New Vessel Extraction Method by Using Skew Normal Distribution for MRA Images

Round 1

Reviewer 1 Report (Previous Reviewer 1)

Line 438: "Luca and Loperfido" should be "De Luca and Loperfido"

Author Response

We have corrected it (in the new version of paper).

Reviewer 2 Report (New Reviewer)

Dear Authors, 

I have read your paper with great interest and recommend you to consider some latest articles and perform more experiments as well as improve the artilce in terms of write-up.

1) Abstract should be based on the Introduction, Motivation, Objective, Results, and Conclusion without the heading but the paragraph should reflect these things. 

2) Better to add the Literature Summary Table at the end of the Introduction Section. 

3) The equations need to well explained. 

4) Must refer the equations, figures and tables in the text. 

5) Abbrevations should be atleast in the full form at first time. 

6) Consider some latest article to improve your articles throughout 

https://aapm.onlinelibrary.wiley.com/doi/full/10.1002/mp.16720?casa_token=w_r3ARVUKjsAAAAA%3Altj4YVDMP6fQVSU1dMM4Sm9gAJvZehSP2C2AS-NzZ62tLFbdxVOFo_mp5CYU1-l23WTLpODxdpizZWHh

https://dl.acm.org/doi/abs/10.1145/3581783.3611718?casa_token=n_WNw92bxPYAAAAA:uVZjKCjYJ0oMZ_oXHIYxRsVfiunPoBnOisPl5N21ZUm78g70k-e47XLAluafn7klG6bG-sNz41EThw

https://link.springer.com/chapter/10.1007/978-3-031-36027-5_20

7) Mention "Article" as paper type.

8) Better to mention the major contributions into built-points after the literature review so that gap should be clear as well as contributions. 

9) Ignore to bold the words if not necessary. 

10) What about to consider other clustering algorithms i.e., Hierarchical-based, centroid based.

Overall paper is weak in terms of write-up and mainly experiments are not sufficient. 

Need to improve

Author Response

Pl find the attached pdf file

Author Response File: Author Response.pdf

Reviewer 3 Report (New Reviewer)

Manuscript Title “New vessel extraction method by using skew normal distribution for MRA images”

General comment:

The structure of this manuscript needs to largely revise for fulfilling the criteria of journal article. The technical term of this manuscript is too few to imply the all idea.

Specific comment:

1.      Abstract; read like part of the introduction and no solid quantified data at all. A good abstract should include definition of keywords, experimental setup in short, result and essential discussions as well.

2.      Introduction; too long the description to imply a simply introduction. A good introduction should contain background review, specific rationale study and the topic of this study.

3.      EM algorithm; Skew normal distribution; statements are clearly described.

4.      Numerical example; this is important to let the reader realize how the assigned algorithm to be explored in this study, the biased weighting intensify the original imaging with various demanded intention.

5.      Conclusion, there is no chapter of discussion and directly jump to this section. It is better to increase the chapter of discussion and elaborate the description of agreement or disagreement among various numerical analysis and give a short and solid statements as strong conclusion in the final end, rather than the format style as it is right now.

Author Response

Pl find the attached pdf file

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report (New Reviewer)

I am satisfied with changes made by authors

Author Response

Reviewer is satisfied with changes made by authors. Pl see below reviewer's coemmnts

Reviewer 3 Report (New Reviewer)

Manuscript Title “New vessel extraction method by using skew normal distribution for MRA images”

General comment:

The quality of revised version is still far away to be accepted. The structure is even worse than the last version although he really adds a new section of short discussion.

Specific comment:

1.      Abstract; the revised one is still lack of solid result or any quantified information.

2.      Introduction; this is not technical report so it is inappropriate arrangement to list a table of paper review, instead, should give a concise summary of integrated rationale study. furthermore, the introduction become even longer than the last one.

3.      EM algorithm; the newly added definition of weighting factor is OK.

4.      Discussion; too short and concise to interpret the variation among various algorithm, plus no quantified information or solid elaboration according to the derived simulation

5.      Conclusion, too long to imply a strong conclusion. The optimal suggestion of a concise conclusion should be among150-200 words, and no reference should be cited in the conclusion. All the correlated discussion should be held in the chapter of discussion.

