Feature-Based Modeling of Subject-Specific Lower Limb Skeletons from Medical Images

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The manuscript presents an interesting method for generating skeletal shape models of distal femurs. While the approach is promising for its potential to lower costs, the reviewer raised concerns about the methodological rigor, error analysis, and clinical applicability, which need to be addressed in a major revision.

The maximum error of 12.88 mm is substantial and unfeasible in clinical settings like implant positioning or surgical planning. Discuss where the errors occur.

The claim of "sufficient accuracy" needs to be better substantiated in the context of this large maximum error. What level of accuracy is "sufficient" for the intended applications?

The manuscript states the base model had 1419 nodes and an average element size of 4.56 mm. How was this specific base model selected? Is it an "average" femur, or a model of a specific individual? The characteristics of the base model could significantly influence the deformation outcome.

Depth Coordinate Assumption for Scanning Features: For scanning feature values obtained from 2D X-ray images, "The depth coordinates of the feature points obtained from the medical images were defined as the same as the coordinates of the relatively co-located points in the model." This is a critical step and needs a much more detailed explanation. How is "relatively co-located" determined? This assumption could introduce significant errors, especially if the base model's depth profile differs substantially from the target patient's.

Exclusion of Bone Axis: The study excluded the bone axis from evaluation because it "did not perform active deformation near the bone axis." How was the "bone axis" region defined and excluded? Could deformation errors propagate into this region?

Why was a standard, well-established surface distance metric (e.g., Hausdorff distance, mean absolute surface distance) not used or compared against? The custom nature of the metric makes comparison with other studies more difficult.

Author Response

Thank you for your polite comments. Our replies to your points are given below. Your comments are in italics. Comments are numbered from 1-1) to 1-7). In the manuscript, revised text is marked in red.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

I read the manuscript entitled “Feature-Based Modeling of Subject-Specific Lower Limb Skeletons from Medical Images” with great interest. This manuscript presents an interesting approach for creating subject-specific 3D skeletal models of the distal femur using only two X-ray images and free-form deformation (FFD). The work addresses an important clinical need by reducing the imaging burden compared to traditional CT/MRI-based methods. The following are my comments:

1. The discussion should better address the clinical significance of the reported errors. Is a mean error of 1.54 mm clinically acceptable for intended applications? In the conclusion, the authors acknowledge that image rotation may cause some errors and that a solution has not yet been determined. However, this is a significant methodological limitation that deserves deeper discussion. Please elaborate on how sensitive the method is to rotational misalignment, how much error it may introduce, and whether this limits the clinical or practical applicability of the approach. Ideally, provide suggestions for mitigating or quantifying rotational effects, or outline future work aimed at solving this challenge.
2. There appears to be duplicated content between Section 2.2 (Adjustment of tilt between medical images and model) and Section 2.3 (Acquisition of feature points and feature values from medical images and shape model), where identical text about the selection of femoral condyle points and tilt adjustment is repeated.
3. In Section 2.3.1, the authors state that they followed the method used by Mohammadi et al. [15] to obtain feature values from tomographic images. While the citation is provided, it would greatly improve clarity and self-containment if the manuscript briefly introduced or summarized the key aspects of Mohammadi et al.’s method, rather than simply referencing it. Readers should not be required to look up external papers to understand the main steps or logic behind the approach applied here. Please provide a short summary of the method and explain how it was adapted in this study.
4. Tables could benefit from better formatting and clearer headers.
5. In Table 2, the large differences between anatomical and scanning methods (39.56 mm vs 10.14 mm maximum errors) need better explanation and discussion of the underlying causes.
6. In Section 2.5, the choice of weight w=200 appears arbitrary. Please provide justification or conduct sensitivity analysis for this parameter selection. How does varying this weight affect the deformation results. 
7. in section 2.6, the evaluation procedure (steps 1-7) is confusing and needs clearer explanation. The threshold selection process and distance calculation methodology require more detailed description for reproducibility. Consider adding a flowchart summarizing the complete methodology.

