427. doi:10.1109/TNSRE.410
2005.847359.411
[10] H. M. Schepers, H. F. J. M. Koopman, P. H. Veltink, Ambulatory as-412
sessment of ankle and foot dynamics, IEEE Transactions on Biomedical413
Engineering 54 (2007) 895–902. doi:10.1109/TBME.2006.889769.414
[11] T. Liu, Y. Inoue, K. Shibata, A wearable force plate system for the con-415
tinuous measurement of triaxial ground reaction force in biomechanical ap-416
plications, Measurement Science and Technology 21 (2010). doi:10.1088/417
0957-0233/21/8/085804.418
[12] A. Forner Cordero, H. J. F. M. Koopman, F. C. T. Van Der Helm, Use of419
pressure insoles to calculate the complete ground reaction forces, Journal420
of Biomechanics 37 (2004) 1427–1432. doi:10.1016/j.jbiomech.2003.12.421
016.422
[13] H. Rouhani, J. Favre, X. Crevoisier, K. Aminian, Ambulatory assessment423
of 3d ground reaction force using plantar pressure distribution, Gait and424
Posture 32 (2010) 311–316. doi:10.1016/j.gaitpost.2010.05.014.425
[14] Y. Jung, M. Jung, K. Lee, S. Koo, Ground reaction force estimation using426
an insole-type pressure mat and joint kinematics during walking, Journal427
of Biomechanics 47 (2014) 2693–2699. doi:10.1016/j.jbiomech.2014.05.428
007.429
[15] J. Van Den Noort, M. Van Der Esch, M. P. Steultjens, J. Dekker, H. M.430
Schepers, P. H. Veltink, J. Harlaar, Influence of the instrumented force431
shoe on gait pattern in patients with osteoarthritis of the knee, Medical432
and Biological Engineering and Computing 49 (2011) 1381–1392. doi:10.433
1007/s11517-011-0818-z.434
17
Accepted manuscript
[16] C. Liedtke, S. A. W. Fokkenrood, J. T. Menger, H. van der Kooij, P. H.435
Veltink, Evaluation of instrumented shoes for ambulatory assessment of436
ground reaction forces, Gait and Posture 26 (2007) 39–47. doi:10.1016/j.437
gaitpost.2006.07.017.438
[17] D. C. Low, S. J. Dixon, Footscan pressure insoles: Accuracy and reliability439
of force and pressure measurements in running, Gait and Posture 32 (2010)440
664–666. doi:10.1016/j.gaitpost.2010.08.002.441
[18] M. L. Audu, R. F. Kirsch, R. J. Triolo, Experimental verification of a com-442
putational technique for determining ground reactions in human bipedal443
stance, Journal of Biomechanics 40 (2007) 1115–1124. doi:10.1016/j.444
jbiomech.2006.04.016.445
[19] L. Ren, R. K. Jones, D. Howard, Whole body inverse dynamics over a446
complete gait cycle based only on measured kinematics, Journal of Biome-447
chanics 41 (2008) 2750–2759. doi:10.1016/j.jbiomech.2008.06.001.448
[20] S. E. Oh, A. Choi, J. H. Mun, Prediction of ground reaction forces during449
gait based on kinematics and a neural network model, Journal of Biome-450
chanics 46 (2013) 2372–2380. doi:10.1016/j.jbiomech.2013.07.036.451
[21] A. Choi, J.-M. Lee, J. H. Mun, Ground reaction forces predicted by us-452
ing artificial neural network during asymmetric movements, International453
Journal of Precision Engineering and Manufacturing 14 (2013) 475–483.454
doi:10.1007/s12541-013-0064-4.455
[22] R. Fluit, M. S. Andersen, S. Kolk, N. Verdonschot, H. F. J. M. Koopman,456
Prediction of ground reaction forces and moments during various activities457
of daily living, Journal of Biomechanics 47 (2014) 2321–2329. doi:10.1016/458
j.jbiomech.2014.04.030.459
[23] S. Skals, M. Jung, M. Damsgaard, M. S. Andersen, Prediction of ground460
reactions forces and moments during sports-related movements, Multibody461
Syst. Dyn (2016).462
18
Accepted manuscript
[24] H. J. Luinge, P. H. Veltink, Measuring orientation of human body segments463
using miniature gyroscopes and accelerometers, Medical and Biological464
Engineering and Computing 43 (2005) 273–282. doi:10.1007/BF02345966.465
[25] D. Roetenberg, H. J. Luinge, C. T. M. Baten, P. H. Veltink, Compen-466
sation of magnetic disturbances improves inertial and magnetic sensing of467
human body segment orientation, IEEE Transactions on Neural Systems468
and Rehabilitation Engineering 13 (2005) 395–405. doi:10.1109/TNSRE.469
2005.847353.470
[26] D. Roetenberg, H. Luinge, P. Slycke, M. Xsens, Full 6DOF Human Motion471
Tracking Using Miniature Inertial Sensors (2009).472
[27] J.-T. Zhang, A. C. Novak, B. Brouwer, Q. Li, Concurrent validation of473
xsens mvn measurement of lower limb joint angular kinematics, Physiolog-474
ical Measurement 34 (2013) N63–N69. doi:10.1088/0967-3334/34/8/N63.475
[28] A. Karatsidis, G. Bellusci, H. M. Schepers, M. de Zee, M. S. Andersen,476
P. H. Veltink, Estimation of ground reaction forces and moments during477
gait using only inertial motion capture, Sensors (Switzerland) 17 (2017).478
doi:10.3390/s17010075.479
[29] B. H. W. Koning, M. M. van der Krogt, C. T. M. Baten, B. F. J. M.480
Koopman, Driving a musculoskeletal model with inertial and magnetic481
measurement units, Computer Methods in Biomechanics and Biomedical482
Engineering 18 (2015) 1003–1013. doi:10.1080/10255842.2013.867481.483
[30] Xsens, Mvn user manual by xsens mvn - issuu, http://issuu.com/484
xsensmvn/docs/mvn_user_manual_71c37181653db5?e=14522406/485
12478179, 2017. (Accessed on 02/03/2018).486
[31] S. Skals, K. Rasmussen, K. Bendtsen, J. Yang, M. Andersen, A muscu-487
loskeletal model driven by dual microsoft kinect sensor data, Multibody488
System Dynamics (2017). doi:10.1007/s11044-017-9573-8.489
19
Accepted manuscript
[32] M. de Zee, L. Hansen, C. Wong, J. Rasmussen, E. B. Simonsen, A generic490
detailed rigid-body lumbar spine model, Journal of Biomechanics 40 (2007)491
1219–1227. doi:10.1016/j.jbiomech.2006.05.030.492
[33] M. D. Klein Horsman, H. F. J. M. Koopman, F. C. T. van der Helm,493
L. P. Pros, H. E. J. Veeger, Morphological muscle and joint parameters for494
musculoskeletal modelling of the lower extremity, Clinical Biomechanics 22495
(2007) 239–247. doi:10.1016/j.clinbiomech.2006.10.003.496
[34] H. E. J. Veeger, F. C. T. Van Der Helm, L. H. V. Van Der Woude, G. M.497
Pronk, R. H. Rozendal, Inertia and muscle contraction parameters for mus-498
culoskeletal modelling of the shoulder mechanism, Journal of Biomechanics499
24 (1991) 615–629. doi:10.1016/0021-9290(91)90294-W.500
[35] H. E. J. Veeger, B. Yu, K.-N. An, R. H. Rozendal, Parameters for modeling501
the upper extremity, Journal of Biomechanics 30 (1997) 647–652. doi:10.502
1016/S0021-9290(97)00011-0.503
[36] F. C. T. Van der Helm, H. E. J. Veeger, G. M. Pronk, L. H. V. Van der504
Woude, R. H. Rozendal, Geometry parameters for musculoskeletal mod-505
elling of the shoulder system, Journal of Biomechanics 25 (1992) 129–144.506
doi:10.1016/0021-9290(92)90270-B.507
[37] M. S. Andersen, M. Damsgaard, B. MacWilliams, J. Rasmussen, A compu-508
tationally efficient optimisation-based method for parameter identification509
of kinematically determinate and over-determinate biomechanical systems,510
Computer Methods in Biomechanics and Biomedical Engineering 13 (2010)511
171–183. doi:10.1080/10255840903067080.512
[38] M. S. Andersen, M. Damsgaard, J. Rasmussen, Kinematic analysis of513
over-determinate biomechanical systems, Computer Methods in Biome-514
chanics and Biomedical Engineering 12 (2009) 371–384. doi:10.1080/515
10255840802459412.516
20
Accepted manuscript
[39] M. A. Marra, V. Vanheule, R. Fluit, B. H. F. J. M. Koopman, J. Ras-517
mussen, N. Verdonschot, M. S. Andersen, A subject-specific musculoskele-518
tal modeling framework to predict in vivo mechanics of total knee arthro-519
plasty, Journal of Biomechanical Engineering 137 (2015). doi:10.1115/1.520
4029258.521
[40] D. C. Frankenfield, W. A. Rowe, R. N. Cooney, J. S. Smith, D. Becker,522
Limits of body mass index to detect obesity and predict body composition,523
Nutrition 17 (2001) 26–30. doi:10.1016/S0899-9007(00)00471-8.524
[41] M. A. Sprague, T. L. Geers, Spectral elements and field separation for an525
acoustic fluid subject to cavitation, Journal of Computational Physics 184526
(2003) 149–162. doi:10.1016/S0021-9991(02)00024-4.527
[42] N. C. Silver, W. P. Dunlap, Averaging correlation coefficients: should528
fisher’s z transformation be used?, Journal of Applied Psychology 72 (1987)529
146.530
[43] R. Taylor, Interpretation of the correlation coefficient: A basic review,531
Journal of Diagnostic Medical Sonography 6 (1990) 35–39. doi:10.1177/532
875647939000600106.533
[44] L. Chiari, U. D. Croce, A. Leardini, A. Cappozzo, L. Chiari, U. D. Croce,534
A. Cappozzo, U. Della Croce, A. Leardini, L. Chiari, Human movement535
analysis using stereophotogrammetry, Gait Posture 21 (2005) 226–237.536
[45] A. Leardini, A. Chiari, U. Della Croce, A. Cappozzo, Human movement537
analysis using stereophotogrammetry part 3. soft tissue artifact assessment538
and compensation, Gait and Posture 21 (2005) 212–225. doi:10.1016/j.539
gaitpost.2004.05.002.540
[46] M. Eltoukhy, C. Kuenze, M. S. Andersen, J. Oh, J. Signorile, Prediction541