Author Response

Comments of Review Report 3

• Abstract; the revised one is still lack of solid result or any quantified information.

Response: We changed and improved the Abstract part.

• Introduction; this is not technical report so it is inappropriate arrangement to list a table of paper review, instead, should give a concise summary of integrated rationale study. furthermore, the introduction become even longer than the last one.

Response: We changed it.

• EM algorithm; the newly added definition of weighting factor is OK.

Response: Thank you.

• Discussion; too short and concise to interpret the variation among various algorithm, plus no quantified information or solid elaboration according to the derived simulation
• Conclusion, too long to imply a strong conclusion. The optimal suggestion of a concise conclusion should be among150-200 words, and no reference should be cited in the conclusion. All the correlated discussion should be held in the chapter of discussion.

Response: We wrote discussion and conclusion in one section.

We highlighted the corrections in the paper with yellow color and are thankful to the reviewer for his/her valuable comments and suggestions. These inputs greatly contributed to improving the presentation and quality of the paper. 

Author Response File: Author Response.pdf

This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.

Round 1

Reviewer 1 Report

Reviewer’s Report on Manuscript Stats-2690473:

“New vessel extraction method by using

skew normal distribution for MRA images”

Summary. The paper uses the univariate skew-normal distribution to model the noise in MRA data, which hampers vessel extraction and therefore vascular disease diagnosis, grading of the stenoses and aneurysms in vessels. Therefore, the correct use of MRA data helps in planning brain surgery and performing angioplasty. The paper uses the expectation-maximization algorithm for obtaining the maximum likelihood estimates of the model parameters. The practical usefulness of the proposed method is assessed with real 2D MRA images.

General. In the last 25 years, the skew-normal distribution and its generalizations have become very popular in the statistical literature. However, their use in subject-matter fields, such as medicine, has been rather limited. The paper contributes to filling the gap between statistical theory and statistical practice. The statistical methodology in the paper is strongly supported by the existing statistical literature and the material in the paper is clearly presented. Therefore, I do believe that the present paper will pave the way to further medical applications of the skew-normal distribution.

Advice. The EM algorithm strongly depends on the starting values. In the statistical literature, they are often derived using the method of moments, using the formulae in Pewsey (2000). Despite its flexibility, the skew-normal distribution can’t model bimodal data. For this reason, the more flexible generalized skew-normal distribution (Loperfido, 2004) has been often used in place of the ordinary skew-normal. I reckon that this generalization falls outside the scope of the present paper, but the authors should at least mention it in the section “Conclusions” as a direction of future research.

References

Loperfido, N. (2004). Generalized Skew-Normal Distributions. In “Skew-Elliptical Distributions and Their Applications: A Journey Beyond Normality”, CRC/Chapman & Hall, 65-80.

Pewsey, A. (2000). Problems of inference for Azzalini’s skew-normal distribution. Journal of Applied Statistics 27, 859-870.

Author Response

Advice. The EM algorithm strongly depends on the starting values. In the statistical literature, they are often derived using the method of moments, using the formulae in Pewsey (2000). Despite its flexibility, the skew-normal distribution can’t model bimodal data. For this reason, the more flexible generalized skew-normal distribution (Loperfido, 2004) has been often used in place of the ordinary skew-normal. I reckon that this generalization falls outside the scope of the present paper, but the authors should at least mention it in the section “Conclusions” as a direction of future research.

Response: Thank you for your valuable feedback. We acknowledge the importance of the starting values in the EM algorithm and the limitations of the skew-normal distribution in modeling bimodal data. While the generalized skew-normal distribution falls outside the scope of our present paper, we will definitely mention it in the "Conclusions" section as a potential direction for future research. We appreciate your suggestion and will ensure to address it in the revised version of the paper.

We have corrected it (in the new version of paper).

Reviewer 2 Report

1.     Deep learning is a new promising segmentation method which have been successfully used in many image segmentations. Why the authors not discuss in the introduction?

2.     on line 309, why 10-6 is used? How many impacts this value on the segmentation performance?

3.     On line 322-323, the author declared that There is no ground truth for MRA images, we can only compare the results qualitatively. If no quantitative comparison, the experimental results is not perfect. Did you try some phantom data sets which the ground truths are known?

