Medical Engineering and Physics 103 (2022) 103790

Contents lists available at ScienceDirect

Medical Engineering and Physics

journal homepage: www.elsevier.com/locate/medengphy

Technical note

EMG optimization in OpenSim: A model for estimating lower back kinetics

in gait
Jacob J. Banks a, b, c, *, Brian R. Umberger d, Graham E. Caldwell a
a
University of Massachusetts Amherst, Department of Kinesiology, 110 Totman Building, 30 Eastman Lane, Amherst, MA 01003, United States
b
Beth Israel Deaconess Medical Center, Center for Advanced Orthopaedic Studies, 330 Brookline Avenue, RN 115, Boston, MA 02215, United States
c
Harvard Medical School, Department of Orthopaedic Surgery, Boston, MA 02115, United States
d
University of Michigan, School of Kinesiology, 830 North University Avenue, Ann Arbor, MI 48109, United States

A R T I C L E I N F O A B S T R A C T

Keywords: Participant-specific musculoskeletal models are needed to accurately estimate lower back internal kinetic de­
In silico mands and injury risk. In this study we developed the framework for incorporating an electromyography opti­
Electromyography mization (EMGopt) approach within OpenSim (https://simtk.org/projects/emg_opt_tool) and evaluated lower
Load carriage
back demands estimated from the model during gait. Kinematic, external kinetic, and EMG data were recorded
Evaluation
Sensitivity
from six participants as they performed walking and carrying tasks on a treadmill. For evaluation, predicted
Static optimization lumbar vertebral joint forces were compared to those from a generic static optimization approach (SOpt) and to
Musculoskeletal previous studies. Further, model-estimated muscle activations were compared to recorded EMG, and model
sensitivity to day-to-day EMG variability was evaluated. Results showed the vertebral joint forces from the model
were qualitatively similar in pattern and magnitude to literature reports. Compared to SOpt, the EMGopt
approach predicted larger joint loads (p<.01) with muscle activations better matching individual participant
EMG patterns. L5/S1 vertebral joint forces from EMGopt were sensitive to the expected variability of recorded
EMG, but the magnitude of these differences (±4%) did not impact between-task comparisons. Despite limita­
tions inherent to such models, the proposed musculoskeletal model and EMGopt approach appears well-suited for
evaluating internal lower back demands during gait tasks.

1. Introduction Several open-source lower back musculoskeletal models are avail­

able in OpenSim [9], most based on a model developed by Christophy
Lower back pain is a multifactorial musculoskeletal disorder with a et al. [10] and then refined and evaluated for specific applications
global point prevalence of 7.5% [1]. An important risk factor associated [11–18]. When applying such models, researchers often utilize a static
with the occurrence of lower back pain is the degree of internal lower optimization (SOpt) approach to estimate the individual muscular forces
back kinetic demands (e.g., vertebral joint and muscle) from everyday responsible for the observed kinematics [19]. SOpt assumes that muscles
tasks such as gait [2,3]. Because these demands are impractical to will be recruited to minimize a criterion objective function such as fa­
measure [4], musculoskeletal models of the lower back have been tigue, and is therefore inherently biased against the prediction of
developed based on established anatomy, physiology and biomechanics antagonistic muscle activation [20]. In contrast, incorporating electro­
[5]. To generate confidence in their calculated outcome measures, such myography (EMG) to estimate muscular contributions can reflect
models should be evaluated through comparisons with known measures, participant-specific recruitment strategies, strength potentials, and
previous model calculations, and tested for input parameter sensitivity antagonistic activations [5,21]. The appropriate distribution of indi­
[6–8]. vidual muscle forces is crucial for accurate estimates of stress on

Abbreviation: s:b15BW, Carrying a dumbbell of 15% of bodyweight in both hands; r15BW, Carrying a dumbbell of 15% of bodyweight in right hand only; EMG,
Electromyography; EMGopt, Electromyography Optimization; L5/S1, Joint connecting Lumbar 5 and Sacrum 1 vertebrae; MVC, Maximal Voluntary Contraction;
m29DoF, Model with 29 degrees-of-freedom; m47DoF, Model with 47 degrees-of-freedom; MTA, Musculotendon Actuator; RMSE, Root Mean Squared Error; SOpt,
Static Optimization; SSS, Self-selected Treadmill Walking Speed.
* Corresponding author.
E-mail addresses: [email protected] (J.J. Banks), [email protected] (B.R. Umberger), [email protected] (G.E. Caldwell).

https://doi.org/10.1016/j.medengphy.2022.103790
Received 9 June 2021; Received in revised form 22 October 2021; Accepted 14 March 2022
Available online 18 March 2022
1350-4533/© 2022 IPEM. Published by Elsevier Ltd. All rights reserved.
J.J. Banks et al. Medical Engineering and Physics 103 (2022) 103790

anatomical tissues [22]. EMG-based approaches can provide improved were recruited from a university population and gave written informed
accuracy and insight into how tasks and recruitment strategies affect consent to an institutional review board approved protocol. Participants
lower back loading. were restricted to those under 40 years old, fit (BMI < 30), without a
Models driven strictly by EMG use measured and calibrated muscle history of lower back pain, physically active, and void of neurological
activations to directly predict muscular forces [5]. Due to physiological issues.
complexities and inherent limitations of EMG [5,23], forces predicted
from EMG-driven approaches will not necessarily satisfy the joint de­
2.2. Equipment and Setup
mands calculated from inverse dynamics [24–26]. To overcome this
limitation, a hybrid EMG optimization method (EMGopt) was developed
Full-body three-dimensional kinematics were defined by an array of
by minimally adjusting EMG activations to match the calculated joint
passive 12.5 mm reflective marker positions (Fig. 1) collected via mo­
demands [27]. Thus EMGopt can alleviate flaws from SOpt and
tion capture at 100 Hz (Qualisys AB, Sweden). For the trunk and pelvis
EMG-driven approaches while retaining advantages of each [24–26,28],
specifically, markers were placed on the xiphoid process, sternal notch,
but this approach has yet to be applied to an OpenSim lower back model.
the C7 and T6 vertebrae, and right and left anterior and posterior iliac
Therefore, our objective was to develop and evaluate an EMGopt
spines, iliac crests, and acromia.
framework for defining muscular contributions in a participant specific
Twelve wireless EMG surface electrodes (Delsys Inc., USA) recorded
OpenSim musculoskeletal model of the lower back during gait. The
muscle activations at a rate of 2000 Hz. Electrodes were positioned
lower back demands predicted by EMGopt were evaluated through
based on previous studies (see Table S.1 in Supplemental Material based
comparison with results from a standard SOpt algorithm, and to de­
on references [30,33]) over six bilateral trunk muscles: rectus abdomi­
mands reported in the literature during similar gait tasks [29–32].
nis, external oblique, internal oblique, latissimus dorsi, longissimus
Further, the sensitivity of estimated joint demands to EMG variability
thoracis pars thoracis, and iliocostalis lumborum pars thoracis. All
was investigated. Our motivation was to develop a model that could be
electrode attachment sites on the skin were prepared in accordance to
applied in subsequent examinations of how different gait and load car­
established standards [34].
riage asymmetries can impact lower back demands.
Isometric maximal voluntary contractions (MVCs) were measured
with a dynamometer (Biodex Medical Systems Inc., USA) to determine
2. Methods
maximal EMG amplitudes and trunk torques for each participant. Gait
tasks took place on an instrumented treadmill (Treadmetrix, USA) that
2.1. Participants
provided pacing and measured ground reaction forces . Hand-held
dumbbells adjusted to 15% body weight were carried both bi- and
Six participants (25±3 years; 65.2±9.6 kgs; 171±9 cm; 3 female)
uni-laterally (right hand) during gait carriage tasks.

