Reconstructing Nerve Structures from Unorganized Points

Round 1

Reviewer 1 Report

The paper's primary goal is to present a 2D contour reconstruction algorithm tailored for unorganized point datasets, specifically for use in neurostimulation simulations. This algorithm is required to handle intricate cross-sections featuring sharp corners, multiple close curves, and other complex geometries.

Overall, the paper is well-structured and easy to follow. The related work section provides a detailed overview of the current state of research.

One major recommendation pertains to the results section. Currently, the results are qualitative and evaluated visually. However, for a more solid research paper, it is essential to introduce a quantitative evaluation. This could involve metrics such as Root Mean Square Error (RMSE) to quantify the discrepancy between your results and the ground truth.

Furthermore, the comparative analysis with existing methods should be more comprehensive. Instead of relying on just two examples to showcase the superiority of your approach over others, consider conducting a more extensive comparison involving multiple methods. Given that you've listed several potential methods in the related work section, you should compare your method with all the potential methods. Otherwise, please explain why you chose the two competitive algorithms to evaluate.

Specific comments

1.       Abstract: Overall, the abstract needs more connections between the sentences for better reader clarity. In Line 2, between the first and second sentences, add one more sentence about why it helps patients and show the impact of stimulation of residual peripheral nerves. Still here, add another sentence to link the first and the second sentence since you jumped from direct electrical stimulation to neurostimulation simulation. Try to use 1-2 sentences here to answer why we can’t use direct electrical stimulation and why we need to simulate neurostimulation. These adjustments will create a more natural and coherent flow in the abstract.

2.       P1, L6: You mentioned in the methods and discussion sections that there are still some parameters that are sensitive and may need tuning. I don’t think “parameter-free” is accurate here.

3.       P1, L7: “Few methods” means there are still some methods that can handle the hard case you talked about. Please briefly emphasize the unsolved gap for these methods.

4.       P1, L8: “successfully reconstruct multiple open and/or closed lines with sharp corners” Please briefly emphasize those are hard cases for algorithms since they are very easy for humans so readers may underestimate the complexity.

5.       P2, L41-43. Add brief explanations of how the modeling is going to fix the electrode position and stimulation protocol challenges. How can it reduce the gap between current technology and real human applications? If real human applications can never be achieved, then the simulation is not very useful.

6.       P2, L51. “unorganized set of points”: Please add connections to nerve structure modeling. Does the modeling output unorganized points? Can it be organized to solve the problem from the source?

7.       P3, L107-108. Any limitation for this method? It sounds like this method can solve your problem.

8.       P3, L113-114. Same here. Any limitation for this method? Readers may think that this method is good enough and why they need to read your method. Try to emphasize the limitation/gap of the existing methods.

9.       P4, L145-146. So only importing to Matlab will have this problem? I am trying to understand whether it is an unavoidable problem for simulation or not.

10.   Equation 5: Where is “n” coming from?

11.   Result: More quantitative metrics and solid comparisons with others are needed. I explained it above at the beginning.

Author Response

Comments and Suggestions for Authors

The paper's primary goal is to present a 2D contour reconstruction algorithm tailored for unorganized point datasets, specifically for use in neurostimulation simulations. This algorithm is required to handle intricate cross-sections featuring sharp corners, multiple close curves, and other complex geometries.

Overall, the paper is well-structured and easy to follow. The related work section provides a detailed overview of the current state of research.

Thank you for your positive feedback regarding the structure and the overview section.

One major recommendation pertains to the results section. Currently, the results are qualitative and evaluated visually. However, for a more solid research paper, it is essential to introduce a quantitative evaluation. This could involve metrics such as Root Mean Square Error (RMSE) to quantify the discrepancy between your results and the ground truth.

We appreciate your suggestion. To make the results more quantitative, we have added a more statistical comparison to the manuscript on pages 15 and 16 – lines 353-370, along with Table 1. The revisions made to the manuscript are indicated in yellow highlighting.

Furthermore, the comparative analysis with existing methods should be more comprehensive. Instead of relying on just two examples to showcase the superiority of your approach over others, consider conducting a more extensive comparison involving multiple methods. Given that you've listed several potential methods in the related work section, you should compare your method with all the potential methods. Otherwise, please explain why you chose the two competitive algorithms to evaluate.

Thank you for this feedback. We have addressed this particular concern in response to comments 7 and 8, where we elaborated on our rationale for selecting the comparison algorithms and clarified additional manuscript improvements. 

Specific comments

1. 1.       Abstract: Overall, the abstract needs more connections between the sentences for better reader clarity. In Line 2, between the first and second sentences, add one more sentence about why it helps patients and show the impact of stimulation of residual peripheral nerves. Still here, add another sentence to link the first and the second sentence since you jumped from direct electrical stimulation to neurostimulation simulation. Try to use 1-2 sentences here to answer why we can’t use direct electrical stimulation and why we need to simulate neurostimulation. These adjustments will create a more natural and coherent flow in the abstract.