Comments on the Quality of English Language

1. Line 100: “…in addition to S-ray images…”? Should be “X-ray images.”
2. Line 124 / 133: There’s a repeated header “2.2.2. Please check section numbering.
3. Several technical terms (e.g., “FFD,” “confidence region method”) are used without definition on first mention. While they are likely familiar to experts, a brief definition or reference might help broader readers.
4. Section numbering seems inconsistent in some parts; please check the numbering and subheading order (especially 2.2.2. Muscle force optimization 124 and knee contact mechanics computation in repeated content of Section 2.2 and section 2.3).

Author Response

Thank you for your polite comments. Our replies to your points are given below. Your comments are in italics. Comments are numbered from 2-1) to 2-8). In the manuscript, revised text is marked in red.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

This is an interesting and relevant study that tackles a practical problem in the field of personalized biomechanics: reducing the dependency on CT and MRI for constructing subject-specific skeletal models. The authors propose a promising method that uses feature points from simple X-ray images combined with free-form deformation (FFD) techniques applied to a base model. It’s a well-motivated idea that could potentially lower both the cost and clinical burden associated with detailed anatomical modeling.

That said, while the concept is solid and the technical approach seems sound, there are a number of issues — both methodological and presentation-related — that should be addressed before this work is ready for publication.

Major Points

1. Limited Sample Size and Validation

   The method was tested on only 13 cases, which feels quite limited considering the variability in human anatomy. Without a larger and more diverse sample, it’s difficult to judge how well this approach would generalize across different patients. Also, there’s no statistical analysis reported to contextualize the mean and maximum errors. Including confidence intervals or hypothesis testing would strengthen the claims of accuracy.

2. Maximum Error is Quite High

   The reported maximum error of 12.88 mm is significant — particularly for applications in joint mechanics where small discrepancies can have meaningful effects. It would be helpful if the authors could discuss what is considered clinically acceptable error for these types of models, or at least compare their results directly to existing alternative methods to give readers a sense of perspective.

3. Manual Feature Point Acquisition

   While this method reduces imaging demands, it still relies on manually selecting feature points from X-ray images. This process is inherently prone to user variability and could limit the scalability of the method in a clinical setting. I would recommend including a reproducibility study or at least discussing potential inter-operator differences. Better yet, if there’s a way to automate or semi-automate this step, it would make the method much more appealing.

4. Use of CT Images for Validation

   A notable limitation is that while the method is designed to work from X-ray images, the validation was performed using CT images as the reference. This somewhat weakens the argument for reducing dependency on high-cost imaging. Ideally, the method should also be validated using real-world clinical X-rays to confirm it performs well with the type of data it’s meant to rely on.

5. No Practical Application Example

   The paper suggests that these models could be used in biomechanical analyses or personalized medicine, but no actual application or case study is shown. Even a small demonstration — like a basic joint range-of-motion simulation or load estimation — would help prove that the models are usable for these purposes.

Minor Suggestions

• The literature review could be strengthened by referencing recent deep learning-based methods for 3D reconstruction from 2D images, which are becoming increasingly relevant in this field.

• The distinction between “feature points” and “feature values” isn’t always clear in the text. It might help to define these explicitly in the Methods section.

• It would be helpful to include demographic details (age, sex, pathology status) for the patient data used in the study to give readers a better sense of the population sampled.

• The figures (Figs. 1–5) were referenced but not visible in the version I reviewed. I would recommend ensuring all figures are included in the final submission as they seem important for understanding the workflow.

Conclusion

In summary, this paper addresses a meaningful problem with a creative and practical approach. However, it would benefit from a more robust validation, clearer reporting, and ideally a small demonstration of its practical utility. I believe this work has potential, but it needs some further refinement before it’s ready for publication.

Author Response

Thank you for your polite comments. Our replies to your points are given below. Your comments are in italics. Comments are numbered from 3-1) to 3-8). In the manuscript, revised text is marked in red.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

Thank you for providing the responses to my initial review and for the corresponding revisions to the manuscript. I appreciate the effort the authors have made to address my comments.