Fig. 1. Full-body OpenSim musculoskeletal model in the calibration pose. Dark grey spheres represent the location of the 64 reflective markers used for motion
tracking, black rectangles the placement for the 12 EMG sensors used to monitor the muscular activity of the trunk, and the cylindrical rods simulate the added inertia
of dumbbells during carrying tasks.

2
J.J. Banks et al. Medical Engineering and Physics 103 (2022) 103790

2.3. Procedure Participant-specific EMGs were then scaled to their muscle-specific MVC
maximum levels and non-linearly re-scaled to better match the reported
Participants donned form-fitting clothing and comfortable walking EMG-to-force relationship [24]. Filtered external forces and EMG linear
footwear prior to measurements of anthropometry, self-selected tread­ envelopes were down sampled to sync with kinematic data (i.e., 100
mill walking speed (SSS), gait tasks, and MVCs. SSSs were systematically Hz).
defined by prompting participants to verbally indicate how a randomly Gait trials were time normalized by strides demarcated by consecu­
presented treadmill speed compared relative to a “comfortable walking tive right heel strikes identified by heel marker anterior-posterior ve­
speed which they can maintain for 3-5 minutes with minimal effort and locities [44]. Maximum and average values from each dependent
discomfort”. Speeds were blinded from the participant and manually variable of interest were derived from an ensemble average of three
adjusted by the researcher until participants consistently settled on their consecutive strides extracted from the end of each 30 second collection.
SSS within ± 0.05 m/s. A 90% SSS value was calculated for each Vertebral joint forces were normalized to bodyweight, except for indi­
participant, representing the reported walking speed adaptation of rect comparisons with in vivo reports where an average equivalent joint
participants during carrying tasks [35]. After SSS determination, EMG force from standing calibrations was used.
electrodes and reflective markers were placed on the participant.
A standing calibration pose was recorded prior to all tasks (Fig. 1). 2.5. Musculoskeletal Model
Participants then performed three randomly-ordered gait tasks in which
lumbar loading has been 1) directly quantified in vivo [31,32], and 2) Two separate full-body lumbar spine models with 27 segments, six
calculated with other EMG-assisted computer models [29,30]. These lumbar joints, and 238 Hill-type trunk musculotendon actuators (MTAs)
tasks included a) treadmill walking at SSS, and walking at 90% of SSS [,45,[46]] were developed within OpenSim 4.0 [9]. Model m29DoF
with b) a dumbbell of 15% of bodyweight in their right hand (r15BW) or with 29 degrees-of-freedom was based on a previous full-body lifting
c) in both hands (b15BW). During SSS gait, participants acclimated for model [12] and used to determine the recorded trial kinematics (Fig. 2).
90 seconds before data were recorded for 30 seconds [36,37]; the The original model was modified for simplicity, anatomical consistency,
acclimation for dumbbell trials was only 20 seconds [38]. and to improve model sensitivity. First, the model was updated to be
Following the gait tasks, participants performed six different MVC compatible with OpenSim 4.0, allowing for improved usability. The
efforts on the dynamometer against a custom bar designed to resist trunk model wrist joints were welded in a neutral posture, and the patellae
flexion, extension, lateral bending, and axial rotation (see Fig. S.1. in segments and associated joints removed. These modeling details mini­
Supplemental Material for setup and references [30,39]). Five MVC mally contribute to lower back demands and were altered to simplify the
exertions (trunk flexion, bi-lateral twisting, and bi-lateral lateral model and the required marker set. Generic Hill-type [45–47] model
bending) took place from a sit-up position with the torso, knees, and hips parameters (e.g., shape factors for the active and passive force-lengths
flexed at approximately 45◦ , 90◦ , and 90◦ , respectively. The trunk and force-velocity) and external oblique and rectus abdominis MTA
extension MVC was performed with the participant prone and safely attachment points were corrected to better represent those in the liter­
suspended horizontally from the dynamometer seat with a slightly ature [46,48]. All MTA physiological cross section areas were adjusted
flexed torso [40]. To lessen fatigue, only two repetitions of each MVC to match a more homogeneous data set [13]. To determine initial MTA
effort were performed and all exertions and tasks were separated by 2 maximal isometric forces, maximal muscular stresses were set to 100
minutes of rest. N/cm2 in the baseline model prior to adjusting to participant-specific
strengths (see subsequent paragraph for more details). MTA optimal
2.4. Data Processing fiber and tendon slack lengths were calculated based on a simulated
neutral standing posture (see Table S.2 in Supplemental Material for
All data were post-processed in MATLAB (MathWorks, USA). Marker details and reference [13]). In m29DoF the eighteen coordinate coupler
positions, ground reaction forces, and dynamometer torques were constraints of the lower back and abdominals were adjusted to represent
smoothed with a fourth-order zero-lag 6-Hz low pass Butterworth filter, the in vivo contributions of the thoracic vertebrae to trunk kinematics
based on residual analysis [41]. EMGs were detrended, band-pass and abdomen movement relative to the L5/S1 posture (see Table S.3 in
filtered (30 – 500 Hz), full wave rectified, and then low-pass filtered Supplemental Material for details and rerference [13]). Finally, model
at 3 Hz with a similar Butterworth filter [42] before time-shifting by 10 m47Dof was constructed by removing the eighteen coordinate coupler
ms to account for physiological electromagnetic delay [43]. constraints, as the additional degrees-of-freedom permitted more

Fig. 2. Flow chart of model input/output processes. Bracketed numbers depict implementation sequences described in Methods.