of ground reaction forces for parkinsons disease patients using a kinect-542
driven musculoskeletal gait analysis model, Medical Engineering & Physics543
21
Accepted manuscript
50 (2017) 75–82. URL: https://doi.org/10.1016/j.medengphy.2017.544
10.004. doi:10.1016/j.medengphy.2017.10.004.545
22
Accepted manuscript
List of Figures546
1 Illustration of the pipeline used in the IMC-PGRF approach. A547
recording from Xsens MVN Studio (a) is exported to a BVH548
file to generate a stick figure model (b), in which virtual markers549
(blue) are placed. Virtual markers (red) are also placed on points550
of the musculoskeletal model (c), and by projecting b on c the551
kinematics of the musculoskeletal model are solved. Finally, in-552
verse dynamic analysis using prediction of ground reaction forces553
and moments is performed to estimate the kinetics. . . . . . . . . 24554
2 Ankle, knee, and hip joint angle estimates (standard deviation555
around mean) of the IMC-PGRF (orange shaded area around556
orange dotted line) and OMC-PGRF models (blue shaded area557
around blue dashed line) versus OMC-MGRF model (thin black558
solid lines around thick black solid line). . . . . . . . . . . . . . . 25559
3 Ground and lower limb joint reaction force estimates (standard560
deviation around mean) of the IMC-PGRF (orange shaded area561
around orange dotted line) and OMC-PGRF models (blue shaded562
area around blue dashed line) versus OMC-MGRF model (thin563
black solid lines around thick black solid line). . . . . . . . . . . . 26564
4 Ground reaction and lower limb net joint moment estimates (stan-565
dard deviation around mean) of the IMC-PGRF (orange shaded566
area around orange dotted line) and OMC-PGRF models (blue567
shaded area around blue dashed line) versus OMC-MGRF model568
(thin black solid lines around thick black solid line). . . . . . . . 27569
23
Accepted manuscript
Figure 1
24
Accepted manuscript
Figure 2
HS TO HS TO HS
-30
-20
-10
0
10
Angle (degrees)
Subtalar Eversion
HS TO HS TO HS
-10
0
10
20
30
Ankle Plantar Flexion
HS TO HS TO HS
-60
-40
-20
0
Knee Flexion
HS TO HS TO HS
-5
0
5
10
15
Angle (degrees)
Hip Abduction
HS TO HS TO HS
0
10
20
30
Hip External Rotation
HS TO HS TO HS
-20
0
20
40
Hip Flexion
OMC-MGRF Mean ± SD
IMC-PGRF Mean ± SD OMC-PGRF Mean ± SD
25
Accepted manuscript
HS TO HS TO
-0.2
0
0.2
Force (BW)
Ground Anteroposterior
HS TO HS TO
-0.05
0
0.05
0.1
Ground Mediolateral
HS TO HS TO
0
0.5
1
Ground Vertical
HS TO HS TO
-1
-0.5
0
Force (BW)
Ankle Anteroposterior
HS TO HS TO
-2
-1.5
-1
-0.5
0
Ankle Mediolateral
HS TO HS TO
0
2
4
6
8
Ankle Proximodistal
HS TO HS TO
0
0.5
1
1.5
Force (BW)
Knee Anteroposterior
HS TO HS TO
-1
-0.5
0
Knee Mediolateral
HS TO HS TO
0
2
4
6
Knee Proximodistal
HS TO HS TO
-0.4
-0.2
0
0.2
0.4
0.6
Force (BW)
Hip Anteroposterior
HS TO HS TO
-1.5
-1
-0.5
0
Hip Mediolateral
HS TO HS TO
0
2
4
6
Hip Proximodistal
OMC-MGRF Mean ± SD
IMC-PGRF Mean ± SD OMC-PGRF Mean ± SD
Figure 3
26
Accepted manuscript
HS TO HS TO
-0.03
-0.02
-0.01
0
0.01
0.02
Moment (BW*BH)
Ground Frontal
HS TO HS TO
-0.1
-0.05
0
Ground Sagittal
HS TO HS TO
-5
0
5
10
10-3 Ground Transverse
HS TO HS TO
0
0.01
0.02
Moment (BW*BH)
Subtalar Eversion
HS TO HS TO
0
0.05
0.1
Ankle Plantar Flexion
HS TO HS TO
-0.01
0
0.01
Ankle Axial
HS TO HS TO
-0.06
-0.04
-0.02
0
Moment (BW*BH)
Knee Abduction
HS TO HS TO
-0.04
-0.02
0
0.02
0.04
0.06
Knee Flexion
HS TO HS TO
-5
0
5
10
10-3 Knee Axial
HS TO HS TO
-0.02
0
0.02
0.04
0.06
Moment (BW*BH)
Hip Abduction
HS TO HS TO
-0.1
-0.05
0
0.05
0.1
Hip Flexion
HS TO HS TO
-0.01
-0.005
0
0.005
0.01
0.015
Hip External Rotation
OMC-MGRF Mean ± SDIMC-PGRF Mean ± SD OMC-PGRF Mean ± SD
Figure 4
27
Accepted manuscript
List of Tables570
1 Comparison of lower limb joint angles between musculoskele-571
tal model driven by the inertial (IMC-PGRF) and optical mo-572
tion capture (OMC-PGRF/OMC-MGRF), using Pearson corre-573
lation coefficient (ρ), absolute and relative root-mean-squared-574
differences (RM SD in degrees and rRMS D in %, respectively).575
Mand Pdenote the % magnitude and phase differences . . . . . 29576
2 IMC-PGRF-based ground and joint reaction forces (first three577
quantities) and net moments (second three quantities) versus578
OMC-MGRF. Pearson correlation coefficient is denoted with ρ.579
Absolute per body weight (or body weight times height) and580
relative root-mean-squared-difference are denoted with RM SD581
(%BW or %BW*BH) and rRM SD (%), respectively. Mand P582
indicate the magnitude and phase differences (%). . . . . . . . . 30583
3 OMC-PGRF-based ground and joint reaction forces (first three584
quantities) and net moments (second three quantities) versus585
OMC-MGRF. Pearson correlation coefficient is denoted with ρ.586
Absolute per body weight (or body weight times height) and587
relative root-mean-squared-difference are denoted with RM SD588
(%BW or %BW*BH) and rRM SD (%), respectively. Mand P589
indicate the magnitude and phase differences (%). . . . . . . . . 31590
28
Accepted manuscript
Table 1
ρRMSD rRMSD M P
Subtalar Eversion 0.81 9.7 (3.2) 32.6 (10.3) 24.0 (34.7) 19.3 (10.2)
Ankle Plantarflexion 0.95 4.1 (1.3) 14.0 ( 4.8) 8.6 (16.4) 9.8 ( 3.9)
Knee Flexion 0.99 4.4 (2.0) 7.2 (3.4) 0.7 (6.2) 4.8 (2.7)
Hip Abduction 0.91 4.1 (2.0) 25.9 (10.7) -12.2 (34.7) 21.2 (9.3)
Hip External Rotation 0.68 6.5 (2.8) 36.9 (15.2) 5.5 (39.0) 12.6 (6.2)
Hip Flexion 0.99 5.7 (2.1) 12.7 ( 5.3) -4.0 (13.9) 8.8 (4.2)
29
Accepted manuscript
Table 2
ρRMSD rRMSD M P
Ground
Anteroposterior 0.91 5.5 (1.2) 15.0 (2.4) -25.4 ( 7.3) 14.4 (3.2)
Mediolateral 0.80 2.1 (0.6) 18.5 (3.2) 7.3 (19.3) 15.4 (3.8)
Vertical 0.97 9.3 (3.0) 7.7 (2.1) -1.5 ( 1.5) 3.4 (1.0)
Frontal 0.64 0.9 (0.6) 38.0 (23.1) 125.5 (319.9) 30.6 (17.3)
Sagittal 0.91 1.6 (0.6) 17.5 ( 6.8) 14.3 ( 18.2) 12.1 ( 4.5)
Transverse 0.82 0.2 (0.1) 23.3 ( 7.2) -8.5 ( 41.9) 17.8 ( 5.3)
Ankle
Anteroposterior 0.84 22.2 (10.3) 26.1 (10.2) 49.0 (45.8) 10.8 (2.1)
Mediolateral 0.93 24.3 ( 8.9) 15.2 ( 5.3) 14.3 (17.1) 7.9 (2.7)
Proximodistal 0.93 88.5 (30.6) 13.6 ( 4.6) 9.8 (14.1) 7.2 (2.3)
Eversion 0.76 0.6 (0.2) 33.3 (20.2) 107.7 (220.3) 18.9 (10.7)
Plantar Flexion 0.93 1.6 (0.6) 15.1 ( 6.6) 10.6 ( 18.1) 9.9 ( 3.6)
Axial 0.67 0.5 (0.2) 30.4 (12.2) 46.5 ( 49.1) 27.2 (13.5)
Knee
Anteroposterior 0.82 30.6 (10.3) 25.8 (9.7) 43.7 (53.5) 13.0 (4.5)
Mediolateral 0.91 12.0 ( 3.5) 14.1 (3.8) 6.6 ( 8.6) 7.0 (2.0)
Proximodistal 0.90 63.1 (26.9) 14.3 (6.6) 5.1 ( 9.1) 7.2 (2.8)
Abduction 0.81 1.1 (0.4) 18.9 ( 6.8) -2.7 (16.1) 10.7 ( 3.8)
Flexion 0.58 1.9 (0.5) 29.8 ( 7.6) 17.9 (45.0) 32.8 ( 9.6)
Axial 0.73 0.3 (0.1) 25.4 (10.3) 2.3 (30.5) 27.9 (13.8)
Hip
Anteroposterior 0.71 17.6 ( 7.6) 27.2 (9.6) 6.8 (24.4) 27.6 (10.9)
Mediolateral 0.73 27.0 (12.5) 23.0 (7.4) 7.7 (14.6) 10.6 ( 4.1)
Proximodistal 0.78 102.8 (30.6) 21.7 (4.5) 20.2 (10.0) 9.0 ( 2.5)
Abduction 0.83 1.4 (0.7) 19.7 (5.8) 6.3 (16.9) 13.7 ( 7.9)
Flexion 0.92 2.2 (0.6) 19.4 (4.2) 73.2 (26.3) 14.8 ( 4.2)
External Rotation 0.50 0.5 (0.2) 31.6 (6.6) -3.9 (36.4) 25.6 (10.1)
30
Accepted manuscript
Table 3
ρRMSD rRMSD M P
Ground
Anteroposterior 0.96 3.7 (1.1) 8.3 (2.0) 7.7 (12.0) 8.8 (1.8)
Mediolateral 0.79 1.9 (0.5) 18.6 (4.1) 2.4 (10.8) 15.2 (4.9)
Vertical 0.99 5.9 (1.9) 4.9 (1.4) -1.2 ( 1.1) 2.1 (0.7)
Frontal 0.66 0.7 (0.2) 30.3 (9.3) 71.0 (122.2) 24.5 (9.1)
Sagittal 0.94 1.2 (0.4) 13.1 (3.8) 15.9 ( 15.3) 9.2 (3.2)
Transverse 0.81 0.2 (0.1) 20.7 (7.5) 7.1 ( 22.9) 17.5 (7.5)
Ankle
Anteroposterior 0.83 18.9 ( 6.9) 23.0 (6.1) 37.3 (28.6) 10.8 (2.3)
Mediolateral 0.96 16.1 ( 4.2) 10.7 (2.6) 6.8 ( 9.6) 5.8 (2.1)
Proximodistal 0.96 62.2 (17.6) 9.8 (2.7) 7.1 ( 9.0) 5.2 (1.8)
Eversion 0.76 0.5 (0.1) 25.5 (7.0) 45.3 (64.1) 18.7 (10.2)
Plantar Flexion 0.96 1.0 (0.3) 10.1 (3.3) 5.9 (10.0) 7.0 ( 2.6)
Axial 0.64 0.5 (0.1) 27.2 (7.3) 33.3 (36.9) 27.5 (11.5)
Knee
Anteroposterior 0.93 11.9 ( 4.5) 12.3 (4.3) -7.3 (8.7) 7.4 (2.0)
Mediolateral 0.96 7.2 ( 2.0) 8.8 (2.6) -4.2 (5.6) 4.4 (1.0)
Proximodistal 0.95 41.7 (12.0) 9.3 (2.6) -2.7 (5.8) 4.9 (1.2)
Abduction 0.91 0.8 (0.2) 12.6 (2.6) -0.1 (10.5) 7.7 (1.6)
Flexion 0.86 0.9 (0.3) 16.7 (4.8) -1.7 (14.3) 16.9 (5.2)
Axial 0.82 0.2 (0.1) 18.5 (6.6) -3.4 (17.7) 20.6 (8.0)
Hip
Anteroposterior 0.89 9.9 ( 3.6) 16.0 (6.7) -10.4 (10.6) 16.6 (7.6)
Mediolateral 0.92 14.7 ( 4.0) 12.7 (3.1) -1.9 ( 6.9) 6.2 (1.5)
Proximodistal 0.92 50.0 (15.9) 11.5 (2.6) -4.6 ( 6.1) 5.5 (1.2)
Abduction 0.91 0.8 (0.2) 13.3 (2.6) -3.2 ( 6.3) 8.7 (2.4)
Flexion 0.86 1.3 (0.4) 16.4 (3.4) -9.3 (12.3) 18.0 (4.1)
External Rotation 0.68 0.3 (0.1) 22.5 (3.7) 6.5 (15.8) 18.8 (4.8)
31
Accepted manuscript
Supplementary material
Title:
Musculoskeletal model-based inverse dynamic analysis under ambulatory conditions using inertial motion
capture
Authors:
Angelos Karatsidis1,2, Moonki Jung3, H. Martin Schepers1, Giovanni Bellusci1, Mark de Zee4, Peter H.