While the revisions have clarified some aspects of the work, several of the most critical scientific and methodological issues have not yet been fully resolved. In its current form, the manuscript would benefit significantly from another round of major revisions to strengthen its arguments and methodological rigour.

My detailed evaluation of each response is below.

Regarding Comment 1-1 (Maximum Error):

Thank you for the response. However, the explanation provided could be strengthened with a more quantitative analysis. The original request was for a detailed analysis of where the large (12.88 mm) errors occur, and the response remains qualitative (e.g., "regions where the femoral contour changes abruptly"). A more in-depth, spatial analysis of the error distribution would be more compelling.

Furthermore, the comparison to the accuracy reported by Laporte et al. (2003) is not fully convincing. The cited study reported a maximum error of approximately 5 mm, which is less than half of the maximum error reported in this manuscript. Citing a study with substantially better accuracy does not, by itself, render the current results "acceptable." Instead, it highlights the need for a more thorough justification of why the present method's larger error margin is permissible for the intended application. The revision to the manuscript reiterates this comparison without providing new data or a deeper analysis to support it.

Regarding Comment 1-2 ('Sufficient Accuracy'):

I commend the authors for scoping their claim to specify that the method's accuracy is intended for biomechanical modelling rather than high-precision clinical tasks. This is a positive step.

However, the assertion that a potential 12.88 mm error on a condylar surface is "acceptable for applications in biomechanical modelling and motion analysis" is a strong claim that requires substantiation. An error of this magnitude could significantly alter calculations of joint contact mechanics, ligament tension, and kinematics. The manuscript would be greatly strengthened by providing evidence or citations from the biomechanical modelling literature to support the claim that this level of inaccuracy is tolerated within the field.

Regarding Comment 1-5 (Depth Coordinate Assumption):

Thank you for the detailed explanation of how depth is estimated by being inherited from the base model's contour geometry. The clarification is helpful. However, the method itself appears to be a significant, unquantified source of potential error, as the 3D reconstruction is constrained by the depth profile of a single, unrelated base model. While the authors acknowledge this limitation in their revision, they have not yet attempted to quantify its impact on the final model's accuracy. This is a key methodological point that, if addressed, would substantially increase confidence in the results.

Regarding Comment 1-6 (Exclusion of Bone Axis):

Further clarification is needed regarding the exclusion of the bone axis region during evaluation. The response defines the bone axis as a vector, which does not directly explain how a region of the 3D mesh was defined and subsequently excluded from the analysis. The second part of the original question, regarding the potential for error propagation into this undefined region, remains unaddressed. An addition to the manuscript clarifying the precise definition of this exclusion zone would be helpful for the reader.

Regarding Comment 1-7 (Evaluation Metric):

The justification for using a custom metric while avoiding standard metrics like the Hausdorff distance could be more persuasive. It is unclear why these standard metrics cannot be computed at the barycentres. The rationale that converting the shell-element model to a point cloud is "outside the scope of the current evaluation framework" seems to be a self-imposed limitation. Interpolation between barycentres and vertices is a standard procedure in 3D model analysis and should be readily achievable with any general-purpose software (MATLAB).

By not employing established metrics, the authors make it difficult for readers to compare this work with the vast body of literature in the field. Adopting standard evaluation practices would help to situate the work more clearly and highlight its specific contributions.

 

Author Response

Thank you for your polite comments. Our replies to your points are given below. Your comments are in italics. Comments are numbered from 1-1) to 1-5). In the manuscript, revised text is marked in red.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

The authors have thoroughly addressed my comments in both the response letter and the revised manuscript. Well done!

Author Response

I sincerely appreciate the reviewer’s time and effort in providing constructive feedback on my work.

Reviewer 3 Report

Comments and Suggestions for Authors

Dear Authors,

Tganks for your efforts. You have edited my revisions completely. Please just recheck the citations, and do grammar check.

Regards,

pourya

 

 

 

 

Comments on the Quality of English Language

It is ok but please recheck it.

Author Response

I sincerely appreciate the reviewer’s time and effort in providing constructive feedback on my work.