3
J.J. Banks et al. Medical Engineering and Physics 103 (2022) 103790

accurate calculation of MTA moment arms (see.Fig. S.2 in Supplemental forces are a product of their current potential, the EMG activations
Material for moment arm comparisons). (emgai), and the adjustment factors (gi). The constraints ensure that the
For each participant the modeling process started by scaling both the optimized MTA forces produce summed moments matching those
m29DoF and m47DoF models using the OpenSim scaling tool to repli­ calculated from inverse dynamics to within ±0.5%. -Specific adjustment
cate their calibration pose and anthropometry (sequence 1 in Fig. 2). The factors kept all MTA forces below their maximal force potential but
muscular strength potentials of the MTAs in the m47DoF model were above 50% (Thres = 0.50) of their measured activation [50]. However,
adjusted further by changing individual maximal muscular stress values MTAs without an assigned electrode were allotted a more lenient Thres
from the default 100 N/cm2 to match the maximal torque outputs from = 0.01 and their activations based on electrode measurements (emgai)
the MVC tasks (sequence 2). Here, the OpenSim inverse dynamics tool from nearby synergists (see Table S.2 in Supplemental Materials for all
integrated external forces and standardized model states to calculate assignments and references [33,51]). EMGopt was implemented in
lumbar joint moments in each of the six MVC positions. Based on EMG MATLAB with the fmincon sequential quadratic programming algorithm.
observations and to incorporate a representative level of antagonistic
muscle activation in the MVC solution, all MTAs were assigned a mini­ 2.7. Model Evaluation and Statistical Analysis
mal lower activation level of 0.05. The highest computed maximal
muscular stress from the six positions was used to determine the Four different evaluation tests were performed, the first comparing
maximal isometric force potential of the trunk flexor and extensor predicted EMGopt and SOpt lumbar joint forces in the three tasks. Three-
muscles. way repeated-measures analyses of variance (ANOVAs) with Tukey
For the gait tasks, generalized segmental coordinates were computed post-hoc testing were performed to test (α < .05) the main effects of the
from recorded kinematic data with the OpenSim inverse kinematics tool two optimization approaches (EMGopt and SOpt), three tasks, and six
using model m29DoF (sequence 3). These segmental coordinates were lumbar vertebral joint levels (L5/S1 thru T12/L1) on dependent vari­
then used to calculate the m47DoF model states and the “top-down” ables representing compressive and shear loading. The second test
lower back joint moments with the OpenSim analysis and inverse dy­ examined how well EMGopt and SOpt predicted activations matched the
namics tools, respectively (sequences 4 and 5). For carrying tasks, recorded EMG. Matches between predicted and recorded activations
dumbbell inertial characteristics were computed as appropriately sized were quantified with cross-correlations and root mean squared errors
solid cylinders welded to the hands. An EMGopt approach (next section) (RMSE) and compared between approaches with paired t-tests. Thirdly,
was applied to partition the lower back joint moments into 238 MTA estimated model lumbar joint forces were indirectly and subjectively
forces (sequence 6). Model MTA, inertial, and external forces were input compared with in silico models from the literature for equivalent
into the OpenSim joint reaction analysis tool to calculate the resultant vertebral joint levels and tasks [29,30]. Literature in vivo pressure
lumbar joint forces, expressed in the caudal lumbar vertebrae coordinate measurements [31] were converted to an estimation of force [4,52,53]
system (sequence 7). For comparison, the analyses were repeated using and vertebral body replacement forces were calculated from publicly
SOpt in OpenSim [9]. All tools and libraries were accessed with MATLAB available data (see Table S.4 in Supplemental Materials for trials used
using the OpenSim application programming interface to minimize the from reference [32]). Finally, the sensitivity of the EMGopt approach
potential for human input error, implement the optimization algorithms, lower back demands to inherent EMG variability was evaluated by
and improve workflow efficiency [49]. artificially adjusting the recorded EMG activations from each of the six
bi-lateral trunk muscles by ±10% [54–56]. For each adjustment, peak
2.6. EMG Optimization Approach L5/S1 joint forces from the three tasks were compared to values from the
unadjusted “baseline” model using paired t-tests, while between task
Individual MTA forces were computed with EMGopt [27] using an differences were assessed using a one-way repeated-measures ANOVA
objective function (JEMG) that balanced vertebral joint moments with and post-hoc tests. All statistical tests were performed using the MAT­
minimal adjustments to the measured EMG activations following an LAB Statistical Toolbox.
established multi-joint convention [33]:
3. Results and Discussion
∑
238
JEMG = min Mnormi (1 − gi ) 2
(1)
i=1
3.1. Evaluation 1: Predicted Joint Forces for EMG and Static
Optimization
with:
√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ Lumbar vertebral joint force estimates from EMGopt and SOpt dis­
∑( )̅
Mnormi = MMTAx(i,j) 2 + MMTAy(i,j) 2 + MMTAz(i,j) 2 (2) played similar patterns across the gait cycle (see Fig. S.3 in Supple­
i mental Material for vertebral joint forces during SSS walking at 1.19 ±
0.19 m/s) and resembled literature reports [29] during comparable
which will minimize the adjustments (gi) made to the predicted walking and carrying tasks. Qualitatively, joint forces peaked following
muscular forces based on the Euclidean moments norm (Mnorm) from heel strikes and were minimal near contralateral toe-off. As expected,
each MTA (i = 1 to 238) about each lumbar joint (j = 1 to 6) subject to joint forces differed across tasks (p < .01), with the b15BW and r15BW
the following constraints: carrying tasks exhibiting larger forces than SSS gait. Average absolute
⃒ ⃒ ⃒ ⃒ peak compressive and resultant shear forces from EMGopt and SOpt
⃒ ⃒ ⃒ ⃒
⃒gi MMTAxj − MIDxj ⃒ ≤ ⃒0.005 ∗ MIDxj ⃒ (3) (Fig. 3) varied across vertebral joint level (p < .01), with lower magni­
tudes at the mid lumbar level where the spinal lordotic curve positioned
⃒ ⃒ ⃒ ⃒
⃒
⃒gi MMTAyj −
⃒ ⃒ ⃒
MIDyj ⃒ ≤ ⃒0.005 ∗ MIDyj ⃒ (4) the vertebrae closer to the center of mass of the trunk during gait [57,
58]. This differs from lifting tasks, where the more superior (e.g.
⃒ ⃒ ⃒ ⃒ T12/L1) lumbar joints are flexed closer to the external load, resulting in
⃒ ⃒ ⃒ ⃒
⃒gi MMTAzj − MIDzj ⃒ ≤ ⃒0.005 ∗ MIDzj ⃒ (5) lower joint forces compared to inferior joints [57,59].
EMGopt predicted significantly larger (p < .01) and more variable
/
Thres ≤ gi ≤ 1 emgai (6) joint forces than SOpt. Unlike SOpt, EMGopt considers participant-
specific MTA activations that may include supplementary antagonistic
where x, y, and z represent the orthogonal moments (MMTA) resulting co-contraction related to joint stability and the inherent variability of
from the MTA forces (i = 1 to 238) and their moment arms. The MTA EMG [6,24,60,61]. Because muscular forces directly influence joint

4
J.J. Banks et al. Medical Engineering and Physics 103 (2022) 103790

Fig. 3. Peak vertebral lumbar joint forces (as a percentage of bodyweight) for EMGopt (maroon bars) and SOpt (grey bars) approach estimates averaged across
participants for each task and lumbar level (indicated by superior/inferior vertebrae number). Data whiskers indicate ±1 standard deviation.

loading, modeling approaches that predict co-contraction using EMG cross-correlation coefficients (r-values; 1.0 perfect correspondence)
will generally lead to greater and more variable joint loads [22,28,62]. [63]. Here, MTA activations correlated well with the recorded EMG
This observed effect suggests that the EMGopt approach was imple­ patterns for both optimization approaches (r-values ranging from .70 to
mented correctly, although by themselves larger and more variable joint .99; Fig. 4 and Table 1). However, EMGopt outperformed SOpt across
force estimates do not necessarily ensure accurate vertebral joint load most temporal comparisons (average r-values SSS: .93 vs. .77; b15BW:
estimates. .95 vs. .82; r15BW: .96 vs. .92; p-values all < .05). The only exception
was the right rectus abdominis during the r15BW carrying task, in which
the r-value was higher for SOpt (.97 vs. .87; p < .01) due to the pref­
3.2. Evaluation 2: Predicted vs. Recorded EMG Activations
erence of the model to minimize antagonistic forces following heel
strike.
In the absence of measured in vivo joint loads, the comparison of
Likewise, the magnitude of model MTA activations were compared
predicted and recorded muscle activations can aid model evaluation [7].
to recorded EMGs using RMSE, with lower RMSE values representing
For our three tasks we quantified how well EMGopt and SOpt MTA ac­
better model performance. Compared to SOpt, EMGopt more closely
tivations correlated temporally with recorded EMG using zero-lag

Fig. 4. Average participant muscle group activity from recorded EMG and predictions from EMGopt and SOpt optimization approaches. Columns depict left and right
trunk muscles during A) self-selected walking speed (SSS), B) carrying 15% bodyweight in each hand (b15BW), and C) carrying 15% bodyweight in the right hand
only (r15BW). Dashed black, maroon, and grey lines represent activations from the EMG, EMGopt, and SOpt, respectively. Standard deviations have been excluded for
clarity. Abbreviations: left (LTO) or right toe-offs (RTO), and left heel-strike (LHS).