We acknowledge and appreciate all your suggestions. In response, we have expanded upon the abstract to provide the reader with a more precise understanding of the original purpose of this algorithm.

2. P1, L6: You mentioned in the methods and discussion sections that there are still some parameters that are sensitive and may need tuning. I don’t think “parameter-free” is accurate here.

We understand the concern that was pointed out here. Still, the term 'parameter-free' has been employed to emphasize that, while the algorithm does possess parameters, there is no need for continual adjustments or manual tuning of these parameters once the initial configuration or initialization is performed. This descriptor underscores the algorithm's ability to automatically adapt its behavior and the fact that, after the initial steps, there is no additional input required from the user.

3. P1, L7: “Few methods” means there are still some methods that can handle the hard case you talked about. Please briefly emphasize the unsolved gap for these methods. 

The problems encountered by the reconstruction methods are outlined in the sentence starting with "Few methods," as well as in the following sentence. This second sentence specifically delineates the challenge that, in cases involving the contours used in this study, proves to be difficult even for the best performing algorithms. We have rephrased both sentences to improve clarity and highlight the main advantage our algorithm offers compared to state-of-the-art algorithms.

4. P1, L8: “successfully reconstruct multiple open and/or closed lines with sharp corners” Please briefly emphasize those are hard cases for algorithms since they are very easy for humans so readers may underestimate the complexity. 

Thank you for your suggestion. We have incorporated an explanation regarding how algorithms perceive tasks of this complexity.

5. P2, L41-43. Add brief explanations of how the modeling is going to fix the electrode position and stimulation protocol challenges. How can it reduce the gap between current technology and real human applications? If real human applications can never be achieved, then the simulation is not very useful. 

While real human application, i.e. generating physiologically relevant sensations in humans, has been demonstrated (as referenced on page 2, line 43), there is still a long road ahead before this technology becomes a part of daily life and widely accessible. The primary advantage of modeling is its contribution to the trial phase, specifically in planning the selection of electrode types, their positions, quantities, and the overall stimulation strategy, which can significantly improve the development process.

Furthermore, this technology holds potential applications extending beyond the realm of neuroprosthetics, encompassing modeling, simulations, and ultimately, facilitating virtual surgeries and similar medical tasks.

6. P2, L51. “unorganized set of points”: Please add connections to nerve structure modeling. Does the modeling output unorganized points? Can it be organized to solve the problem from the source? 

The mentioned section of the manuscript serves as an introduction to the broader issue of curve reconstruction from unorganized points, where we present the overall problem and highlight the most prominent approaches. The issue we address falls under this category, but it is introduced later, in Section Motivation of This Study. The answer to the second part of this comment is closely related to Comment 9, where we explain the stage at which the problem arises and why it was an integral step of a larger, intricate framework called ProprioStim. 

7. P3, L107-108. Any limitation for this method? It sounds like this method can solve your problem. 

Even though Funke and Ramos show that their method is capable of reconstructing both open and closed curves with sharp corners, according to the authors, the mentioned algorithm implies specific sampling conditions. This is also the reason why this method is not often mentioned or subject to in-depth analysis, as observed in the review paper by Ohrhallinger from 2021 (referenced in our paper) or other relater papers. Additionally, its numerical complexity and the unavailability of open-source code are also the reasons why others (e.g. the paper presenting Vicur algorithm) or we did not analyze this method further. We inserted a brief explanation on why the method is not more widely used on the page 3 in the lines 13-15. We thank you for your observation.

8. P3, L113-114. Same here. Any limitation for this method? Readers may think that this method is good enough and why they need to read your method. Try to emphasize the limitation/gap of the existing methods.

The reason we employed the Vicur algorithm, but not the Discur algorithm, is that the former method is, to a significant extent, an improved version of the Discur algorithm, as it was developed by two authors who had previously contributed to Discur. While recognizing that even the more advanced of these two algorithms encounters challenges in reconstructing realistic nerve contours in specific cases, we believed it was appropriate to concentrate on and present the results of the Vicur algorithm in our findings. In our Results and Discussion section, from the line 341-347,  on page 14 we emphasized the reasons for selecting the comparison algorithms.

9. P4, L145-146. So only importing to Matlab will have this problem? I am trying to understand whether it is an unavoidable problem for simulation or not.