4.     In Fig. 2, why do not use the manual segmentation in (a) as the ground truth?

5.     More references within recent 3 years are needed.

Some descriptions can be improved.

Author Response

Deep learning is a new promising segmentation method which have been successfully used in many image segmentations. Why the authors not discuss in the introduction?

Response: Thank you for your comment. We appreciate the suggestion to include deep learning as a segmentation method in the introduction. We value your input and addressed this issue in introduction.

On line 309, why 10-6 is used? How many impacts this value on the segmentation performance?

Response: We changed it to ϵ. In this paper, we set. We can use other values for ϵ. If the value of ϵ is very small, we need more iterations to get accurate results, and if the value of ϵ is larger, the results may be less precise.

On line 322-323, the author declared that There is no ground truth for MRA images, we can only compare the results qualitatively. If no quantitative comparison, the experimental results is not perfect. Did you try some phantom data sets which the ground truths are known?

Response: Thank you for your insightful comment. We agree that the absence of a ground truth for MRA images limits the ability to quantitatively compare results. While we did not use phantom datasets with known ground truths in this study, we recognize the importance of such validation in evaluating the performance of segmentation methods. In future research, we will consider incorporating phantom datasets to allow for quantitative comparisons and provide a more comprehensive assessment of the experimental results. We appreciate your suggestion and will take it into account for future studies.

In Fig. 2, why do not use the manual segmentation in (a) as the ground truth?

Response: Thank you for your question. Using manual segmentation as the ground truth is a common and valid approach in evaluating segmentation methods. In the context of Fig. 2, if manual segmentation was available, it would indeed be beneficial to use it as the ground truth for comparison with the results obtained from the K-means, EMS, TFA, Wilson, and the presented method. In case where manual segmentations are available, they can provide a reliable reference for evaluating the accuracy of automated segmentation methods. Manual segmentation is a common and handy role for image segmentation. However, if good results are obtained from manual segmentation, less effort is needed for further attempts.

More references within recent 3 years are needed.

Response: We added some new references within recent 3 years.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Reference number 39 should be

G. De Luca and N. Loperfido, “A Skew-in-Mean GARCH Model for Financial Returns”, in “Skew-Elliptical Distributions and Their Applications: A Journey beyond normality.” Chapman and Hall London, UK:,205-222, 2004.

And not

G. De Luca and N. Loperfido, “Skew-Elliptical Distributions and Their Applications: A Journey beyond 521 Normality. CRC.” Chapman and Hall London, UK:, 2004.

Author Response

Advice. Reference number 39 should be,

2004. De Luca and N. Loperfido, “A Skew-in-Mean GARCH Model for Financial Returns”, in “Skew-Elliptical Distributions and Their Applications: A Journey beyond normality.” Chapman and Hall London, UK:,205-222, 2004.

Response: Thanks a lot for pointing out this. We have corrected it (in the new version of paper). Pl see ref #45.

Reviewer 2 Report

All my comments are disregarded but not considered and revised seriously.

Author Response

All my comments are disregarded but not considered and revised seriously.

Response: With all due respect, it is noted that we have completed the introduction and added more references as per your request.

In image processing, various factors are used to evaluate the effectiveness of a method, such as PSNR, SNR, ENL, Entropy, CV rate, and so on. Typically, the comparison with an original image is necessary to obtain meaningful results. However, in the context of vessel extraction, there is a unique challenge.

Vessel extraction involves extracting vessels from an image, resulting in an output that only contains the vessels themselves. As a result, it becomes impossible to directly compare the extracted vessels with the original image. Consequently, reporting results in the form of a table, which is commonly done when comparing images, becomes impractical in this case.

In such situations, an alternative means of evaluation, such as Visual Comparison, can be employed. During Visual Comparison, instead of relying on numerical metrics, you can visually compare the extracted vessels with the original image or with ground truth annotations, if available. This subjective assessment can provide valuable insights into the quality of the extraction method.

We highlighted the corrections in the paper. We are grateful for your excellent comments, which certainly helped to improve the quality of the paper.

Author Response File: Author Response.pdf