5
J.J. Banks et al. Medical Engineering and Physics 103 (2022) 103790

matched recorded EMG magnitudes with lower RMSE values across

Average temporal cross-correlation (r) and root mean squared error (RMSE) comparisons between muscle group activations predicted from either optimization approach and recorded EMG across three tasks (SSS, b15BW,
and r15BW). Muscle Group abbreviations: lower case letter denotes trunk side; capital letters denote external oblique (EO), iliocostalis lumborum (IL), internal oblique (IO), latissimus dorsi (LD), longissimus thoracis (LT),

Significant (p < .05) differences between optimization approaches are represented by greater (>) and less (<) than signs between columns to indicate relationship direction. Refer to text for optimization approach, task,
(.04)
(.03)
(.14)
(.05)
(.13)
(.16)
(.02)
(.04)
(.03)
(.01)
(.04)
(.04)
(.10)
most muscle groups and all tasks (Fig. 4 and Table 1; average RMSE SSS:
.03 vs. .06; b15BW .03 vs. .08; r15BW: .05 vs. .11; p-values all < .05).

SOpt

.07
.07
.20
.04
.19
.25
.07
.07
.11
.02
.15
.05
.11
For both approaches, temporal and magnitude correspondence with
recorded EMG was comparable to published lower back model evalua­

<

<
<
<

<

<
<
<
tion studies [11,12,14,15,64,65], but overall EMGopt produced the
better match. This again suggests that the EMGopt algorithm was suc­
(.05)
(.03)
(.07)
(.04)
(.02)
(.05)
(.01)
(.02)
(.02)
(.01)
(.02)
(.02)
(.04)
EMGopt
cessfully implemented and is preferable if sufficient EMG data are
r15BW

available [66].
.09
.04
.10
.07
.03
.03
.02
.03
.03
.01
.09
.03
.05
3.3. Evaluation 3: Comparisons with the Literature
(.02)
(.04)
(.09)
(.06)
(.13)
(.11)
(.03)
(.05)
(.03)
(.03)
(.03)
(.04)
(.08)
SOpt

A new model should demonstrate some level of consistency with

.04
.05
.07
.06
.22
.18
.07
.08
.03
.04
.10
.08
.08
previous evaluated models and in vivo measurements at equivalent tasks
[7]. However, discrepancies between computer models will arise due to
<
<
<
<

<
<
<
<
architectural differences (e.g., MTA geometry, joint design, control
scheme) and there is no definitive method to determine which estimate,
(.03)
(.06)
(.02)
(.02)
(.02)
(.01)
(.01)
(.02)
(.01)
(.01)
(.02)
(.02)
(.03)
EMGopt
b15BW

if either, is an accurate representation. Similarly, in vivo measurements

.04
.06
.04
.03
.02
.02
.01
.02
.01
.01
.05
.04
.03
are limited by factors such as their invasiveness, low participant
numbers with age and health variation, and unaccounted tissue loading
[4,31]. Therefore, taking these limitations into consideration we only
(.02)
(.02)
(.05)
(.03)
(.07)
(.06)
(.02)
(.02)
(.02)
(.03)
(.02)
(.03)
(.05)

qualitatively compared our EMGopt results with previous literature

SOpt

.02
.03
.04
.04
.15
.13
.04
.05
.04
.04
.07
.06
.06

reports.
In general, the model results compared well to published in silico and
in vivo data, with similar force ranges and trends across lumbar levels
<
<
<
<
<
<
<
<
<

and the three tasks (Fig. 5). The largest discrepancy was in resultant
shear force estimates, but the force values were still ranked similarly
.06 (.04)
.08 (.06)
.03 (.02)
.04 (.02)
.01 (.01)
.02 (.01)
.01 (.01)
.01 (.00)
.01 (.01)
.01 (.01)
.04 (.01)
.03 (.01)
.03 (.03)
EMGopt
RMSE

between tasks [29–32]. Shear loads are typically more sensitive than
SSS

compression forces to model and task kinematic differences [67].

Overall the level of agreement with published reports is comparable to
(.04)
(.04)
(.06)
(.02)
(.05)
(.02)
(.06)
(.08)
(.02)
(.02)
(.05)
(.03)
(.06)

other lower back models [11–13,15,16,68–70].

SOpt

.87
.97
.89
.86
.89
.98
.90
.87
.95
.97
.87
.97
.92

3.4. Evaluation 4: Sensitivity of model to EMG magnitudes

>
<

>
>

>
>
>
>
>

>

A critical component in the EMGopt approach is the inclusion of

recorded EMG, which is absent from the SOpt approach. While allowing
(.03)
(.05)
(.04)
(.03)
(.01)
(.03)
(.02)
(.03)
(.01)
(.01)
(.01)
(.02)
(.04)
EMGopt

for participant specificity, the EMG data can introduce additional vari­
r15BW

ability. To test the sensitivity of L5/S1 joint force estimates to potential

.93
.87
.95
.92
.97
.96
.98
.96
.99
.99
.99
.99
.96

day-to-day variation in EMG, we artificially adjusted the recorded EMG

activation of each bilateral muscle group electrode pairing by ±10%
(.14)
(.11)
(.07)
(.06)
(.10)
(.10)
(.08)
(.03)
(.02)
(.02)
(.07)
(.06)
(.09)

[54–56]. Table 2 shows that the degree of sensitivity in L5/S1 joint

SOpt

.75
.80
.75
.78
.80
.79
.84
.81
.93
.94
.84
.87
.82

forces was muscle group dependent. L5/S1 joint loads were more sen­
sitive to activation variation in internal obliques, longissimus thoracis,
and iliocostalis lumborum, MTAs that exhibited better correspondence
>
>
>
>
>
>
>
>
>
>
>

between model activities and measured EMGs in Evaluation 2 or had

more MTAs crossing the L5/S1 joint. In contrast, L5/S1 forces were less
(.09)
(.06)
(.04)
(.04)
(.04)
(.04)
(.01)
(.03)
(.01)
(.01)
(.00)
(.00)
(.06)
EMGopt
b15BW

sensitive to variation for MTAs that exhibited poorer correspondence

.87
.89
.93
.90
.94
.94
.97
.97
.99
.99
.99
.99
.95

between predicted and recorded activations (external obliques and

rectus abdominis) or those with fewer MTAs crossing the L5/S1 joint
(latissimus dorsi). When present, joint force changes were directly
(.05)
(.09)
(.09)
(.09)
(.06)
(.08)
(.08)
(.08)
(.04)
(.02)
(.09)
(.09)
(.09)

related to EMG changes (i.e., L5/S1 loads usually were increased in the
SOpt

.70
.70
.72
.71
.76
.80
.80
.75
.84
.87
.81
.85
.77

+10% condition, decreased in the -10% condition).

