All reviews of published articles are made public. This includes manuscript files, peer review comments, author rebuttals and revised materials. Note: This was optional for articles submitted before 13 February 2023.
Peer reviewers are encouraged (but not required) to provide their names to the authors when submitting their peer review. If they agree to provide their name, then their personal profile page will reflect a public acknowledgment that they performed a review (even if the article is rejected). If the article is accepted, then reviewers who provided their name will be associated with the article itself.
Dr Nitschke, the original Academic Editor is no longer available and so I am making this decision in my capacity as Section Editor.
Thank you for addressing all of the concerns raised by the Reviewers for your manuscript. I am pleased to therefore recommend your amended manuscript for publication.
Thank you for supporting PeerJ and we look forward to future submissions.
Thanks, A/Prof Mike Climstein
no comment
Overall I'm very pleased with the work that the authors have done to address my comments. There are 3 further comments which the authors should consider addressing in the paper.
1. In the response to 1.8 and 1.13 it is mentioned:
"Although the scaling of activation is equivalent to the scaling of maximum isometric force, we are aware that scaling the activation is less common."
Solely scaling the activation is not equivalent to scaling the maximum isometric force: the force-length curves of the tendon and the parallel element are usually constructed so that they scale with the maximum isometric force, but do not scale with activation. By only scaling activation you may have created an MTU that has a CE that can generate 5x the active force, but has a tendon and parallel element that has (1/5) the stiffness that would be expected. For example, applying activation scaling to a typical soleus muscle and Achilles tendon with an isometric strain of 4.9% [1] would result in a soleus muscle with 5x the strength and an Achilles tendon with an isometric strain of around 24.5%. This is extreme: the Achilles tendon has gone from a typical value [1] to being larger than any in-vivo measurements of which I'm aware [2].
Were the force-length curves of the tendon and the parallel element also scaled by 5? Or are the simulated tendons very compliant?
[1] Magnusson, S. P., Aagaard, P., Dyhre-Poulsen, P. & Kjaer, M. Load-displacement properties of the human triceps surae aponeurosis in vivo. The Journal of Physiology 531, 277–288 (2001).
[2] Waugh, C. M., Blazevich, A. J., Fath, F. & Korff, T. Age-related changes in mechanical properties of the Achilles tendon. Journal of anatomy 220, 144–155 (2012).
2. [Optional] In the response to Comment 1.13 and 1.8 the authors helpfully mentioned the range of normalized CE lengths and CE velocities that occurred during the simulations. However, these ranges were not mentioned in the paper. It would be nice to see one more sentence that notes:
"Some of the simulations had wide range of active fiber lengths (0.24 to 1.59 lopt) and velocities (-15.76 to 9.25 lopt/sec) like to due to the speed of the movement and small differences between the model and the participant."
These differences are not uncommon and the field would benefit if there were more publications that were transparent that these kinds of differences can exist when a scaled generic model is used to follow rapid movements.
3. [Optional] In response to comment 1.12 the authors presented CoP profiles but have decided not to include these plots because the profiles can differ between reconstruction methods. This is very interesting and valuable information, and it is a great reason to include these plots in the paper. The COP is a valuable output that many people reading this paper will care out. It is ultimately your choice, but I think the paper is more valuable if the CoP data is included, and the differences between the CoP profiles is addressed in the discussion somewhere.
no comment
no comment
None
None
None
The authors responded to the comments raised by this reviewer well. This reviewer did not have any further comments and appreciated all the sincere responses.
The paper is very well written and organized.
The experiments and simulations are appropriately designed to address the posed research question. The study is rigorous regarding the methodology applied and its description in the manuscript.
The data is made available in an organized and accessible manner, allowing for verification and use in future studies. The paper offers insightful results and discussion on the performance of the marker-tracking optimal control approach when applied to complex 3D musculoskeletal models and activities.
My comments and concerns were fully and properly addressed by the authors.
The reviewers were generally impressed by the fairly novel technical element of the study in performing this variety of data-tracking with a complex model. There were some concerns on the clarity and motivation of the methodological choices and acknowledgement of previous related work that should be addressed.