Veltink2, Michael Skipper Andersen5
Affiliations:
1Xsens Technologies B.V., Enschede 7521 PR, The Netherlands
2Institute for Biomedical Technology and Technical Medicine (MIRA), University of Twente, Enschede
7500 AE, The Netherlands
3AnyBody Technology A/S, Aalborg 9220, Denmark
4Department of Health Science and Technology, Aalborg University, Aalborg 9220, Denmark
5Department of Materials and Production, Aalborg University, Aalborg 9220, Denmark
Corresponding author:
Angelos Karatsidis, Address: Xsens Technologies B.V., Enschede 7521 PR, The Netherlands, E-mail:
angelos.karatsidis@xsens.com, Tel.: +31 88 97367 36, Fax: +31 88 97367 01
Keywords:
musculoskeletal modeling, inertial motion capture, inverse dynamics, ground reaction forces and mo-
ments, gait analysis
1
Accepted manuscript
1. Virtual marker placement
Table 1: Desription of the placement of virtual markers (VM) on the segments of the Xsens MVN model (stick figure model)
and the musculoskeletal model constructed based on the AnyBody Managed Model Repository (AMMR).
VM Name VM Placement on MVN VM Placement on AMMR VM Weight
T1C7 jT1C7 T1/C7 Joint (1,1,1)
SPNE jT9T8 Inferior to T1/C7 Joint (1,1,1)
CHST Anterior to jT9T8 Inferior and Anterior to T1/C7 Joint (1,1,1)
SACR jL5S1 Anterior to Pelvis/Sacrum Joint (10,0,0)
RHJC jRightHip Right Hip Joint (10,10,10)
RKJC jRightKnee Right Knee Joint (2,2,2)
RKJL Lateral to jRightKnee Lateral to Right Knee Joint (1,0,0)
RAJC jRightAnkle Right Ankle Joint (1,1,1)
RTOE jRightBallFoot Right Big Toe Node (1,1,1)
RTOL Lateral to jRightBallFoot Lateral to Right Big Toe Node (0,1,0)
RSJC jRightShoulder Right Glenohumeral Joint (0,2,2)
REJC jRightElbow Elbow Joint (2,2,2)
RELA Lateral to jRightElbow Lateral to Elbow Joint (1,1,1)
RWJC jRightWrist Right Wrist Joint (2,2,2)
RHT1 Inferior and Medial to jRightWrist Inferior and Medial to Right Wrist Joint (0.5,0.5,0.5)
RHT2 Inferior and Lateral to jRightWrist Inferior and Lateral to Right Wrist Joint (0.5,0.5,0.5)
LHJC jLeftHip Left Hip Joint (10,10,10)
LKJC jLeftKnee Left Knee Joint (2,2,2)
LKJL Lateral to jLeftKnee Lateral to Left Knee Joint (1,0,0)
LAJC jLeftAnkle Left Ankle Joint (1,1,1)
LTOE jLeftBallFoot Left Big Toe Node (1,1,1)
LTOL Lateral to jLeftBallFoot Lateral to Left Big Toe Node (0,1,0)
LSJC jLeftShoulder Left Glenohumeral Joint (0,2,2)
LEJC jLeftElbow Elbow Joint (2,2,2)
LELA Lateral to jLeftElbow Lateral to Elbow Joint (1,1,1)
LWJC jLeftWrist Left Wrist Joint (2,2,2)
LHT1 Inferior and Medial to jLeftWrist Inferior and Medial to Left Wrist Joint (0.5,0.5,0.5)
LHT2 Inferior and Lateral to jLeftWrist Inferior and Lateral to Left Wrist Joint (0.5,0.5,0.5)
2
Accepted manuscript
(a) Xsens MVN (stick-figure) model (b) AnyBody Musculoskeletal Model (AMMR)
Figure 1: Illustration of the placement of the virtual markers (VM) on the segments of the Xsens MVN model (stick figure
model) and the musculoskeletal model constructed based on the AnyBody Managed Model Repository (AMMR).
3
Accepted manuscript
2. Accuracy analysis per walking speed
2.1. Comfortable walking speed
Table 2: Comfortable walking speed; comparison of lower limb joint angles between musculoskeletal model driven by the inertial
(IMC-PGRF) and optical motion capture (OMC-PGRF/OMC-MGRF), using Pearson correlation coefficient (ρ), absolute and
relative root-mean-squared-differences (RMSD in degrees and rRM SD in %, respectively). Mand Pdenote the % magnitude
and phase differences .
Normal Walking
ρRMSD rRMSD M P
Subtalar Eversion 0.79 9.7 (3.1) 32.6 (10.1) 25.5 (36.2) 18.9 (9.6)
Ankle Plantarflexion 0.95 4.0 (1.3) 13.1 ( 4.9) 10.3 (16.6) 9.3 (3.6)
Knee Flexion 0.98 4.6 (2.0) 7.4 (3.1) 2.1 (5.5) 4.9 (2.3)
Hip Abduction 0.91 3.9 (1.9) 25.2 ( 9.1) -15.9 (28.8) 20.4 (8.0)
Hip External Rotation 0.66 6.5 (2.6) 35.7 (14.1) 7.9 (36.7) 12.6 (5.5)
Hip Flexion 0.99 5.6 (2.2) 12.5 ( 5.5) -3.7 (13.0) 8.8 (4.4)
Figure 2: Comfortable walking speed; ankle, knee, and hip joint angle estimates (standard deviation around mean) of the
IMC-PGRF (orange shaded area around orange dotted line) and OMC-PGRF models (blue shaded area around blue dashed
line) versus OMC-MGRF model (thin black solid lines around thick black solid line).
HS TO HS TO HS
-30
-20
-10
0
10
Angle (degrees)
Subtalar Eversion
HS TO HS TO HS
-10
0
10
20
30
Ankle Plantar Flexion
HS TO HS TO HS
-60
-40
-20
0
Knee Flexion
HS TO HS TO HS
-5
0
5
10
15
Angle (degrees)
Hip Abduction
HS TO HS TO HS
0
10
20
30
Hip External Rotation
HS TO HS TO HS
-20
0
20
40
Hip Flexion
OMC-MGRF Mean ± SDIMC-PGRF Mean ± SD OMC-PGRF Mean ± SD
4
Accepted manuscript
Table 3: Comfortable walking speed; IMC-PGRF-based ground and joint reaction forces (first three quantities) and net moments
(second three quantities) versus OMC-MGRF. Pearson correlation coefficient is denoted with ρ. Absolute per body weight
(or body weight times height) and relative root-mean-squared-difference are denoted with RMS D (%BW or %BW*BH) and
rRM SD (%), respectively. Mand Pindicate the magnitude and phase differences (%).
ρRMSD rRMSD M P
Ground
Anteroposterior 0.91 5.5 (1.1) 14.6 (2.4) -24.8 ( 6.6) 14.2 (3.0)
Mediolateral 0.80 2.1 (0.5) 18.7 (2.5) 4.8 (17.4) 15.5 (3.0)
Vertical 0.97 8.6 (2.3) 7.1 (1.8) -1.4 ( 1.4) 3.1 (0.8)
Frontal 0.66 0.8 (0.5) 34.1 (15.3) 105.5 (334.8) 28.5 (13.8)
Sagittal 0.91 1.5 (0.6) 16.8 ( 7.0) 12.2 ( 17.2) 11.6 ( 4.4)
Transverse 0.83 0.2 (0.1) 22.3 ( 6.4) -11.7 ( 31.7) 17.4 ( 4.3)
Ankle
Anteroposterior 0.84 21.9 (10.3) 25.7 (10.1) 48.0 (47.2) 10.9 (2.1)
Mediolateral 0.93 23.9 ( 8.3) 14.8 ( 4.9) 12.8 (15.4) 8.0 (2.5)
Proximodistal 0.93 85.6 (27.7) 13.0 ( 4.4) 8.0 (12.6) 7.2 (2.2)
Eversion 0.75 0.6 (0.2) 31.5 (16.3) 98.7 (234.1) 18.3 ( 8.6)
Plantar Flexion 0.94 1.5 (0.6) 14.3 ( 6.0) 8.2 ( 15.3) 9.6 ( 3.5)
Axial 0.70 0.5 (0.2) 29.2 (11.6) 38.7 ( 46.6) 25.4 (12.8)
Knee
Anteroposterior 0.84 29.8 ( 9.3) 26.0 (9.8) 49.4 (50.6) 12.3 (4.7)
Mediolateral 0.93 11.4 ( 3.0) 13.5 (3.8) 8.1 ( 8.7) 6.4 (2.2)
Proximodistal 0.92 58.4 (29.2) 13.3 (6.9) 5.1 ( 8.1) 6.7 (3.1)
Abduction 0.83 1.0 (0.4) 17.6 (6.4) -2.1 (14.7) 10.3 ( 4.0)
Flexion 0.59 1.8 (0.5) 29.8 (6.8) 16.7 (38.8) 32.9 ( 8.2)
Axial 0.73 0.3 (0.2) 25.4 (9.9) 2.7 (31.6) 27.8 (13.1)
Hip
Anteroposterior 0.74 16.7 ( 7.4) 26.2 (8.8) 5.7 (17.7) 26.6 (8.8)
Mediolateral 0.75 26.1 (13.4) 22.7 (7.6) 7.1 (15.4) 10.3 (4.2)
Proximodistal 0.81 99.0 (25.3) 21.6 (4.4) 21.2 (10.1) 8.5 (2.7)
Abduction 0.84 1.3 (0.7) 18.8 (5.5) 9.5 (17.9) 12.8 (8.2)
Flexion 0.92 2.2 (0.6) 18.9 (3.5) 69.5 (20.9) 14.6 (4.6)
External Rotation 0.47 0.5 (0.2) 30.9 (6.9) -6.7 (32.6) 25.6 (9.9)
5
Accepted manuscript
Table 4: Comfortable walking speed; OMC-PGRF-based ground and joint reaction forces (first three quantities) and net
moments (second three quantities) versus OMC-MGRF. Pearson correlation coefficient is denoted with ρ. Absolute per body
weight (or body weight times height) and relative root-mean-squared-difference are denoted with RM SD (%BW or %BW*BH)
and rRM SD (%), respectively. Mand Pindicate the magnitude and phase differences (%).