Cross-Correlation (r)

Previous studies [71–74] have reported sensitivity of lower back

>
>
>
>
>
>
>
>
>
>
>
>
>

models to various model characteristics (e.g. MTA design, EMG-force

relation, daily variation in static/dynamic exertions), but not to the
.88 (.03)
.89 (.04)
.89 (.07)
.88 (.07)
.92 (.05)
.92 (.06)
.93 (.03)
.94 (.03)
.98 (.02)
.99 (.01)
.99 (.00)
.99 (.01)
.93 (.06)

potential daily variation in EMG for EMGopt or during level walking and
EMGopt

and muscle group abbreviations.

carrying tasks. When implementing EMG into a lower back model,

SSS

added provisions should be taken to ensure the measured activations

and rectus abdominis (RA).

from MTAs crossing joints of interest are accurate. Still, although there
rRA
rEO
lRA

rLD
lEO

lLD

rLT
rIO

lLT
lIO

rIL

were significant differences in 38 of 72 variability tests, the largest

lIL

L5/S1 joint force changes we observed were less than ±4% from base­
line. Furthermore, all sensitivity tests resulted in similar between-task
Muscle Group

trends. These results suggest that even with the potential for daily
Average

variation, the EMGopt approach with our model is appropriate to

Table 1

compare different gait tasks in a repeated-measures study design with

normal day-to-day variability in recorded EMG.

6
J.J. Banks et al. Medical Engineering and Physics 103 (2022) 103790

Fig. 5. Indirect comparison of lumbar joint forces between the proposed model (maroon bars) and published (white bars) estimates across equivalent tasks and
lumbar levels. Row A) in silico compression (left panel) and resultant shear (right panel) forces (as a percentage of bodyweight); row B) in vivo forces (as a percentage
of a relaxed standing posture). All forces reflect the vertebral joint level and measure from each referenced model.

Table 2
Peak L5/S1 joint forces (as a percentage of bodyweight) for Baseline and when EMG activation levels were modified by ±10% of their measured Baseline activity.
Adjustment Compression Resultant Shear
SSS b15BW r15BW SSS b15BW r15BW

Baseline 311.3 (89.4)a 360.6 (75.7)a,b 403.1 (111.8)b 46.8 (14.6)a 52.2 (14.5)a 65.1 (19.6)b
Muscle Group Rectus þ10% 311.0 (88.9)a 359.8 (75.6)a,b 402.9 (111.0)b 46.7 (14.5)a 51.8 (14.4)a 65.2 (19.4)b
Abdominis -10% 311.7 (89.9)a 359.9 (75.5)a,b 403.2 (112.6)b 47.0 (14.7)a 52.1 (14.4)a 65.1 (19.8)b
External þ10% 312.0 (89.4)a 360.2 (76.0)a,b 404.5 (112.1)b 47.1 (14.6)a 52.1 (14.4)a 65.7 (19.5)b
Obliques -10% 310.7 (89.4)a 359.9 (74.8)a,b 401.6 (111.4)b 46.5 (14.5)a 51.7 (14.4)a 64.5 (19.8)b
Internal þ10% 312.8 (89.8)a 361.6 (76.0)a,b 404.8 (112.4)b 46.5 (14.6)a 51.5 (14.5)a 64.7 (19.7)b
Obliques -10% 309.8 (89.1)a 358.4 (74.8)a,b 401.2 (111.1)b 47.1 (14.5)a 52.6 (14.4)a 65.6 (19.5)b
Latissimus þ10% 311.6 (89.7)a 360.0 (75.7)a,b 403.6 (112.4)b 47.0 (14.6)a 52.1 (14.4)a 65.3 (19.8)b
Dorsi -10% 311.1 (89.1)a 360.9 (74.2)a,b 402.5 (111.1)b 46.6 (14.5)a 52.0 (14.5)a 64.9 (19.4)b
Longissimus þ10% 316.7 (92.3)a 364.7 (78.4)a,b 411.5 (118.6)b 48.5 (15.3)a 53.5 (14.9)a 67.6 (21.0)b
Thoracis -10% 305.7 (86.5)a 354.8 (72.5)a,b 394.1 (104.9)b 45.0 (13.9)a 50.5 (14.0)a 62.8 (17.8)b
Iliocostalis þ10% 317.1 (92.0)a 365.8 (77.3)a,b 412.0 (117.8)b 47.7 (15.1)a 52.7 (14.4)a 66.8 (20.3)b
Lumb -10% 305.4 (86.7)a 354.1 (73.4)a,b 393.7 (105.1)b 45.8 (14.1)a 51.1 (14.3)a 63.7 (18.6)b

Significant (p < .05) differences from Baseline model are in bold and letters (a & b) denote between task groupings within each scenario. Refer to text for task and
muscle group abbreviations.

3.5. Limitations of Model and Evaluation The EMGopt approach optimizes a generic gain adjustment factor
which was not directly related to any physiological characteristic(s) of
Our results are subject to limitations associated with the musculo­ the MTAs. Other modeling studies attempt to optimize physiological
skeletal model, EMGopt approach, and our experimental setup. The properties of the MTAs to closely match the kinetics of calibration tasks
current model neglects passive force contributions from non-muscular [23,79,80] but have their own limitations [5,24]. Ultimately, the
tissues and intra-abdominal pressure, but these should be minimal for frame-by-frame gain adjustment of EMGopt can be viewed as an
the upright postures and low exertion levels of gait [29,75]. Similar aggregate of all unknown physiological factors for the MTAs and thereby
reasoning justifies our omission of vertebral joint and axis of rotation reduces the number of parameters being optimized and simplifies the
translations [76]. The vertebral joint coupling in model m29DoF was process. In addition, the availability, reliability, and interpretation of
generic and assigns the total thoracic rotational contributions to the EMG recordings is often problematic [66]. We applied reasonable
T12/L1 joint. For our experimental setup, the complex measurement of muscle synergies, activation to force relationships, and lower threshold
each joint individually was not possible. While the accurate assignment bounds to solve kinetic demands during relatively neutral and low
of thoracic contributions to trunk movement can impact thoracic impact tasks. The sensitivity of the predicted model forces to these as­
loading, its impact on lumbar loads is often insignificant and not justi­ sumptions is unknown.
fied [77]. Further, the musculoskeletal model described here only pro­ All tested conditions were performed on a treadmill to allow for
vides an estimate of the net joint contact forces. Developing a multiscale precise control of gait speed and facilitate efficient data collection of
modeling approach including a finite-element model capable of consecutive gait cycles and conditions, differences in spine kinematics
distributing contact forces across soft tissues may be beneficial [78]. and ground reaction forces have been reported for treadmill versus over