Please see 'Additional comments' for my full review
Please see 'Additional comments' for my full review
Please see 'Additional comments' for my full review
Summary:
The authors have presented a comparison between 3 different reconstruction methods: inverse methods (inverse kinematics + inverse dynamics), optimal control problem (OCP) that tracks the generalized coordinates (from IK), and an OCP that tracks the markers. The comparison was made to 3 trials of straight running, curved running, and a v-cut trial across 10 participants. Inverse kinematics yielded the most accurate marker tracking, with a close second from the marker tracking OCP, and the worst performance from the generalized coordinate tracking.
This is a very useful paper to have in the literature. I hope that the methods available improve so that more people are able to take advantage of OCP reconstruction methods.
About your reviewer:
I've applied optimal control methods to musculoskeletal models in my own research for the past five years, have authored several muscle models, several foot-ground contact models. Although I have mostly used direct multiple shooting, rather than direct collocation, the methods and numerical issues are similar.
Abstract
Introduction
1. Somewhere in the paragraph that begins on line 88 there should be a reference to bioptim: because its open-source and supports both direct multiple shooting and direct collocation. This sentence would be a good spot for it:
"Furthermore, the increasing availability of toolboxes facilitates access to the methodology (Dembia et al., 2020; Patterson and Rao, 2014)."
Michaud B, Bailly F, Charbonneau E, Ceglia A, Sanchez L, Begon M. Bioptim, a python framework for musculoskeletal optimal control in biomechanics. IEEE Transactions on Systems, Man, and Cybernetics: Systems. 2022 Jun 28.
https://github.com/pyomeca/bioptim
2. Lin et al. is related work that should be mentioned somewhere in the literature review
Lin YC, Walter JP, Pandy MG. Predictive simulations of neuromuscular coordination and joint-contact loading in human gait. Annals of biomedical engineering. 2018 Aug;46(8):1216-27.
3. line 114: add a pair of commas around "for example"
4. line 118: the paragraph that begins "Furthermore, evaluation" belongs with the previous paragraph. The lines from 118-122 are punctuated as a paragraph, but this is not a paragraph: it does not meet the minimum requirements of having a topic sentence, a body sentence, and a sentence linking to the next paragraph.
Methods:
5. line 150: add a pair of commas around "but did not extract"
6. Eqns. 1-5: This is optional, as it is a matter of taste: The very long subscripts (e.g. N_{states}, N_{controls}) are a bit awkward to read. Consider just shortening the subscripts to a single character, for example N_{X} for N_{states}, N_{U} for N_{controls}, N_{\tau} for N_{tor}, etc. Most of these single variables have been defined already, so using them as subscripts is quite clear.
7. line 158-160
It's unclear to me what is meant by:
"In contrast to other musculoskeletal models, runMaD has an adapted pelvis rotation sequence that makes the pelvis orientation interpretable independent of the movement direction."
If the pelvis defined with respect to the inertial frame using a 6 Dof free-flyer joint then the orientation of the pelvis would always be independent of the movement direction (by which I assume you mean CoM velocity). What am I missing?
8. Did the parameters of the model's muscles have to be adjusted in any way to each participant? I see from Nitschke et al. 2020 that the runMaD model makes use of Sam Hamner's musculotendon properties. Having just read the supplementary material for Nitschke et al. 2020 I didn't see any mention of the strength of the modelled muscles being adjusted. I know that Sam made a number of manual adjustments to the model he used (with parameters that originally came from a cadaver) so that it could run at 5 m/s. Was this model strong, fast, and flexible enough to complete the simulated motions without any subject specific adjustment?
9. line 257
The convergence tolerance is listed as 10^{-4}:
a. Please describe exactly how this tolerance is defined.
b. In the results, please provide the norm of the gradient of the Lagrangian so that the reader has some idea of how close the numerical solutions are to a minima.
Results
10. Optional: This is a style point, but the results text begins with a statement that the work you've done is available online. The opening sentence and first paragraph of the results section are usually reserved for the most impactful result of the study. You are doing the paper a disservice by leading with a weak statement. If this is the first time you've heard a comment like this, I suggest you give Brand & Huiskes a read and then consider restructuring the results section.