ρRMSD rRMSD M P
Ground
Anteroposterior 0.96 3.6 (1.0) 8.2 (1.8) 5.8 ( 8.5) 8.9 (1.8)
Mediolateral 0.77 1.9 (0.4) 19.1 (3.7) 1.6 (10.8) 15.4 (4.1)
Vertical 0.99 5.7 (1.2) 4.8 (1.0) -1.2 ( 0.9) 2.1 (0.5)
Frontal 0.66 0.7 (0.2) 30.9 (9.4) 68.9 (139.9) 24.7 (8.2)
Sagittal 0.94 1.1 (0.2) 12.0 (2.6) 15.8 ( 11.3) 8.4 (2.0)
Transverse 0.82 0.2 (0.1) 19.8 (7.2) 3.8 ( 21.4) 16.6 (6.4)
Ankle
Anteroposterior 0.85 18.8 ( 6.5) 23.2 (6.6) 40.1 (29.7) 10.3 (2.1)
Mediolateral 0.96 15.0 ( 2.9) 9.8 (1.9) 6.7 ( 7.5) 5.5 (1.6)
Proximodistal 0.97 57.2 (12.6) 8.8 (2.0) 7.0 ( 6.7) 4.8 (1.3)
Eversion 0.75 0.5 (0.1) 25.5 (6.8) 45.4 (75.0) 18.0 ( 8.7)
Plantar Flexion 0.97 0.9 (0.2) 8.9 (2.0) 5.7 ( 7.3) 6.3 ( 1.7)
Axial 0.63 0.5 (0.1) 27.3 (6.5) 29.9 (35.3) 26.6 (10.1)
Knee
Anteroposterior 0.93 11.0 (4.3) 11.2 (2.7) -6.9 (5.4) 7.2 (1.4)
Mediolateral 0.97 6.5 (1.6) 8.0 (1.8) -4.2 (3.8) 4.1 (0.7)
Proximodistal 0.96 37.7 (7.8) 8.4 (1.7) -2.7 (4.0) 4.6 (0.9)
Abduction 0.91 0.7 (0.1) 11.9 (2.3) -0.3 ( 8.1) 7.8 (1.7)
Flexion 0.86 0.9 (0.2) 16.7 (4.5) -1.2 (12.9) 17.2 (5.2)
Axial 0.81 0.2 (0.1) 18.7 (6.2) -6.5 (17.2) 20.6 (6.9)
Hip
Anteroposterior 0.89 9.6 ( 3.0) 15.6 (6.2) -11.8 (9.3) 16.8 (7.0)
Mediolateral 0.91 14.7 ( 3.2) 12.6 (2.2) -2.2 (7.0) 6.4 (1.0)
Proximodistal 0.92 47.5 (13.1) 11.2 (2.2) -4.9 (4.9) 5.4 (1.0)
Abduction 0.90 0.8 (0.1) 13.3 (2.4) -3.6 ( 5.7) 8.9 (2.0)
Flexion 0.86 1.3 (0.3) 16.0 (2.6) -9.0 (11.8) 17.6 (3.4)
External Rotation 0.67 0.3 (0.1) 22.7 (3.5) 7.0 (15.7) 18.8 (4.4)
6
Accepted manuscript
HS TO HS TO
-0.2
-0.1
0
0.1
0.2
0.3
Force (BW)
Ground Anteroposterior
HS TO HS TO
-0.05
0
0.05
0.1
Ground Mediolateral
HS TO HS TO
0
0.5
1
Ground Vertical
HS TO HS TO
-1
-0.5
0
Force (BW)
Ankle Anteroposterior
HS TO HS TO
-2
-1.5
-1
-0.5
0
Ankle Mediolateral
HS TO HS TO
0
2
4
6
8
Ankle Proximodistal
HS TO HS TO
0
0.5
1
1.5
Force (BW)
Knee Anteroposterior
HS TO HS TO
-1
-0.8
-0.6
-0.4
-0.2
0
Knee Mediolateral
HS TO HS TO
0
1
2
3
4
5
Knee Proximodistal
HS TO HS TO
-0.4
-0.2
0
0.2
0.4
0.6
Force (BW)
Hip Anteroposterior
HS TO HS TO
-1.5
-1
-0.5
0
Hip Mediolateral
HS TO HS TO
0
2
4
6
Hip Proximodistal
OMC-MGRF Mean ± SD
IMC-PGRF Mean ± SD OMC-PGRF Mean ± SD
Figure 3: Comfortable walking speed; ground and lower limb joint reaction force estimates (standard deviation around mean)
of the IMC-PGRF (orange shaded area around orange dotted line) and OMC-PGRF models (blue shaded area around blue
dashed line) versus OMC-MGRF model (thin black solid lines around thick black solid line).
7
Accepted manuscript
HS TO HS TO
-0.03
-0.02
-0.01
0
0.01
0.02
Moment (BW*BH)
Ground Frontal
HS TO HS TO
-0.1
-0.05
0
Ground Sagittal
HS TO HS TO
-5
0
5
10
10-3 Ground Transverse
HS TO HS TO
0
0.01
0.02
Moment (BW*BH)
Subtalar Eversion
HS TO HS TO
0
0.05
0.1
Ankle Plantar Flexion
HS TO HS TO
-0.015
-0.01
-0.005
0
0.005
0.01
Ankle Axial
HS TO HS TO
-0.06
-0.04
-0.02
0
Moment (BW*BH)
Knee Abduction
HS TO HS TO
-0.04
-0.02
0
0.02
0.04
Knee Flexion
HS TO HS TO
-5
0
5
10
10-3 Knee Axial
HS TO HS TO
-0.02
0
0.02
0.04
0.06
Moment (BW*BH)
Hip Abduction
HS TO HS TO
-0.1
-0.05
0
0.05
0.1
Hip Flexion
HS TO HS TO
-0.01
-0.005
0
0.005
0.01
0.015
Hip External Rotation
OMC-MGRF Mean ± SD
IMC-PGRF Mean ± SD OMC-PGRF Mean ± SD
Figure 4: Comfortable walking speed; ground reaction and lower limb net joint moment estimates (standard deviation around
mean) of the IMC-PGRF (orange shaded area around orange dotted line) and OMC-PGRF models (blue shaded area around
blue dashed line) versus OMC-MGRF model (thin black solid lines around thick black solid line).
8
Accepted manuscript
2.2. Slow walking speed
Table 5: Slow walking speed; comparison of lower limb joint angles between musculoskeletal model driven by the inertial
(IMC-PGRF) and optical motion capture (OMC-PGRF/OMC-MGRF), using Pearson correlation coefficient (ρ), absolute and
relative root-mean-squared-differences (RMSD in degrees and rRM SD in %, respectively). Mand Pdenote the % magnitude
and phase differences .
Slow Walking
ρRMSD rRMSD M P
Corr RMSE rRMSE M P
Subtalar Eversion 0.81 10.1 (3.5) 32.9 (9.6) 29.5 (36.3) 17.6 (10.1)
Ankle Plantarflexion 0.96 3.9 (1.2) 13.7 (4.0) 5.1 (14.0) 9.5 ( 3.4)
Knee Flexion 0.99 4.1 (2.4) 7.0 (4.3) -0.3 (7.5) 4.7 (3.6)
Hip Abduction 0.91 4.1 (2.1) 27.6 (12.4) -3.3 (42.8) 23.2 (10.9)
Hip External Rotation 0.76 6.7 (3.1) 39.5 (17.9) 12.9 (46.9) 13.0 ( 6.7)
Hip Flexion 0.99 5.2 (1.9) 13.3 ( 5.7) -3.8 (16.4) 8.6 ( 4.2)
Figure 5: Slow walking speed; ankle, knee, and hip joint angle estimates (standard deviation around mean) of the IMC-PGRF
(orange shaded area around orange dotted line) and OMC-PGRF models (blue shaded area around blue dashed line) versus
OMC-MGRF model (thin black solid lines around thick black solid line).
HS TO HS TO HS
-40
-20
0
20
Angle (degrees)
Subtalar Eversion
HS TO HS TO HS
-10
0
10
20
30
Ankle Plantar Flexion
HS TO HS TO HS
-60
-40
-20
0
Knee Flexion
HS TO HS TO HS
-5
0
5
10
15
Angle (degrees)
Hip Abduction
HS TO HS TO HS
0
10
20
30
Hip External Rotation
HS TO HS TO HS
-10
0
10
20
30
40
Hip Flexion
OMC-MGRF Mean ± SDIMC-PGRF Mean ± SD OMC-PGRF Mean ± SD
9
Accepted manuscript
Table 6: Slow walking speed; IMC-PGRF-based ground and joint reaction forces (first three quantities) and net moments
(second three quantities) versus OMC-MGRF. Pearson correlation coefficient is denoted with ρ. Absolute per body weight
(or body weight times height) and relative root-mean-squared-difference are denoted with RMS D (%BW or %BW*BH) and
rRM SD (%), respectively. Mand Pindicate the magnitude and phase differences (%).
ρRMSD rRMSD M P
Ground
Anteroposterior 0.88 4.7 (0.8) 16.3 (2.3) -26.3 ( 8.6) 15.8 (3.4)
Mediolateral 0.84 1.7 (0.5) 17.6 (3.6) 13.2 (23.1) 13.3 (3.5)
Vertical 0.97 8.1 (2.3) 7.4 (2.1) -1.5 ( 1.2) 3.0 (0.9)
Frontal 0.64 1.0 (0.8) 45.7 (32.5) 177.8 (340.9) 32.6 (21.9)
Sagittal 0.90 1.5 (0.7) 18.7 ( 8.2) 10.2 ( 20.2) 12.2 ( 4.9)
Transverse 0.81 0.2 (0.1) 23.3 ( 5.8) -0.7 ( 56.1) 17.5 ( 4.7)
Ankle
Anteroposterior 0.85 22.8 (12.2) 29.0 (12.1) 56.5 (52.1) 10.7 (2.5)
Mediolateral 0.93 24.2 (11.3) 16.0 ( 6.7) 12.9 (21.3) 7.7 (2.9)
Proximodistal 0.93 87.4 (38.0) 14.3 ( 5.6) 8.2 (18.1) 7.1 (2.3)
Eversion 0.78 0.6 (0.3) 37.6 (27.2) 140.7 (259.7) 18.7 (13.5)
Plantar Flexion 0.93 1.6 (0.8) 16.8 ( 8.5) 9.6 ( 24.0) 10.1 ( 4.1)
Axial 0.65 0.5 (0.2) 33.7 (14.9) 47.8 ( 61.5) 27.9 (15.2)
Knee
Anteroposterior 0.84 23.6 ( 6.8) 25.9 (11.3) 50.0 (65.3) 11.5 (3.0)
Mediolateral 0.89 10.5 ( 3.1) 14.7 ( 4.2) 3.8 ( 7.4) 7.3 (2.1)
Proximodistal 0.87 65.0 (29.7) 16.8 ( 7.3) 5.2 (11.7) 8.1 (3.1)
Abduction 0.74 1.2 (0.5) 23.2 ( 7.5) -5.8 (20.4) 12.0 ( 4.5)
Flexion 0.46 1.8 (0.5) 34.5 ( 7.2) 35.3 (59.5) 36.9 (10.3)
Axial 0.60 0.3 (0.1) 29.8 (10.7) 8.2 (32.3) 33.0 (14.6)
Hip
Anteroposterior 0.55 17.3 ( 7.2) 32.5 (10.3) 11.0 (30.6) 34.7 (12.2)
Mediolateral 0.66 23.8 (11.1) 25.1 ( 7.4) 5.8 (15.0) 11.1 ( 4.4)
Proximodistal 0.71 88.4 (23.7) 23.1 ( 3.8) 17.8 ( 8.4) 9.3 ( 2.4)
Abduction 0.83 1.4 (0.7) 21.0 (6.9) -1.1 (13.4) 14.0 ( 8.3)
Flexion 0.92 2.0 (0.4) 22.8 (3.5) 93.3 (23.2) 15.5 ( 4.1)
External Rotation 0.57 0.4 (0.2) 32.4 (7.0) -0.2 (40.5) 24.8 (10.4)
10
Accepted manuscript
Table 7: Slow walking speed; OMC-PGRF-based ground and joint reaction forces (first three quantities) and net moments
(second three quantities) versus OMC-MGRF. Pearson correlation coefficient is denoted with ρ. Absolute per body weight
(or body weight times height) and relative root-mean-squared-difference are denoted with RMS D (%BW or %BW*BH) and
rRM SD (%), respectively. Mand Pindicate the magnitude and phase differences (%).