7
J.J. Banks et al. Medical Engineering and Physics 103 (2022) 103790

ground walking [81]. We hypothesize the proposed model and approach studies. J Biomech 2016;49:833–45. https://doi.org/10.1016/j.
jbiomech.2015.12.038.
can be applied to most gait scenarios given that the literature reports
[5] Davis KG, Jorgensen MJ. Biomechanical modeling for understanding of low back
indirectly compared to our results were from over ground trials per­ injuries : A systematic review. Occup Ergon 2005;5:57–76.
formed at SSS, but this should be confirmed. A bottom-up model to [6] Cholewicki J, McGill SM. Mechanical stability of the in vivo lumbar spine:
calculate the kinetic demands at each joint is often more sensitive to the Implications for injury and chronic low back pain. Clin Biomech 1996;11:1–15.
https://doi.org/10.1016/0268-0033(95)00035-6.
impulsive heel strike forces [29,82]. However, to avoid any noise [7] Hicks JL, Uchida TK, Seth A, Delp SL, Rajagopal A, Delp SL. Is My Model Good
resulting from the moving treadmill belt and accommodate future setups Enough? Best Practices for Verification and Validation of Musculoskeletal Models
without ground reaction forces, the current model applied a top-down and Simulations of Movement. J Biomech Eng 2015;137:020905. https://doi.org/
10.1115/1.4029304.
inverse dynamics approach. Future work should compare top-down [8] Neptune RR. Neptune_2000_PMRCNA.pdf. Sci Princ Sport Rehabil 2000;11:
and bottom-up approaches within this full-body OpenSim model dur­ 417–34.
ing a variety of tasks performed on stationary force plates. [9] Delp SL, Anderson FC, Arnold AS, Loan JP, Habib A, John CT, et al. OpenSim: Open
source to create and analyze dynamic simulations of movement. IEEE Trans
Biomed Eng 2007;54:1940–50. https://doi.org/10.1109/TBME.2007.901024.
4. Conclusion [10] Christophy M, Senan NAF, Lotz JC, O’Reilly OM. A Musculoskeletal model for the
lumbar spine. Biomech Model Mechanobiol 2012;11:19–34. https://doi.org/
10.1007/s10237-011-0290-6.
In this work, we presented the framework for incorporating an [11] Actis JA, Honegger JD, Gates DH, Petrella AJ, Nolasco LA, Silverman AK.
EMGopt approach in OpenSim (now available through https://simtk.org Validation of lumbar spine loading from a musculoskeletal model including the
/projects/emg_opt_tool) using a modified lower back model and then lower limbs and lumbar spine. J Biomech 2018;68:107–14. https://doi.org/
10.1016/j.jbiomech.2017.12.001.
evaluated its efficacy in studying lower back demands during gait and [12] Beaucage-Gauvreau E, Robertson WSP, Brandon SCE, Fraser R, Freeman BJC,
carrying tasks. Across different gait and carrying tasks, the proposed Graham RB, et al. Validation of an OpenSim full-body model with detailed lumbar
EMGopt approach predicted similar responses to models previously re­ spine for estimating lower lumbar spine loads during symmetric and asymmetric
lifting tasks. Comput Methods Biomech Biomed Engin 2019;epub:1–14. https://
ported in the literature and our own measured EMG data. By incorpo­ doi.org/10.1080/10255842.2018.1564819.
rating EMG, this approach estimated larger joint loads and more closely [13] Bruno AG, Bouxsein ML, Anderson DE. Development and Validation of a
matched individual participant muscular recruitment strategies than a Musculoskeletal Model of the Fully Articulated Thoracolumbar Spine and Rib Cage.
J Biomech Eng 2015;137:081003. https://doi.org/10.1115/1.4030408.
more traditional SOpt approach. Joint loads from EMGopt were sensi­
[14] Molinaro DD, King AS, Young AJ. Biomechanical analysis of common solid waste
tive to the inherent variability in recorded EMG, but the magnitude of collection throwing techniques using OpenSim and an EMG-assisted solver.
these differences did not impact the between task comparisons. Overall, J Biomech 2020;104:109704. https://doi.org/10.1016/j.jbiomech.2020.109704.
[15] Raabe ME, Chaudhari AMW. An investigation of jogging biomechanics using the
the EMGopt optimization approach was successfully implemented, and
full-body lumbar spine model: Model development and validation. J Biomech
the model appears well-suited for the inclusion of participant-specific 2016;49:1238–43. https://doi.org/10.1016/j.jbiomech.2016.02.046.
muscle recruitment strategies and evaluating the lower back joint de­ [16] Senteler M, Weisse B, Rothenfluh DA, Snedeker JG. Intervertebral reaction force
mands of walking and carrying tasks in a repeated-measures study prediction using an enhanced assembly of OpenSim models. Comput Methods
Biomech Biomed Engin 2016;19:538–48. https://doi.org/10.1080/
design. 10255842.2015.1043906.
[17] Zhu XY, Kim HK, Zhang Y. Development of an Enhanced Musculoskeletal Model for
Simulating Lumbar Spine Loading During Manual Lifting Tasks. Lect Notes Comput
Ethical Approval
Sci 2017;(July):229–37. https://doi.org/10.1007/978-3-319-58463-8.
[18] Favier CD, Finnegan ME, Quest RA, Honeyfield L, McGregor AH, Phillips ATM. An
This study protocol (Protocol ID: 2018-5208) was reviewed and open-source musculoskeletal model of the lumbar spine and lower limbs: a
validation for movements of the lumbar spine. Comput Methods Biomech Biomed
approved by the University of Massachusetts Amherst IRB, Federal Wide
Engin 2021;0:1–16. https://doi.org/10.1080/10255842.2021.1886284.
Assurance #00003909. [19] Crownshield RD, Brand RA. A Physiologically Based Criterion of Muscle Force
Prediction in Locomotion. J Biomech 1981;14:793–801.
[20] Marras WS. Predictions of Force Acting Upon the Lumbar Spine Under Isometric
Declaration of Competing Interest and Isokinetic Conditions: A Model - Experimental Comparison. Int J Ind Ergon
1988;3:19–27.
[21] Le P, Best TM, Khan SN, Mendel E, Marras WS. A review of methods to assess
None to declare. coactivation in the spine. J Electromyogr Kinesiol 2017;32:51–60. https://doi.org/
10.1016/j.jelekin.2016.12.004.
[22] Marras WS. Occupational Low Back Disorders Causation and Control. Ergonomics
Acknowledgements
2000;43:880–902.
[23] van Dieën JH, Visser B. Estimating net lumbar sagittal plane moments from EMG
This work was funded in part by a dissertation research grant from data. The validity of calibration procedures. J Electromyogr Kinesiol 1999;9:
the University of Massachusetts Amherst Graduate School (JJB) and a 309–15. https://doi.org/10.1016/S1050-6411(99)00004-8.
[24] Cholewicki J, McGill SM, Norman RW. Comparison of muscle forces and joint load
UMass Department of Kinesiology Priscilla Clarkson scholarship (JJB). from an optimization and EMG assisted lumbar spine model: Towards development
of a hybrid approach. J Biomech 1995;28. https://doi.org/10.1016/0021-9290
(94)00065-C.
Supplementary materials
[25] Gagnon D, Larivière C, Loisel P. Comparative ability of EMG, optimization, and
hybrid modelling approaches to predict trunk muscle forces and lumbar spine
Supplementary material associated with this article can be found, in loading during dynamic sagittal plane lifting. Clin Biomech 2001;16:359–72.
https://doi.org/10.1016/S0268-0033(01)00016-X.
the online version, at doi:10.1016/j.medengphy.2022.103790.
[26] Choi H, Vanderby RJ. Comparison of Biomechanical Human Neck Models: Muscle
Forces and Spinal Loads at C4/5 Level. J Appl Biomech 1999;15:120–38.
References [27] Cholewicki J, McGill SM. EMG Assisted Optimization: A Hybrid Approach for
Estimating Muscle Forces in an Indeterminate Biomechanical Model. J Biomech
1994;27:1287–9.
[1] Wu A, March L, Zheng X, Huang J, Wang X, Zhao J, et al. Global low back pain
[28] Li SSW, Chow DHK. Comparison of Predictions Between an EMG- Assisted
prevalence and years lived with disability from 1990 to 2017: estimates from the
Approach and Two Optimization-Driven Approaches for Lumbar Spine Loading
Global Burden of Disease Study 2017. Ann Transl Med 2020;8:1–14. https://doi.
During Walking With Backpack Loads. Hum Factors 2019. https://doi.org/
org/10.21037/atm.2020.02.175.
10.1177/0018720819851299. epub.
[2] da Costa BR, Vieira ER. Risk factors for work-related musculoskeletal disorders : A
[29] Callaghan JP, Patla AE, McGill SM. Low back three-dimensional joint forces,
systematic review of recent longitudinal studies. Am J Ind Med 2010;53:285–323.
kinematics, and kinetics during walking. Clin Biomech 1999;14:203–16. https://
https://doi.org/10.1002/ajim.20750.
doi.org/10.1016/S0268-0033(98)00069-2.
[3] Norman RW, Wells RP, Neumann P, Frank JS, Shannon H, Kerr M, et al.
[30] McGill SM, Marshall LW, Andersen J. Low back loads while walking and carrying:
A comparison of peak vs cumulative physical work exposure risk factors for the
comparing the load carried in one hand or in both hands. Ergonomics 2013;56:
reporting of low back pain in the automotive industry. Clin Biomech 1998;13:
293–302.
561–73. https://doi.org/10.1016/S0268-0033(98)00020-5.
[4] Dreischarf M, Shirazi-Adl A, Arjmand N, Rohlmann A, Schmidt H. Estimation of
loads on human lumbar spine: A review of in vivo and computational model