The issue of concern arose during the data transfer process from Comsol to Matlab, where it became challenging to preserve the order of points defining the contour's cross-section upon export from Comsol to Matlab. The ProprioStim framework, which served as the inspiration for the presented algorithm, has a strong dependence on Comsol Multiphysics for modeling and simulating physical processes, as well as on Matlab programming, which plays a significant role in various stages of the pipeline. This situation prompted the authors addressing this specific challenge to seek solutions to overcome this obstacle. Nonetheless, we firmly believe that similar challenges may arise in numerous other fields and scenarios, where our proposed method could also prove valuable.

10. Equation 5: Where is “n” coming from?

The parameter “n” from Equation 5 is obtained through the expansion of the expression M(X) using a Taylor series, as detailed in the book “Introduction to statistical pattern recognition” by K. Fukunaga, published by Elsevier in 2013, in subchapter 11.2 Nonparametric Clustering. It reflects the dimensionality of the input vector X. The reason we didn't provide further elaboration on this parameter is because, for all the analyzed points surrounding point X, the first part of the expression remains constant, hence it has no bearing on the selection of the successor.

We have refined the exact definition of this variable within the text, specifically in 187 on page 6, and we have also provided clarification to ensure that it is apparent that the influence of the initially mentioned variable is negligible, leaving only the expression cosθxy at play in determining the successor point (lines 195-199 on page 7). 

11. Result: More quantitative metrics and solid comparisons with others are needed. I explained it above at the beginning.

Thank you for your feedback. As mentioned earlier in our response, we have taken your valuable input into consideration. We have incorporated additional quantitative metrics in the Results and Discussion section and conducted solid comparisons with other relevant methods to address your concerns. We believe these enhancements will provide a more comprehensive and detailed evaluation of our work.

Reviewer 2 Report

results and discussion part need to be elaborated. enough explanation is not given to understand and justify the novelty of your research.

use grammarly tool to get complete corrections suggested.

Author Response

Comments and Suggestions for Authors

results and discussion part need to be elaborated. enough explanation is not given to understand and justify the novelty of your research. 

Thank you for your feedback. We have taken your suggestions into account and made several improvements. Specifically, we have introduced additional quantitative metrics in the Results and Discussion section and conducted thorough comparisons with other relevant methods to address your concerns. Moreover, we have elaborated on the key distinctions and the novelty of our research in the Conclusion section to provide a clearer understanding of how our proposed approach improves upon existing methods. 

Comments on the Quality of English Language

use grammarly tool to get complete corrections suggested.

We thank you for the suggestions regarding the English language. We have made language improvements throughout the entire manuscript to rectify grammatical mistakes and improve the flow of the text.

Reviewer 3 Report

This manuscript addresses an important problem and the proposed methodology shows promise. However, there are some key issues that need to be addressed before this manuscript can be considered for publication.

Major comments:

The literature review and motivation sections could be strengthened. While many related techniques are discussed, a more thorough review of the state-of-the-art is needed. In particular, how the proposed approach improves upon or addresses limitations of existing methods needs clearer explanation.

The algorithm description lacks important details needed for replication. Pseudocode should be provided to clearly outline the steps. Key aspects like parameter selection and initialization processes require more details.

Experimental validation is limited. Quantitative comparisons to other methods on synthetic and/or real datasets are needed to demonstrate effectiveness. Error metrics, sensitivity analyses, and limitations of the approach should be characterized.

The potential applications and significance of this work could be better conveyed. While neuroprosthetics is discussed as motivation, more context is needed on how this technique may impact modeling, simulations, and device development.

The conclusions overstate what is shown and several claims require further support. The ability to handle various curve types and proximity challenges requires demonstration rather than assertion.

• The English language used is generally good but could benefit from minor edits in some sections for improved clarity and readability.

• A few sentences are quite long and complex, making the ideas difficult to follow. Breaking these up into simpler structures would aid comprehension.

• Some key technical terms are not defined on first use, such as "fascicles" and "FEM". Providing the full terminology would help general readers.

• Consistent terminology would improve the flow (e.g. sometimes "points", sometimes "vertices").

• A handful of grammatical errors were noticed, such as missing articles, incorrect verb forms. Proofreading is recommended.

• The large number of acronyms and abbreviations, like "NN" and "TSP", make certain parts difficult to parse for those unfamiliar. Spelling them out initially would help.

• Figure and equation labels could be presented more clearly for reader reference throughout the text.

Author Response

Comments and Suggestions for Authors

This manuscript addresses an important problem and the proposed methodology shows promise. However, there are some key issues that need to be addressed before this manuscript can be considered for publication.

Major comments:

The literature review and motivation sections could be strengthened. While many related techniques are discussed, a more thorough review of the state-of-the-art is needed. In particular, how the proposed approach improves upon or addresses limitations of existing methods needs clearer explanation.