ρRMSD rRMSD M P
Ground
Anteroposterior 0.97 3.2 (1.2) 8.8 (2.6) 16.5 (13.0) 8.5 (1.7)
Mediolateral 0.82 1.5 (0.3) 17.2 (2.8) 4.4 (10.9) 13.8 (4.3)
Vertical 0.99 5.1 (1.6) 4.7 (1.5) -1.4 ( 1.4) 1.8 (0.6)
Frontal 0.70 0.6 (0.2) 29.7 (10.2) 83.9 (124.3) 22.3 (9.5)
Sagittal 0.94 1.1 (0.4) 13.4 ( 5.0) 5.0 ( 12.4) 9.7 (4.2)
Transverse 0.80 0.2 (0.1) 21.2 ( 6.9) 16.2 ( 24.5) 18.1 (7.9)
Ankle
Anteroposterior 0.83 16.8 ( 6.8) 22.7 (6.2) 32.4 (31.0) 11.2 (2.8)
Mediolateral 0.96 14.4 ( 3.8) 10.6 (3.1) 1.9 (10.0) 5.6 (2.4)
Proximodistal 0.97 55.3 (16.7) 9.8 (3.5) 1.2 ( 8.9) 5.2 (2.3)
Eversion 0.79 0.4 (0.1) 24.7 (8.3) 43.2 (59.0) 18.0 (13.3)
Plantar Flexion 0.96 0.9 (0.4) 10.6 (4.4) -0.9 ( 9.2) 7.2 ( 3.2)
Axial 0.68 0.4 (0.1) 27.3 (7.3) 37.2 (42.0) 25.6 (11.9)
Knee
Anteroposterior 0.92 9.9 ( 3.5) 14.8 (5.7) -11.4 (11.5) 7.6 (2.7)
Mediolateral 0.96 7.0 ( 2.4) 9.9 (3.5) -7.3 ( 6.1) 4.6 (1.3)
Proximodistal 0.95 39.1 (14.7) 10.1 (3.6) -6.8 ( 5.6) 5.0 (1.6)
Abduction 0.90 0.7 (0.1) 14.0 (2.9) -5.3 (11.3) 7.7 (1.6)
Flexion 0.84 0.8 (0.2) 18.7 (5.4) -2.8 (18.0) 18.2 (5.8)
Axial 0.73 0.2 (0.1) 21.7 (7.0) 2.7 (16.2) 24.7 (9.2)
Hip
Anteroposterior 0.87 8.3 ( 2.7) 17.0 (7.4) -10.0 (9.0) 18.4 (9.3)
Mediolateral 0.90 12.9 ( 4.3) 13.7 (3.9) -2.3 (5.7) 6.5 (2.0)
Proximodistal 0.91 44.3 (13.9) 12.4 (3.0) -6.0 (5.9) 5.5 (1.3)
Abduction 0.93 0.7 (0.1) 12.4 (2.7) -2.4 ( 6.2) 7.1 (1.7)
Flexion 0.81 1.1 (0.3) 19.2 (2.4) -12.4 (11.7) 20.7 (3.3)
External Rotation 0.71 0.3 (0.1) 22.4 (4.1) 12.5 (16.0) 16.9 (4.2)
11
Accepted manuscript
HS TO HS TO
-0.2
-0.1
0
0.1
0.2
Force (BW)
Ground Anteroposterior
HS TO HS TO
-0.05
0
0.05
0.1 Ground Mediolateral
HS TO HS TO
0
0.5
1
Ground Vertical
HS TO HS TO
-1
-0.5
0
Force (BW)
Ankle Anteroposterior
HS TO HS TO
-1.5
-1
-0.5
0
Ankle Mediolateral
HS TO HS TO
0
2
4
6
8Ankle Proximodistal
HS TO HS TO
0
0.5
1
1.5
Force (BW)
Knee Anteroposterior
HS TO HS TO
-0.8
-0.6
-0.4
-0.2
0
Knee Mediolateral
HS TO HS TO
0
1
2
3
4
Knee Proximodistal
HS TO HS TO
-0.4
-0.2
0
0.2
0.4
Force (BW)
Hip Anteroposterior
HS TO HS TO
-1
-0.5
0
Hip Mediolateral
HS TO HS TO
0
1
2
3
4
5
Hip Proximodistal
OMC-MGRF Mean ± SD
IMC-PGRF Mean ± SD OMC-PGRF Mean ± SD
Figure 6: Slow walking speed; ground and lower limb joint reaction force estimates (standard deviation around mean) of the
IMC-PGRF (orange shaded area around orange dotted line) and OMC-PGRF models (blue shaded area around blue dashed
line) versus OMC-MGRF model (thin black solid lines around thick black solid line).
12
Accepted manuscript
HS TO HS TO
-0.03
-0.02
-0.01
0
0.01
Moment (BW*BH)
Ground Frontal
HS TO HS TO
-0.08
-0.06
-0.04
-0.02
0
0.02
Ground Sagittal
HS TO HS TO
0
5
10
10-3 Ground Transverse
HS TO HS TO
0
0.01
0.02
Moment (BW*BH)
Subtalar Eversion
HS TO HS TO
0
0.05
0.1
Ankle Plantar Flexion
HS TO HS TO
-15
-10
-5
0
5
10-3 Ankle Axial
HS TO HS TO
-0.06
-0.04
-0.02
0
Moment (BW*BH)
Knee Abduction
HS TO HS TO
-0.04
-0.02
0
0.02
0.04
Knee Flexion
HS TO HS TO
-5
0
5
10-3 Knee Axial
HS TO HS TO
-0.04
-0.02
0
0.02
0.04
0.06
Moment (BW*BH)
Hip Abduction
HS TO HS TO
-0.05
0
0.05
Hip Flexion
HS TO HS TO
-0.01
-0.005
0
0.005
0.01
0.015
Hip External Rotation
OMC-MGRF Mean ± SD
IMC-PGRF Mean ± SD OMC-PGRF Mean ± SD
Figure 7: Slow walking speed; ground reaction and lower limb net joint moment estimates (standard deviation around mean)
of the IMC-PGRF (orange shaded area around orange dotted line) and OMC-PGRF models (blue shaded area around blue
dashed line) versus OMC-MGRF model (thin black solid lines around thick black solid line).
13
Accepted manuscript
2.3. Fast walking speed
Table 8: Fast walking speed; comparison of lower limb joint angles between musculoskeletal model driven by the inertial
(IMC-PGRF) and optical motion capture (OMC-PGRF/OMC-MGRF), using Pearson correlation coefficient (ρ), absolute and
relative root-mean-squared-differences (RMSD in degrees and rRM SD in %, respectively). Mand Pdenote the % magnitude
and phase differences .
Fast Walking
ρRMSD rRMSD M P
Subtalar Eversion 0.83 9.3 (2.8) 32.2 (11.4) 16.1 (29.9) 21.6 (10.9)
Ankle Plantarflexion 0.95 4.6 (1.3) 15.4 ( 5.1) 10.4 (18.3) 10.8 ( 4.7)
Knee Flexion 0.98 4.6 (1.6) 7.3 (2.6) -0.0 (5.0) 4.6 (1.9)
Hip Abduction 0.9 4.2 (2.0) 25.0 (10.3) -17.6 (29.6) 20.0 (8.4)
Hip External Rotation 0.62 6.2 (2.6) 35.3 (12.8) -5.4 (29.1) 12.0 (6.4)
Hip Flexion 0.99 6.3 (1.9) 12.5 ( 4.4) -4.6 (12.0) 9.1 (3.8)
Figure 8: Fast walking speed; ankle, knee, and hip joint angle estimates (standard deviation around mean) of the IMC-PGRF
(orange shaded area around orange dotted line) and OMC-PGRF models (blue shaded area around blue dashed line) versus
OMC-MGRF model (thin black solid lines around thick black solid line).
HS TO HS TO HS
-30
-20
-10
0
10
Angle (degrees)
Subtalar Eversion
HS TO HS TO HS
-20
-10
0
10
20
30
Ankle Plantar Flexion
HS TO HS TO HS
-60
-40
-20
0
Knee Flexion
HS TO HS TO HS
-10
0
10
20
Angle (degrees)
Hip Abduction
HS TO HS TO HS
0
10
20
30
Hip External Rotation
HS TO HS TO HS
-20
0
20
40
Hip Flexion
OMC-MGRF Mean ± SDIMC-PGRF Mean ± SD OMC-PGRF Mean ± SD
14
Accepted manuscript
Table 9: Fast walking speed; IMC-PGRF-based ground and joint reaction forces (first three quantities) and net moments
(second three quantities) versus OMC-MGRF. Pearson correlation coefficient is denoted with ρ. Absolute per body weight
(or body weight times height) and relative root-mean-squared-difference are denoted with RMS D (%BW or %BW*BH) and
rRM SD (%), respectively. Mand Pindicate the magnitude and phase differences (%).
ρRMSD rRMSD M P
Ground
Anteroposterior 0.92 6.5 (1.2) 14.1 (2.0) -25.2 ( 6.6) 13.0 (2.7)
Mediolateral 0.75 2.5 (0.7) 19.3 (3.4) 3.9 (15.2) 17.5 (3.8)
Vertical 0.95 11.5 (3.2) 8.8 (2.2) -1.7 ( 1.8) 4.1 (1.1)
Frontal 0.61 0.9 (0.5) 34.1 (15.5) 91.7 (269.9) 30.9 (15.0)
Sagittal 0.90 1.7 (0.4) 17.1 ( 4.2) 21.6 ( 15.2) 12.7 ( 4.1)
Transverse 0.81 0.2 (0.1) 24.4 ( 9.3) -13.1 ( 33.0) 18.4 ( 6.8)
Ankle
Anteroposterior 0.84 22.0 ( 8.0) 23.4 (6.9) 41.8 (34.5) 10.7 (1.5)
Mediolateral 0.94 25.0 ( 6.2) 14.9 (3.7) 17.7 (13.1) 8.0 (2.9)
Proximodistal 0.93 93.3 (23.8) 13.6 (3.2) 13.8 ( 9.3) 7.5 (2.5)
Eversion 0.75 0.7 (0.2) 30.7 (13.7) 81.9 (137.1) 19.8 ( 9.8)
Plantar Flexion 0.93 1.6 (0.4) 14.1 ( 4.2) 14.8 ( 12.5) 10.1 ( 3.3)
Axial 0.65 0.6 (0.1) 28.3 ( 8.4) 54.9 ( 33.2) 28.6 (12.1)
Knee
Anteroposterior 0.75 39.4 ( 8.2) 25.3 (7.4) 29.3 (38.3) 15.7 (4.6)
Mediolateral 0.90 14.5 ( 3.0) 14.3 (3.3) 8.0 ( 8.9) 7.3 (1.5)
Proximodistal 0.92 66.8 (19.0) 12.8 (4.0) 5.0 ( 7.0) 6.7 (1.8)
Abduction 0.86 1.1 (0.2) 15.5 (3.3) 0.1 (10.9) 9.8 ( 2.1)
Flexion 0.67 2.0 (0.4) 24.6 (5.3) -0.2 (18.7) 28.2 ( 8.3)
Axial 0.83 0.3 (0.1) 20.6 (8.1) -4.9 (25.8) 22.2 (11.5)
Hip
Anteroposterior 0.80 19.0 ( 8.3) 22.5 (6.7) 3.3 (23.4) 20.7 (5.9)
Mediolateral 0.77 31.8 (11.6) 21.1 (6.8) 10.5 (13.0) 10.6 (3.7)
Proximodistal 0.82 123.9 (32.5) 20.3 (4.9) 21.7 (11.2) 9.2 (2.4)
Abduction 0.80 1.5 (0.6) 19.5 (4.5) 10.7 (16.6) 14.4 (6.9)
Flexion 0.91 2.6 (0.6) 16.4 (3.1) 55.2 (20.0) 14.5 (3.8)
External Rotation 0.44 0.5 (0.2) 31.7 (5.6) -4.5 (36.2) 26.6 (9.9)
15
Accepted manuscript
Table 10: Fast walking speed; OMC-PGRF-based ground and joint reaction forces (first three quantities) and net moments
(second three quantities) versus OMC-MGRF. Pearson correlation coefficient is denoted with ρ. Absolute per body weight
(or body weight times height) and relative root-mean-squared-difference are denoted with RMS D (%BW or %BW*BH) and
rRM SD (%), respectively. Mand Pindicate the magnitude and phase differences (%).