8
J.J. Banks et al. Medical Engineering and Physics 103 (2022) 103790

[31] Wilke H-J, Neef P, Caimi M, Hoogland T, Claes LE. New In Vivo measurements of [58] Raabe ME, Chaudhari AMW. Biomechanical consequences of running with deep
pressures in the invertebral disc in daily life. Spine (Phila Pa 1976) 1999;24: core muscle weakness. J Biomech 2018;67:98–105. https://doi.org/10.1016/j.
755–62. jbiomech.2017.11.037.
[32] Bergmann G (ed.), Charite Universitaetsmedizin Berlin. “Orthoload” 2008. htt [59] Arjmand N, Shirazi-Adl A. Sensitivity of kinematics-based model predictions to
ps://orthoload.com (accessed December 30, 2020). optimization criteria in static lifting tasks. Med Eng Phys 2006;28:504–14. https://
[33] Gagnon D, Arjmand N, Plamondon A, Shirazi-Adl A, Larivière C. An improved doi.org/10.1016/j.medengphy.2005.10.001.
multi-joint EMG-assisted optimization approach to estimate joint and muscle forces [60] Granata KP, Orishimo KF. Response of trunk muscle coactivation to changes in
in a musculoskeletal model of the lumbar spine. J Biomech 2011;44:1521–9. spinal stability. J Biomech 2001;34:1117–23. https://doi.org/10.1016/S0021-
https://doi.org/10.1016/j.jbiomech.2011.03.002. 9290(01)00081-1.
[34] Tankisi H, Burke D, Cui L, de Carvalho M, Kuwabara S, Nandedkar SD, et al. [61] Huebner A, Faenger B, Schenk P, Scholle HC, Anders C. Alteration of Surface EMG
Standards of instrumentation of EMG. Clin Neurophysiol 2020;131:243–58. amplitude levels of five major trunk muscles by defined electrode location
https://doi.org/10.1016/j.clinph.2019.07.025. displacement. J Electromyogr Kinesiol 2015;25:214–23. https://doi.org/10.1016/
[35] Crosbie J, Flynn W, Rutter L. Effect of side load carriage on the kinematics of gait. j.jelekin.2014.11.008.
Gait Posture 1994;2:103–8. [62] Caldwell GE, Chapman AE. The general distrubtion problem: A physiological
[36] Matsas A, Taylor NF, McBurney H. Knee joint kinematics from familiarised solution which includes antagonism. Hum Mov Sci 1991;10:355–92.
treadmill walking can be generalised to overground walking in young unimpaired [63] Hinkle D, Wiersma W, Jurs SG. Applied Statistics for Behavioral Sciences. 5th ed.
subjects. Gait Posture 2000;11:46–53. Houghton Mifflin Harcourt; 2003.
[37] Taylor NF, Evans OM, Goldie PA. Angular movements of the lumbar spine and [64] Cappozzo A, Gazzani F. Spinal Loading During Abnormal Walking. In: Huiskes R,
pelvis can be reliably measured after 4 minutes of treadmill walking. Clin Biomech van Campen D, De Wijn J, editors. Biomech. Princ. Appl. Martinus Nijhoff
1996;11:484–6. Publishers; 1982. p. 141–8.
[38] Dick RB, Hudock SD, Lu M-LL, Waters TR, Putz-Anderson V. Manual Materials [65] El-Rich M, Shirazi-Adl A, Arjmand N. Muscle Activity, Internal Loads, and Stability
Handling. In: Stave GM, Wald PH, editors. Phys. Biol. Hazards Work. 3rd ed. John of the Human Spine in Standing Postures: Combined Model and In Vivo Studies.
Wiley & Sons Inc.; 2017. p. 33–52. https://doi.org/10.5334/bbe.j. Spine (Phila Pa 1976) 2004;29:2633–42.
[39] McGill SM. Electromyographic Activity of the Abdominal and Low Back [66] De Luca CJ. The Use of Surface Electromyography in Biomechanics. J Appl
Musculature During the Generation of Isometric and Dynamic Axial Trunk Torque: Biomech 1997;13:135–63.
Implications for Lumbar Mechanics. J Orthop Res 1991;9:91–103. [67] Kingma I, Faber GS, van Dieën JH. Supporting the upper body with the hand on the
[40] Biering-Sorensen F. Physical Measurements as Risk Indicators for Low-Back thigh reduces back loading during lifting. J Biomech 2016;49:881–9. https://doi.
Trouble Over a One-Year Period. Spine (Phila Pa 1976) 1984;9:106–19. org/10.1016/j.jbiomech.2015.09.035.
[41] Winter DA. Biomechanics and Motor Control of Human Movement. 4th ed. [68] Arjmand N, Shirazi-Adl A. Model and in vivo studies on human trunk load
Hoboken, NJ: John Wiley & Sons, Inc.; 2009. partitioning and stability in isometric forward flexions. J Biomech 2006;39:
[42] Brereton LC, McGill SM. Frequency response of spine extensors during rapid 510–21. https://doi.org/10.1016/j.jbiomech.2004.11.030.
isometric contractions: Effects of muscle length and tension. J Electromyogr [69] Malakoutian M, Street J, Wilke H-J, Stavness I, Fels S, Oxland T. A musculoskeletal
Kinesiol 1998;8:227–32. https://doi.org/10.1016/S1050-6411(98)00009-1. model of the lumbar spine using ArtiSynth–development and validation. Comput
[43] Corcos DM, Gottlieb GL, Latash ML, Almeida GL, Agarwal GC. Electromechanical Methods Biomech Biomed Eng Imaging Vis 2018;6:483–90. https://doi.org/
Delay: An Experimental Artifact. J Electromyogr Kinesiol 1992;2:59–68. 10.1080/21681163.2016.1187087.
[44] Zeni JAA, Richards JGG, Higginson JS. Two simple methods for determining gait [70] Schultz AB, Andersson GBJ, Ortengren R, Haderspeck K, Nachemson A. Loads on
events during treadmill and overground walking using kinematic data. Gait Posture the Lumbar Spine: Validation of a Biomechanical Analysis by Measurement of
2008;27:710–4. https://doi.org/10.1016/j.gaitpost.2007.07.007. Intradiscal Pressures and Myoelectric Signals. J Bone Jt Surg 1982;64A:713–20.