We thank you for your valuable insights. To provide a clearer understanding of how our proposed approach improves upon existing methods, we have emphasized key distinctions in the Conclusion section of our paper. We have emphasized the distinctive feature of our algorithm - that, unlike many existing methods that primarily consider geometry, our approach integrates dynamics and a fascicle model through an adaptive Kalman filter. Specifically, our approach assumes that most curve paths are smooth and slightly convex, treating sharp bends or angles as specific occurrences. This distinction is made possible through the expansion of the Q matrix in the adaptive Kalman filter, which deviates from the standard model.

Additionally, The Results and Discussion section now provides quantitative insights that highlight the distinctive aspects of our algorithm, setting it apart from existing models.

The algorithm description lacks important details needed for replication. Pseudocode should be provided to clearly outline the steps. Key aspects like parameter selection and initialization processes require more details.

Thank you for your feedback regarding the algorithm description and parameter selection. To enhance the clarity and replicability of our approach, we have included pseudocode in the appendix. This pseudocode provides a comprehensive outline of the algorithm's steps, making it easier for readers to understand and replicate our method. We'd like to emphasize that this pseudocode is closely related to the flowchart presented in Figure 8, which visually represents the algorithm's workflow.

Additionally, we have elaborated on the nature and selection of the parameters in the Results and Discussion section of our paper (lines 393-415 on pages 16 and 17). We believe that these added details provide a much better understanding of our parameter choices and their significance in our approach. These revisions are indicated in yellow highlighting.

We hope that the inclusion of pseudocode and the expanded parameter descriptions in the appendix and Results section, along with the visual representation in Figure 8, will assist readers in replicating our work and gaining insights into the intricacies of our algorithm.

Experimental validation is limited. Quantitative comparisons to other methods on synthetic and/or real datasets are needed to demonstrate effectiveness. Error metrics, sensitivity analyses, and limitations of the approach should be characterized.

Thank you for your valuable feedback regarding the experimental validation of our approach. We have made several additions to address these concerns:

Quantitative Comparisons: In the Results and Discussion section, specifically on lines 353 to 370 on pages 15 and 16, we have included quantitative comparisons with other methods. These comparisons are also summarized in Table 1, providing readers with a clear overview of our approach's effectiveness compared to existing methods.

Limitations of the Approach: We have elaborated on the limitations of our approach in subsequent paragraphs, providing readers with a comprehensive understanding of the boundaries and constraints of our method.

We believe that these enhancements strengthen the validity and clarity of our approach.

The potential applications and significance of this work could be better conveyed. While neuroprosthetics is discussed as motivation, more context is needed on how this technique may impact modeling, simulations, and device development.

In the Conclusions section, we demonstrated that our proposed method presents a novel approach to curve reconstruction, with a particular focus on intricate nerve contours. By comparing our approach to existing state-of-the-art methods, we have highlighted instances where our method exhibits superior performance. We believe these findings underscore the significance of our work in the context of neuroprosthetics and beyond.

Our method offers precise reconstruction of complex structures, improving simulation accuracy and aiding in the advancement of neuroprosthetic devices. Additionally, it has potential applications beyond neuroprosthetics, including modeling, simulations, and, in result, virtual surgeries and similar tasks in the medical fields. We hope these additions provide a clearer understanding of the novelty and advantages of our proposed approach. 

The conclusions overstate what is shown and several claims require further support. The ability to handle various curve types and proximity challenges requires demonstration rather than assertion.

Thank you for your comments and suggestions. We have added statistical analysis in the Results and Discussion section to better demonstrate our algorithm's performance relative to other methods and compared to the expert knowledge. 

Comments on the Quality of English Language

• The English language used is generally good but could benefit from minor edits in some sections for improved clarity and readability.

• A few sentences are quite long and complex, making the ideas difficult to follow. Breaking these up into simpler structures would aid comprehension.

• Some key technical terms are not defined on first use, such as "fascicles" and "FEM". Providing the full terminology would help general readers.

• Consistent terminology would improve the flow (e.g. sometimes "points", sometimes "vertices").

• A handful of grammatical errors were noticed, such as missing articles, incorrect verb forms. Proofreading is recommended.

• The large number of acronyms and abbreviations, like "NN" and "TSP", make certain parts difficult to parse for those unfamiliar. Spelling them out initially would help.
• Figure and equation labels could be presented more clearly for reader reference throughout the text.

We would like to express our gratitude for your valuable feedback and constructive comments regarding the quality of writing and the English language in our paper. We acknowledge the need for improvements in certain sections to enhance clarity and readability. Based on your feedback, we have made extensive language and style enhancements throughout the paper.

Round 2

Reviewer 1 Report

I can see all my previous questions are well addressed. Thank authors for their effort on improving the paper quality. I have no more questions.

Reviewer 3 Report

English language fine. No issues detected.