ρRMSD rRMSD M P
Ground
Anteroposterior 0.96 4.2 (0.8) 7.9 (1.5) 0.1 ( 7.4) 9.0 (1.7)
Mediolateral 0.76 2.2 (0.5) 19.4 (5.3) 1.2 (10.6) 16.5 (6.0)
Vertical 0.98 7.2 (2.3) 5.4 (1.6) -1.0 ( 0.9) 2.6 (0.8)
Frontal 0.59 0.8 (0.2) 30.1 (8.1) 59.1 (92.8) 26.9 (9.3)
Sagittal 0.94 1.5 (0.3) 14.3 (2.9) 28.3 (13.3) 9.7 (2.8)
Transverse 0.80 0.2 (0.1) 21.0 (8.5) 1.1 (19.8) 17.9 (8.3)
Ankle
Anteroposterior 0.83 21.3 ( 7.0) 22.9 (5.2) 39.4 (23.6) 11.0 (1.9)
Mediolateral 0.95 19.4 ( 4.2) 11.9 (2.3) 12.5 ( 8.6) 6.6 (2.2)
Proximodistal 0.96 76.4 (15.8) 11.0 (2.0) 13.7 ( 6.9) 5.7 (1.7)
Eversion 0.72 0.6 (0.1) 26.2 (5.6) 47.5 (55.2) 20.3 ( 7.5)
Plantar Flexion 0.96 1.3 (0.3) 10.9 (2.6) 13.8 ( 8.0) 7.7 ( 2.7)
Axial 0.60 0.5 (0.1) 27.0 (8.2) 33.3 (32.7) 30.6 (12.1)
Knee
Anteroposterior 0.94 15.2 (4.1) 10.8 (2.3) -3.0 (5.8) 7.4 (1.6)
Mediolateral 0.96 8.2 (1.6) 8.5 (1.4) -0.6 (4.7) 4.5 (0.8)
Proximodistal 0.95 49.6 (9.0) 9.4 (1.6) 1.7 (4.7) 5.3 (0.9)
Abduction 0.92 0.9 (0.2) 11.8 (1.8) 6.1 ( 9.0) 7.6 (1.5)
Flexion 0.90 1.2 (0.3) 14.6 (3.4) -1.1 (11.2) 15.2 (4.0)
Axial 0.89 0.2 (0.1) 14.8 (4.7) -6.5 (18.5) 16.1 (4.9)
Hip
Anteroposterior 0.91 11.9 ( 4.0) 15.4 (6.4) -9.1 (13.3) 14.4 (5.4)
Mediolateral 0.94 16.6 ( 3.5) 11.7 (2.4) -1.2 ( 7.9) 5.8 (1.2)
Proximodistal 0.93 59.5 (17.0) 10.8 (2.5) -2.6 ( 7.2) 5.5 (1.3)
Abduction 0.88 1.0 (0.2) 14.3 (2.5) -3.5 ( 7.2) 10.4 (2.4)
Flexion 0.89 1.6 (0.3) 13.8 (2.9) -6.1 (13.0) 15.3 (3.8)
External Rotation 0.67 0.4 (0.1) 22.2 (3.5) -0.9 (12.6) 20.7 (5.3)
16
Accepted manuscript
HS TO HS TO
-0.2
0
0.2
Force (BW)
Ground Anteroposterior
HS TO HS TO
-0.05
0
0.05
0.1
Ground Mediolateral
HS TO HS TO
0
0.5
1
1.5
Ground Vertical
HS TO HS TO
-1
-0.5
0
Force (BW)
Ankle Anteroposterior
HS TO HS TO
-2
-1.5
-1
-0.5
0
Ankle Mediolateral
HS TO HS TO
0
2
4
6
8
Ankle Proximodistal
HS TO HS TO
0
0.5
1
1.5
2
Force (BW)
Knee Anteroposterior
HS TO HS TO
-1
-0.5
0
Knee Mediolateral
HS TO HS TO
0
2
4
6
Knee Proximodistal
HS TO HS TO
-0.4
-0.2
0
0.2
0.4
0.6
0.8
Force (BW)
Hip Anteroposterior
HS TO HS TO
-1.5
-1
-0.5
0
Hip Mediolateral
HS TO HS TO
0
2
4
6
8
Hip Proximodistal
OMC-MGRF Mean ± SD
IMC-PGRF Mean ± SD OMC-PGRF Mean ± SD
Figure 9: Fast walking speed; ground and lower limb joint reaction force estimates (standard deviation around mean) of the
IMC-PGRF (orange shaded area around orange dotted line) and OMC-PGRF models (blue shaded area around blue dashed
line) versus OMC-MGRF model (thin black solid lines around thick black solid line).
17
Accepted manuscript
HS TO HS TO
-0.03
-0.02
-0.01
0
0.01
0.02
Moment (BW*BH)
Ground Frontal
HS TO HS TO
-0.1
-0.05
0
0.05
Ground Sagittal
HS TO HS TO
0
5
10
10-3 Ground Transverse
HS TO HS TO
-0.01
0
0.01
0.02
Moment (BW*BH)
Subtalar Eversion
HS TO HS TO
-0.05
0
0.05
0.1
Ankle Plantar Flexion
HS TO HS TO
-0.02
-0.01
0
0.01
0.02
Ankle Axial
HS TO HS TO
-0.08
-0.06
-0.04
-0.02
0
Moment (BW*BH)
Knee Abduction
HS TO HS TO
-0.04
-0.02
0
0.02
0.04
0.06
0.08
Knee Flexion
HS TO HS TO
-0.01
-0.005
0
0.005
0.01
0.015
Knee Axial
HS TO HS TO
-0.02
0
0.02
0.04
0.06
0.08
Moment (BW*BH)
Hip Abduction
HS TO HS TO
-0.1
-0.05
0
0.05
0.1
Hip Flexion
HS TO HS TO
-0.01
-0.005
0
0.005
0.01
0.015
Hip External Rotation
OMC-MGRF Mean ± SD
IMC-PGRF Mean ± SD OMC-PGRF Mean ± SD
Figure 10: Fast walking speed; ground reaction and lower limb net joint moment estimates (standard deviation around mean)
of the IMC-PGRF (orange shaded area around orange dotted line) and OMC-PGRF models (blue shaded area around blue
dashed line) versus OMC-MGRF model (thin black solid lines around thick black solid line).
18
Accepted manuscript
... Combined motion and force datasets have also been used on a small scale to evaluate the models that estimate moments and forces from motion. One approach uses measured motion and force data to reconstruct dynamics with offline trajectory optimization methods and biomechanically accurate models (i.e., using muscle actuators) [62,63,26,64,65]. Optimization is computationally expensive (e.g., 0.001x real time [65,26]), can only run on a few seconds of data at a time, has errors on the order of 10% of body weight compared to measured ground reaction forces [65,26] and depends on a task-specific movement objective (e.g., minimize energy, maximize speed). ...
... One approach uses measured motion and force data to reconstruct dynamics with offline trajectory optimization methods and biomechanically accurate models (i.e., using muscle actuators) [62,63,26,64,65]. Optimization is computationally expensive (e.g., 0.001x real time [65,26]), can only run on a few seconds of data at a time, has errors on the order of 10% of body weight compared to measured ground reaction forces [65,26] and depends on a task-specific movement objective (e.g., minimize energy, maximize speed). A second approach leverages motion and force data to train deep learning models to predict ground reaction forces for specific tasks, such as running [40,42] [66], and stair climbing [46]. ...
... One approach uses measured motion and force data to reconstruct dynamics with offline trajectory optimization methods and biomechanically accurate models (i.e., using muscle actuators) [62,63,26,64,65]. Optimization is computationally expensive (e.g., 0.001x real time [65,26]), can only run on a few seconds of data at a time, has errors on the order of 10% of body weight compared to measured ground reaction forces [65,26] and depends on a task-specific movement objective (e.g., minimize energy, maximize speed). A second approach leverages motion and force data to train deep learning models to predict ground reaction forces for specific tasks, such as running [40,42] [66], and stair climbing [46]. ...
Preprint
Full-text available
While reconstructing human poses in 3D from inexpensive sensors has advanced significantly in recent years, quantifying the dynamics of human motion, including the muscle-generated joint torques and external forces, remains a challenge. Prior attempts to estimate physics from reconstructed human poses have been hampered by a lack of datasets with high-quality pose and force data for a variety of movements. We present the AddBiomechanics Dataset 1.0, which includes physically accurate human dynamics of 273 human subjects, over 70 hours of motion and force plate data, totaling more than 24 million frames. To construct this dataset, novel analytical methods were required, which are also reported here. We propose a benchmark for estimating human dynamics from motion using this dataset, and present several baseline results. The AddBiomechanics Dataset is publicly available at https://addbiomechanics.org/download_data.html.
... The wearable sensor used for motion capture, as shown in Fig. 3, uses an Inertial Measurement Unit (IMU) to measure angular motion and linear acceleration of a body segment in three dimensions, employing a gyroscope and an accelerometer, respectively. The kinematic data form IMUs can be used by MSK modeling software like AMS or OpenSim [24][25][26][27] to predict joint reaction and muscle forces. Bailey et al. [25] conducted an assessment of the IMU-driven MSK model's sensitivity and validity, concluding its suitability for evaluating joint angle time series, variability magnitude, ROM, and dynamic stability. ...
... Secondly, capturing both feet and multiple steps often necessitates the use of multiple force plates, which is costly. Thirdly, incorporating force plates in motion analysis can lead to dynamic inconsistencies, resulting in residual forces and moments during ID computations [27]. ...