[45] Hill AV. The Heat of Shortening and the Dynamic Constants of Muscle. Proc R Soc B [71] Brand RA, Pedersen DR, Friederich JA. The Sensitivity of Muscle Force Predictions
Biol Sci 1938;126:136–95. https://doi.org/10.1098/rspb.1938.0050. to Changes in Physiological Cross-Sectional Area. J Biomech 1986;19:589–96.
[46] Thelen DG. Adjustment of muscle mechanics model parameters to simulate [72] Marras WS, Granata KP, Davis KG. Variability in spine loading model performance.
dynamic contractions in older adults. J Biomech Eng 2003;125:70–7. https://doi. Clin Biomech 1999;14:505–14. https://doi.org/10.1016/S0268-0033(98)90090-0.
org/10.1115/1.1531112. [73] Nussbaum MA, Chaffin DB, Rechtien CJ. Muscle lines-of-action affect predicted
[47] Zajac FE. Muscle and tendon: properties, models, scaling, and application to forces in optimization-based spine muscle modeling. J Biomech 1995;28:401–9.
biomechanics and motor control. Crit Rev Biomed Eng 1989;17:359–411. https:// https://doi.org/10.1016/0021-9290(94)00078-I.
doi.org/10.1016/j.pcad.2015.11.006. [74] Sparto PJ, Parnianpour M, Marras WS, Granata KP, Reinsel TE, Simon SR. Effect of
[48] Stokes IAF, Gardner-Morse MG. Quantitative anatomy of the lumbar musculature. Eltromyogram-Force Relationships and Method of Gain Estimation on the
J Biomech 1999;32:311–6. https://doi.org/10.1016/S0021-9290(98)00164-X. Predictions of an Electromyogram-Driven Model of Spinal Loading. Spine (Phila Pa
[49] Lee L-F, Umberger BR. Generating optimal control simulations of musculoskeletal 1976) 1998;23:423–9.
movement using OpenSim and MATLAB. PeerJ 2016;4:1–18. https://doi.org/ [75] Arshad R, Angelini L, Zander T, Di Puccio F, El-Rich M, Schmidt H. Spinal loads
10.7717/peerj.1638. and trunk muscles forces during level walking – A combined in vivo and in silico
[50] Gagnon D, Plamondon A, Larivière C. A biomechanical comparison between expert study on six subjects. J Biomech 2018;70:113–23. https://doi.org/10.1016/j.
and novice manual materials handlers using a multi-joint EMG-assisted jbiomech.2017.08.020.
optimization musculoskeletal model of the lumbar spine. J Biomech 2016;49: [76] Ghezelbash F, Arjmand N, Shirazi-Adl A. Effect of intervertebral translational
2938–45. https://doi.org/10.1016/j.jbiomech.2016.07.009. flexibilities on estimations of trunk muscle forces, kinematics, loads, and stability.
[51] McGill SM, Juker D, Kropf P. Appropriately placed surface EMG electrodes reflect Comput Methods Biomech Biomed Engin 2015;18:1760–7. https://doi.org/
deep muscle activity (psoas, quadratus lumborum, abdominal wall) in the lumbar 10.1080/10255842.2014.961440.
spine. J Biomech 1996;29:1503–7. https://doi.org/10.1016/0021-9290(96) [77] Ignasiak D, Ferguson SJ, Arjmand N. A rigid thorax assumption affects model
84547-7. loading predictions at the upper but not lower lumbar levels. J Biomech 2016;49:
[52] Nachemson A. Lumbar intradiscal pressure. Experimental studies on post-mortem 3074–8. https://doi.org/10.1016/j.jbiomech.2016.07.006.
material. Acta Orthop Scand Suppl 1960;43:1–104. https://doi.org/10.3109/ [78] Lloyd DG, Besier TF. An EMG-driven musculoskeletal model to estimate muscle
ort.1960.31.suppl-43.01. forces and knee joint moments in vivo. J Biomech 2003;36:765–76. https://doi.
[53] Nachemson A. The load on lumbar disks in different positions of the body. Clin org/10.1016/S0021-9290(03)00010-1.
Orthop Relat Res 1966;45:107–22. https://doi.org/10.1097/00003086- [79] Granata KP, Marras WS. An EMG-assisted model of loads on the lumbar spine
196600450-00014. during asymmetric trunk extensions. J Biomech 1993;26:1429–38. https://doi.
[54] Ahern DK, Follick MJ, Council JR, Laser-Wolston N. Reliability of Lumbar org/10.1016/0021-9290(93)90093-T.
Paravertebral EMG Assessment in Chronic Low Back Pain. Arch Phys Med Rehabil [80] Hendershot BD, Wolf EJ. Three-dimensional joint reaction forces and moments at
1986;67:762–5. the low back during over-ground walking in persons with unilateral lower-
[55] Lehman GJ. Clinical Considerations in the Use of Surface Electromyography: Three extremity amputation. Clin Biomech 2014;29:235–42. https://doi.org/10.1016/j.
Experimental Studies. J Manipulative Physiol Ther 2002;25:293–9. https://doi. clinbiomech.2013.12.005.
org/10.1067/mmt.2002.121410. [81] Honegger JD, Actis JA, Gates DH, Silverman AK, Munson AH, Petrella AJ, et al.
[56] Yang JF, Winter DA. Electromyography Reliability in Maximal and Submaximal Development of a multiscale model of the human lumbar spine for investigation of
Isometric Contractions. Arch Phys Med Rehabil 1983;64:417–20. tissue loads in people with and without a transtibial amputation during sit-to-
[57] Bruno AG, Mokhtarzadeh H, Allaire BT, Velie KR, De Paolis Kaluza MC, stand. Biomech Model Mechanobiol 2020. https://doi.org/10.1007/s10237-020-
Anderson DE, et al. Incorporation of CT-based measurements of trunk anatomy into 01389-2.
subject-specific musculoskeletal models of the spine influences vertebral loading [82] Riley PO, Paolini G, Della Croce U, Paylo KW, Kerrigan DC. A kinematic and kinetic
predictions. J Orthop Res 2017;35:2164–73. https://doi.org/10.1002/jor.23524. comparison of overground and treadmill walking in healthy subjects. Gait Posture
2007;26:17–24. https://doi.org/10.1016/j.gaitpost.2006.07.003.