Brand RA, Huiskes R. Structural outline of an archival paper for the Journal of Biomechanics. Journal of Biomechanics. 2001 Nov 1;34(11):1371-4.
11. Please add a plot or table comparing the residual forces of a typical IK + ID approach to the residual forces that are present in each of the simulations. It is mentioned several times in the paper that dynamic consistency is a benefit of using optimal control for reconstruction, but the benefit is not quantified and presented in the results. I found it strange that dynamic consistency was not reported.
I was expecting that this would be the biggest result of the paper. Why? Residual forces can be huge when using traditional IK + ID, and these forces hang like a dark cloud over any kind of later analysis that depends on forces. In contrast, optimal control based reconstruction should have residual forces that are zero, or close to it, and yet offer a similar level of marker error. This is much more important than a 6-7mm improvement in average marker error. I think the paper could be much more impactful if you reported the difference in residual forces as a result.
12. Figure 3
a. Figure 3 has a wide variety of different scales for each plot which makes it challenging to interpret the plot. To make the plots easier to interpret please: first, indicate in each plot the point in which the biggest pair-wise difference between the methods occurs; second, add a label that denotes the size of the difference. For example ToeL in Z would probably have the largest disagreement at frame 15, and the biggest pair-wise difference would be roughly 60 mm.
b. The last row of plots have y-axis labels of normalized moment but the titles of angle.
c. Please add plots comparing the center-of-pressure location between the measurements and the models. I realize that technically this information is present in ankle, subtalar, and mtp joint moment plots. However, it is a lot easier to interpret a difference in center-of-pressure location than it is to interpret subtalar joint moments, for example. If this doesn't fit, consider breaking up Figure 3 into a kinematics plot, and a separate kinetics plot.
13. Did any of the modelled muscles:
a. Saturate with a maximum excitation?
b. Reach a very short or very long normalized length (<0.6, or >1.5)?
c. Reach a very fast contraction speed (< -7 lopt/s or > 7 lopt/s)
It would be useful to know this information. If the answer is no to all 3 questions, then it would be nice to see a sentence somewhere in the paper stating this fact.
Discussion:
14. Line 324
"In this paper, we showed that it is feasible to reconstruct measured motions directly from marker and GRF data by optimal control simulation of a 3D full-body musculoskeletal model."
This sentence makes it sound like this paper is the first paper to present an optimal control solution in which a 3D musculoskeletal model tracks recorded data. Please qualify this statement to make it clear that other optimal control reconstructions using 3D musculoskeletal models exist (Lin et al. is comparable), but that this is the first to directly track marker data.
Lin YC, Walter JP, Pandy MG. Predictive simulations of neuromuscular coordination and joint-contact loading in human gait. Annals of biomedical engineering. 2018 Aug;46(8):1216-27.
15. Line 382 [Comment - no response required]
"Hence, estimated kinematics should be evaluated with bone markers or medical imaging."
Sadly even bone markers, if you mean bone pins, have drawbacks: skin pulls on the pin and distorts the location of the marker.
16. Line 404 [Comment - no response required]
In the future, marker tracking simulations could also be driven by virtual marker positions extracted from video data, depth images, or radar technology instead of using marker-based optical motion capturing.
Having seen 77 GHz radar captures of people moving indoors, I think it will be many years before tracking can be applied solely to the output of a radar chip.
This manuscript is well-written and easy to read.
Their study rationale needs improvement.
#100-108
This reviewer suggests that including more rationale about tracking marker positions directly is more critical that tracking generalized coordinates for describing human motion using the musculoskeletal model with optimal control theory. The authors stated that the inaccuracies caused by tracking generalized coordinates in an optimization framework could be reduced by directly tracking marker position data. However, marker positions are also prone to factors including skin movement below markers. In the method (#276) and discussion (#380), the authors also mentioned this issue by stating that no ground truth can be compared with the methods used in this study. Thus, readers may be interested in hearing more about what benefit they have when tracking marker positions directly in the optimization framework.