Article
Full-text available
Beyond qualitative assessment, gait analysis involves the quantitative evaluation of various parameters such as joint kinematics, spatiotemporal metrics, external forces, and muscle activation patterns and forces. Utilizing multibody dynamics-based musculoskeletal (MSK) modeling provides a time and cost-effective non-invasive tool for the prediction of internal joint and muscle forces. Recent advancements in the development of biofidelic MSK models have facilitated their integration into clinical decision-making processes, including quantitative diagnostics, functional assessment of prosthesis and implants, and devising data-driven gait rehabilitation protocols. Through an extensive search and meta-analysis of over 116 studies, this PRISMA-based systematic review provides a comprehensive overview of different existing multibody MSK modeling platforms, including generic templates, methods for personalization to individual subjects, and the solutions used to address statically indeterminate problems. Additionally, it summarizes post-processing techniques and the practical applications of MSK modeling tools. In the field of biomechanics, MSK modeling provides an indispensable tool for simulating and understanding human movement dynamics. However, limitations which remain elusive include the absence of MSK modeling templates based on female anatomy underscores the need for further advancements in this area.
... Inertial measurement units (IMUs) and 3D depth image motion capture methods have been increasingly adopted to collect gait kinematic data, which offer the advantage of being more portable and cost-effective [19][20][21]. Karatsidis and colleagues had developed an IMU driven musculoskeletal multibody model to obtain ground reaction force and joint contact forces [22]. While this system is portable, it should be noted that the inertial motion capture system used in their experiment could still be expensive. ...
... However, it is important to mention that significant discrepancies were observed under gait conditions [23]. Overall, the current systems often require several wired cameras, proprietary software, and specialized computing resources, hampering their widespread adoption [22,23]. ...
Article
Full-text available
The estimation of joint contact forces in musculoskeletal multibody dynamics models typically requires the use of expensive and time-consuming technologies, such as reflective marker-based motion capture (Mocap) system. In this study, we aim to propose a more accessible and cost-effective solution that utilizes the dual smartphone videos (SPV)-driven musculoskeletal multibody dynamics modeling workflow to estimate the lower limb mechanics. Twelve participants were recruited to collect marker trajectory data, force plate data, and motion videos during walking and running. The smartphone videos were initially analyzed using the OpenCap platform to identify key joint points and anatomical markers. The markers were used as inputs for the musculoskeletal multibody dynamics model to calculate the lower limb joint kinematics, joint contact forces, and ground reaction forces, which were then evaluated by the Mocap-based workflow. The root mean square error (RMSE), mean absolute deviation (MAD), and Pearson correlation coefficient (ρ) were adopted to evaluate the results. Excellent or strong Pearson correlations were observed in most lower limb joint angles (ρ = 0.74 ~ 0.94). The averaged MADs and RMSEs for the joint angles were 1.93 ~ 6.56° and 2.14 ~ 7.08°, respectively. Excellent or strong Pearson correlations were observed in most lower limb joint contact forces and ground reaction forces (ρ = 0.78 ~ 0.92). The averaged MADs and RMSEs for the joint lower limb joint contact forces were 0.18 ~ 1.07 bodyweight (BW) and 0.28 ~ 1.32 BW, respectively. Overall, the proposed smartphone video-driven musculoskeletal multibody dynamics simulation workflow demonstrated reliable accuracy in predicting lower limb mechanics and ground reaction forces, which has the potential to expedite gait dynamics analysis in a clinical setting.
... The user's body mass, stature, and other anthropometric measures (e.g., arm span, shoulder width, hip width, knee height) are used to scale body segment masses and lengths in the AMS. Ground reaction forces are predicted [22,23], and inverse dynamics analysis is used to compute trunk torques [24]. To simulate the effects of BSELOW and BSEHIGH conditions, the external torques that would be provided by respective BSE condition are subtracted from the calculated trunk torques. ...
Article
Full-text available
While occupational exoskeletons have shown the potential to control and prevent work-related musculoskeletal disorders, there is limited information available that can guide users in making informed decisions about adopting exoskeletons. Hence, we developed a virtual reality (VR) program that enables users to perform tasks required in real-life scenarios and visualize the positive and negative effects of using an exoskeleton. As a first step, we simulated a specific passive back-support exoskeleton (BSE; backX™) for assisting box lifting tasks in a virtual environment. The VR program was designed for users to customize the lifting environment (i.e., load magnitude and lifting distance, height, and angle), execute virtual lifting tasks (i.e., without actual physical loads and without wearing the BSE), and access infographics illustrating the simulated positive (i.e., decrease in spinal forces) and negative (i.e., increase in chest discomfort) effects of using the BSE for the given lifting task. An experimental study, involving 12 participants, was conducted to explore how perceived usefulness of the BSE changes when individuals experience the VR program. Participants completed four different types of virtual box lifting tasks (i.e., symmetric ankle-to-hip lift and asymmetric hip-to-hip lift; each with light and heavy loads). Perceived usefulness was higher for lifts with heavy load (vs. light load) and for ankle-to-hip lifts (vs. hip-to-hip lifts), suggesting that the infographics presented in the VR program effectively conveyed the potential biomechanical risks associated with lifting tasks as well as the extent to which wearing a BSE can alleviate these risks. Future work should investigate whether the experience in such virtual settings aligns with real-world BSE utilization, and/or contributes to influencing an individuals’ eventual understanding/acceptance of exoskeleton technologies.
... Constrained optimization via biomechanical modeling, both static and dynamic, has also been used for estimation of both kinematics and kinetics. Staticoptimization approaches rely on zero-velocity detection algorithms from joint constraints, external contacts, and additional sensors (e.g., GPS, RF-based local positioning sensors, barometers) to correct the position of the model at each step [82,85] , while dynamic optimization approaches currently require that the motion be periodic [38] , both of which limit ease of implementation and generalizability. ...
Thesis
Full-text available
Biomechanical analysis and musculoskeletal simulation techniques have been developed in laboratory settings to help us understand the body and the mechanisms of motion with enough precision to evaluate health and suggest rehabilitative treatments that aim to improve musculoskeletal function. To achieve this same result outside of a laboratory, portable sensors must be made available that can monitor the motion of the body, kinematics, the contact interactions between the body and the surrounding environment, kinetics, and the forces generated by the body that enable locomotion, muscle dynamics. While many portable sensing options have been developed in recent years, the accuracy of portable biomechanics monitoring techniques has yet to match the accuracy of traditional laboratory-based tools and remains insufficient for many rehabilitative applications. One cause for insufficient accuracy is that portable sensing techniques are often explored in isolation and unable to overcome their unique limitations. Another cause is that many possible alternative portable sensing approaches developed outside of the biomechanics field have yet to be fully investigated for their potential to serve as biomechanics monitoring tools in combination with existing portable techniques. Here, I show how developments in inertial sensing and computer vision techniques can be intelligently synthesized using either kinematics equations of motion or rigid body dynamics equations of motion to enable more accurate portable predictions of body kinematics than approaches which utilize only inertial sensors or computer vision (Chapter 2). I also show how nuance exists in the choice of fusion approach depending on the quality of inertial sensing data and computer vision estimates. A prominent trade-off exists when adding rigid body dynamics into the synthesis paradigm, and adding dynamics is helpful so long as the dynamics equations provide more accurate estimates of angular velocities and accelerations than inertial sensing data, which is likely to occur during real-world applications due to soft-tissue motion artifacts. Next, I show how capacitive sensing, a sensing technique that has been understudied in biomechanics, can be adapted for use as a customizable, comfortable, lightweight, and sensitive biomechanics monitoring wearable sensor that enables muscle-activity measurements with the fidelity of gold-standard laboratory-based techniques (Chapter 3). Capacitive sensing muscle-activity measurements can then be synthesized with inertial sensors to enable full-body kinematics, kinetics, and muscle dynamics predictions with comparable accuracy to that of marker-based motion capture. Altogether, these findings show the importance of extensively validating and incorporating new sensing approaches into biomechanics monitoring tools that seamlessly integrate with other sensors to cover their weaknesses. I suggest that future biomechanics monitoring emphasize more nuanced applications, where multiple sensing modalities are fused intelligently and optimized for specific applications to maximize monitoring accuracy and intervention efficacy in each local domain, rather than sacrificing local accuracy to reach for a one-size-fits-all solution. In the future, I envision the development of a list of locally optimized biomechanics monitoring best practices, where specific sensor combinations with precise placements and parameterized computational algorithms are tuned to maximize monitoring accuracy for use on specific clinical populations and pathologies. I believe this more nuanced approach to biomechanics monitoring will enable the development of the next generation of rehabilitative strategies to improve and sustain more widespread musculoskeletal well-being.
... A study reported a technique to predict the GRF and GRM using a biomechanical model without training data [87]. A recent finding proposed a method for prediction using a contact model with the biomechanical model [88,89]. This technique can predict the GRF and GRM acting on each foot independently using the contact model between the ground and the foot of the biomechanical model, devoid of a statistical model. ...
Article
Full-text available
Diabetes foot ulcers (DFU) have been connected to rising global mortality rates. An increased plantar tissue stress beneath the numb foot of neuropathy patients is associated with DFU. This article summarises essential therapeutic footwear concepts for the treatment of DFUs and shows how new sensor technologies have made this possible. People with neuropathic DFUs can address their foot problems with therapeutic footwear without being constrained in their daily activities. This intervention in maintaining DFUs must be discrete, power-efficient, accurate, and user-friendly in order to offer a trustworthy and acceptable way for effective health monitoring. The device can be inserted into a person's shoes from the plantar aspect of the foot to gather more information about their health. The findings showed that weight-bearing activities and sensor offloading devices can be employed as effective offloading therapies to heal DFU. This article examines the therapeutic footwear-based technological sensor system, which keeps track of many elements that are helpful in tracking foot ulcers, available treatments, and potential research areas in the future Full Text available online at: https://www.linkedin.com/posts/sir-okoduwa_in-this-study-my-research-team-and-i-conceptualize-activity-7208571588483379200-bwDR?utm_source=share&utm_medium=member_android
Article
Bone-anchored limbs (BALs) are socket prosthesis alternatives, directly fixing to residual bone via osseointegrated implant. There is a need to quantify multi-level effects of rehabilitation for transfemoral BAL users (i.e. changes in joint loading and movement patterns). Our primary objective was determining feasibility of using optimal control to predict gait biomechanics compared to ground-truth experimental data from transfemoral BAL users. A secondary objective was examining biomechanical effects from estimated changes in hip abductor muscle strength. We developed and validated a workflow for predicting gait biomechanics in four transfemoral BAL users and investigated the biomechanical effects of altered hip abductor strengths.
Article
Musculoskeletal modeling, typically implemented using marker-based systems in laboratory environments, is commonly used for noninvasive estimations of loads. Inertial measurement units (IMUs) have become an alternative for the evaluation of kinematics. However, estimates of spine joint contact forces using IMUs have yet to be thoroughly evaluated. Dynamics tasks and static postures from activities of daily living were captured on 11 healthy subjects using both systems simultaneously. Spine kinematics obtained from IMU- and marker-based systems and L4–L5 joint contact forces were compared. Lateral bending resulted in a weak agreement with significant differences between the 2 systems ( P = .02, average root mean-squared error = 4.81), whereas flexion–extension and axial rotation exhibited the highest agreement with no significant differences ( P < .05, average root mean-squared error = 5.51 and P < .31, average root mean-squared error = 5.08, respectively). All tasks showed excellent correlations ( R ² = .76–.99) in estimated loads between systems. Differences in predicted loads at the L4–L5 were only observed during flexion–extension (1041 N vs 947 N, P = .0004) and walking with weights (814 N vs 727 N, P = .004). Different joint reaction force outcomes were obtained in 2 of the 8 tasks between systems, suggesting that IMUs can be robust tools allowing for convenient and less expensive evaluations and for longitudinal assessments inside and outside the laboratory setting.
Article
Full-text available
Musculoskeletal modeling is becoming a standard method to estimate muscle, ligament and joint forces non-invasively. As input, these models often use kinematic data obtained using marker-based motion capture, which, however, is associated with several limitations, such as soft tissue artefacts and the time-consuming task of attaching markers. These issues can potentially be addressed by applying marker-less motion capture. Therefore, we developed a musculoskeletal model driven by marker-less motion capture data, based on two Microsoft Kinect Sensors and iPi Motion Capture software, which incorporated a method for predicting ground reaction forces and moments. For validation, selected model outputs (e.g. ground reaction forces, joint reaction forces, joint angles and joint range-of-motion) were compared to musculoskeletal models driven by simultaneously recorded marker-based motion capture data from 10 males performing gait and shoulder abduction with and without external load. The primary findings were that the vertical ground reaction force during gait and the shoulder abduction/adduction angles, glenohumeral joint reaction forces and deltoideus forces during both shoulder abduction tasks showed comparable results. In addition, shoulder abduction/adduction range-of-motions were not significantly different between the two systems. However, the lower extremity joint angles, moments and reaction forces showed discrepancies during gait with correlations ranging from weak to strong, and for the majority of the variables, the marker-less system showed larger standard deviations. Although discrepancies between the systems were identified, the marker-less system shows potential, especially for tracking simple upper-body movements.
Article
Full-text available
Ground reaction forces and moments (GRF&M) are important measures used as input in biomechanical analysis to estimate joint kinetics, which often are used to infer information for many musculoskeletal diseases. Their assessment is conventionally achieved using laboratory-based equipment that cannot be applied in daily life monitoring. In this study, we propose a method to predict GRF&M during walking, using exclusively kinematic information from fully-ambulatory inertial motion capture (IMC). From the equations of motion, we derive the total external forces and moments. Then, we solve the indeterminacy problem during double stance using a distribution algorithm based on a smooth transition assumption. The agreement between the IMC-predicted and reference GRF&M was categorized over normal walking speed as excellent for the vertical (ρ = 0.992, rRMSE = 5.3%), anterior (ρ = 0.965, rRMSE = 9.4%) and sagittal (ρ = 0.933, rRMSE = 12.4%) GRF&M components and as strong for the lateral (ρ = 0.862, rRMSE = 13.1%), frontal (ρ = 0.710, rRMSE = 29.6%), and transverse GRF&M (ρ = 0.826, rRMSE = 18.2%). Sensitivity analysis was performed on the effect of the cut-off frequency used in the filtering of the input kinematics, as well as the threshold velocities for the gait event detection algorithm. This study was the first to use only inertial motion capture to estimate 3D GRF&M during gait, providing comparable accuracy with optical motion capture prediction. This approach enables applications that require estimation of the kinetics during walking outside the gait laboratory.
Article
Full-text available
When performing inverse dynamic analysis (IDA) of musculoskeletal models to study human motion, inaccuracies in experimental input data and a mismatch between the model and subject lead to dynamic inconsistency. By predicting the ground reaction forces and moments (GRF&Ms) this inconsistency can be reduced and force plate measurements become unnecessary. In this study, a method for predicting GRF&Ms was validated for an array of sports-related movements. The method was applied to ten healthy subjects performing, for example, running, a side-cut manoeuvre, and vertical jump. Pearson’s correlation coefficient (\(r\)) and root-mean-square deviation were used to compare the predicted GRF&Ms and associated joint kinetics to the traditional IDA approach, where the GRF&Ms were measured using force plates. The main findings were that the method provided estimates comparable to traditional IDA across all movements for vertical GRFs (\(r\) ranging from 0.97 to 0.99, median 0.99), joint flexion moments (\(r\) ranging from 0.79 to 0.98, median 0.93), and resultant joint reaction forces (\(r\) ranging from 0.78 to 0.99, median 0.97). Considering these results, this method can be used instead of force plate measurements, hereby, facilitating IDA in sports science research and enabling complete IDA using motion analysis systems that do not incorporate force plate data.
Article
Full-text available
Kinetic analysis of walking requires joint kinematics and ground reaction force (GRF) measurement, which are typically obtained from a force plate. GRF is difficult to measure in certain cases such as slope walking, stair climbing, and track running. Nevertheless, estimating GRF continues to be of great interest for simulating human walking. The purpose of the study was to develop reaction force models placed on the sole of the foot to estimate full GRF when only joint kinematics are provided (Type-I), and to estimate ground contact shear forces when both joint kinematics and foot pressure are provided (Type-II and Type-II-val). The GRF estimation models were attached to a commercial full body skeletal model using the AnyBody Modeling System, which has an inverse dynamics-based optimization solver. The anterior-posterior shear force and medial-lateral shear force could be estimated with approximate accuracies of 6% BW and 2% BW in all three methods, respectively. Vertical force could be estimated in the Type-I model with an accuracy of 13.75% BW. The accuracy of the force estimation was highest during the mid-single-stance period with an average RMS for errors of 3.10% BW, 1.48% BW, and 7.48% BW for anterior-posterior force, medial-lateral force, and vertical force, respectively. The proposed GRF estimation models could predict full and partial GRF with high accuracy. The design of the contact elements of the proposed model should make it applicable to various activities where installation of a force measurement system is difficult, including track running and treadmill walking.
Article
Kinetic gait abnormalities result in reduced mobility among individuals with Parkinson’s disease (PD). Currently, the assessment of gait kinetics can only be achieved using costly force plates, which makes it difficult to implement in most clinical settings. The Microsoft Kinect v2 has been shown to be a feasible clinic-based alternative to more sophisticated three-dimensional motion analysis systems in producing acceptable spatiotemporal and kinematic gait parameters. In this study, we aimed to validate a Kinect- driven musculoskeletal model using the AnyBody modeling system to predict three-dimensional ground reaction forces (GRFs) during gait in patients with PD. Nine patients with PD performed over-ground walking trials as their kinematics and ground reaction forces were measured using a Kinect v2 and force plates, respectively. Kinect v2 model-based and force-plate measured peak vertical and horizontal ground reaction forces and impulses produced during the braking and propulsive phases of the gait cycle were compared. Additionally, comparison of ensemble curves and associated 90% confidence intervals (CI90) of the three-dimensional GRFs were constructed to investigate if the Kinect sensor could provide consistent and accurate GRF predictions throughout the gait cycle. Results showed that the Kinect v2 sensor has the potential to be an effective clinical assessment tool for predicting GRFs produced during gait for patients with PD. However, the observed findings should be replicated and model reliability established prior to integration into the clinical setting.
Article
This study examined the influence of force plate targeting, via stride length adjustments, on the magnitude and consistency of ground reaction force and segment angle profiles of the stance phase of human running. Seven male subjects (height, 1.77 m ± 0.081; mass, 72.4 kg ± 7.52; age range, 23 to 32 years) were asked to run at a mean velocity of 3.2 m · s-1 under three conditions (normal, short, and long strides). Four trials were completed for each condition. For each trial, the ground reaction forces were measured and the orientations of the foot, shank, and thigh computed. There were no statistically significant differences (p > .05) between the coefficients of variation of ground reaction force and segment angle profiles under the three conditions, so these profiles were produced consistently. Peak active vertical ground reaction forces, their timings, and segment angles at foot off were not significantly different across conditions. In contrast, significant differences between conditions were found for peak vertical impact forces and their timings, and for the three lower limb segment angles at the start of force plate contact. These results have implications for human gait studies, which require subjects to target the force plate. Targeting may be acceptable depending on the variables to be analyzed.
Article
Musculoskeletal (MS) models should be able to integrate the patient-specific MS architecture and undergo thorough validation prior to their introduction into the clinical practice. We present a streamlined methodology to develop subject-specific models able to simultaneously predict body-level dynamics, muscle forces, ligament forces, knee joint contact forces and secondary knee kinematics. The MS architecture of a generic cadaver-based model was scaled using an advanced morphing technique to the subject-specific morphology of a patient implanted with an instrumented total knee arthroplasty available in the fifth "Grand Challenge Competition to Predict in Vivo Knee Loads" dataset. Inverse dynamics-like analyses of a hinge-like knee model and an 11-degree-of-freedom force-dependent kinematics (FDK) knee model were simulated for one gait, one right-turn and one unloaded leg-swing trial. Predicted tibiofemoral (TF) forces and secondary knee kinematics were evaluated using experimental data available in the Grand Challenge dataset. Total TF contact forces were predicted with a root-mean-square error (RMSE) and a coefficient of determination (R^2) smaller than 0.3 BW and higher than 0.9, respectively, for both gait and right-turn trials. Secondary knee kinematics from the leg-swing trial were overall better approximated using the FDK model (average Sprague and Geers' combined error C = 0.06) than when using a hinged knee model (C = 0.34). The proposed modeling approach allows detailed subject-specific scaling and personalization, and does not contain any non-physiological parameters. This modeling framework has potential applications in aiding the clinical decision-making in orthopedics procedures, and as a tool for virtual implant design.
Article
We developed and evaluated a new kinematic driver for musculoskeletal models using ambulatory inertial and magnetic measurement units (IMMUs). The new driver uses the orientation estimates based on sensor fusion of each individual IMMU and benefits from two important properties of musculoskeletal models. First, these models contain more complex, anatomical, kinematic models than those currently used for sensor fusion of multiple IMMUs and are continuously improved. Second, they allow movement between segment and measured sensor. For three different tasks, the new IMMU driver, (optical) marker drivers and a combination of both were used to reconstruct the motion. Maximal root mean square (RMS) joint angle differences with respect to the silver standard (combined IMMU/marker drivers) were found for the hip joint; 4°, 2° and 5° during squat, gait and slideboard tasks for IMMU-driven reconstructions, compared with 6°, 5° and 5° for marker-driven reconstructions, respectively. The measured angular velocities corresponded best to the IMMU-driven reconstructions, with a maximal RMS difference of 66°/s, compared with 108°/s and 91°/s for marker-driven reconstructions and silver standard. However, large oscillations in global accelerations occurred during IMMU-driven reconstructions resulting in a maximal RMS difference with respect to measured acceleration of 23 m/s(2), compared with 9 m/s(2) for reconstructions that included marker drivers. The new driver facilitates direct implementation of IMMU-based orientation estimates in currently available biomechanical models. As such, it can help in the rapid expansion of biomechanical analysis based on outdoor measurements.
Article
During the motion analysis, joint moments are calculated by the inverse dynamics method. The ground reaction forces are representative input data for an application of the inverse dynamics technique, and extracted from an installed force platform. However, when the vast majority of human body movement, continuous and wide spaced, is considered, usage of the fixed force platform on analyzing exercises has many limitations. Therefore, the purpose of this study is to suggest a method to predict the ground reaction forces which occur in complex planes during asymmetric movements. Thirteen healthy male subjects performed static posture, gait and asymmetric movements. The experimental equipment included six infrared cameras and two force platforms, and the artificial neural network was used to solve indeterminate problems occurring in a double support phase. The final 13 input variables from a variety of kinematic and kinetic data for the model were carried out by the Self Organizing Map-Genetic Algorithm General Regression Neural Network, and calculated for one side of ground reaction force from the ‘13-26-1’ of a multi-layer neural network and other one from a dynamic equation. As a result, the correlation coefficients between predicted and measured values were 0.88, 0.47, 0.99 and 0.19 N/kg, 0.25 N/kg, 0.61 N/kg of RMSE at the mediolateral, anterior-posterior, vertical axis, respectively. Especially, the highest prediction rate was shown at the vertical direction ground reaction force which has the biggest variations in motion and a small error at the anterior-posterior and lateral-medial axis compared with the measured value. The results of this study show that ground reaction forces can predict general human movements without force platforms, and we expect them to be used as basic information to interpret inverse dynamics.