#118-122
This reviewer suggests removing this paragraph because it does not support the study rationale.
#180
Please provide a list of tracking variables (e.g., y included #, what is j, all 42 markers were tracked).
#272
This reviewer would like to see more details about what animations the authors used to evaluate the quality of the optimal solutions. How were they generated?
# 404
This reviewer liked the idea described here. Pose estimation technique has been growing interest in the biomechanics community (e.g., OpenCap suggested by Uhlrich et al. (2022)). This reviewer thought tracking points of interest with the help of musculoskeletal modeling and optimal control could expand their application to real-world problems. Consistent with the earlier comment, the study may be better supported by including the application (tracking marker positions directly) in the Introduction.
#160-161
Please provide more details about the musculoskeletal model (e.g., the number of muscles per leg and arm, degrees of freedom for legs and arms) so that readers do not need to find the model paper. This reviewer does not expect too many details for that model though.
Their finding may be impacted by different weighting values, which it needs to be addressed as a study limitation.
#169, #313, #366, #388
Would authors provide their thoughts on the effects of unequal weighting values for markers and tracking variables on their findings? For example, the authors stated that all markers were weighted equally in inverse kinematics. The observed discrepancies, especially for foot-related makers and angles, may be reduced or better estimated with greater weighting values for those in the inverse kinematics and objective function. If using unequal weighting values would affect their findings, please add this as a study limitation.
This reviewer includes comments for the manuscript in the attached document.
The paper is very well written and organized.
The experiments and simulations are appropriately designed to address the posed research question. The study is rigorous regarding the methodology applied and its description in the manuscript.
The data is made available in an organized and accessible manner, allowing for verification and use in future studies. The paper offers insightful results and discussion on the performance of the marker-tracking optimal control approach when applied to complex 3D musculoskeletal models and activities.
General Comments
The study investigates the feasibility and the performance of a marker-tracking optimal control approach to reconstruct kinematics and kinetics of running and change-of-direction (COD) running using a complex full-body musculoskeletal model. The analysis was performed on 10 participants (3 trials each). The authors present a detailed and careful comparison with the coordinate tracking simulation approach and the more commonly used inverse kinematics/dynamics approach.
Although marker tracking is not new, previous studies have used either models with few degrees of freedom or torque-driven models. This study investigates the performance of marker tracking for a complex, full-body musculoskeletal model, offering a straightforward and high-quality assessment of performance compared to the other two methods commonly used in the literature. The manuscript is very well written and offers many insights on the advantages and disadvantages of using a marker-based tracking optimal control approach, also regarding computational costs. Therefore, although the methodology is not new, the quality and completeness of the study and its application to a complex model and a complex activity (COD running) substantially contribute to the body of knowledge in the area. All the data is made available in a well-organized manner.
Specific Comments
- Abstract: authors mention that marker tracking is recommended when precise marker tracking is required. It would be essential to elaborate on this in the discussion session. In which instances would accurate marker tracking be required?
- Line 97 - the isolated claim that the coordinate tracking is computationally efficient because generalized coordinates are used as kinematic states is misleading. I suggest authors rephrase or displace this sentence to put it in a comparative context with the marker tracking approach.
- Lines 99 to 101 – the authors could be more precise when discussing error propagation. All methods are ultimately tracking markers. The difference is how and to which extent the marker trajectories' inconsistencies are propagated. Although the discussion is very well-written, the authors could shed more light on the problem from this perspective.
- In equations (1) to (4), I assume the denominator should be "N+N_{tra,mus,tor}" and not "N*N_{tra,mus,tor}". Please, check.
- Lines 193 to 201 - please, provide a reference or a rationale to justify the sum of volume-weighted cubed neural excitations to solve the muscle redundancy problem instead of other, more commonly used cost functions.
- Lines 208-230: perhaps incorporating initial states as additional optimization parameters could address this issue more straightforwardly. Have authors tried this? Authors don't need to perform any changes to the utilized approach, as this comment is meant to be a mere contribution to the discussion.
- Lines 328: please, mention in which sense marker tracking is comparable to inverse methods.
All text and materials provided via this peer-review history page are made available under a Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.