Paper: Training a Supervised Cilia Segmentation Model from Self-SupervisionPaper added

[salirezav]

If you are creating this PR in order to submit a draft of your paper, please name your PR with Paper: <title>. An editor will then add a paper label and GitHub Actions will be run to check and build your paper.

See the project readme for more information.

Editor: Sanhita Joshi @sanhitamj

Reviewers:

• Chong Shen Ng @chongshenng
• Tek Kshetri @iamtekson

[github-actions]

Curvenote Preview

Directory	Preview	Checks	Updated (UTC)
papers/Alireza_Vaezi	🔍 Inspect	✅ 54 checks passed (5 optional)	Sep 3, 2024, 1:53 AM

[ameyxd]

@salirezav Thanks for your submission. Can you update the DOIs that fail the checks?

You may be able to add citation keys you want to ignore if their DOIs don't exist in myst.yml under error_rules:

error_rules:
  - rule: doi-exists
    severity: ignore
    keys:
      - abc
      - def01

[salirezav]

Hi,

Thank you for your help. I made sure that every bibliography entry has a valid DOI, and otherwise appended them to the error_rule section. However, the check step still fails to complete.

[iamtekson]

Hi @salirezav, I am Tek Kshetri, A graduate student at the University of Calgary, Canada. and one of the reviewers of your paper. I am currently reading your paper and will provide you with some comments here.

[iamtekson]

Overall, the paper is well-written and easy to understand, even for someone not from the medical field. I appreciate the clarity and thoroughness of the authors' explanations. Thank you for giving me the opportunity to review this work.

Summary of paper:

The study introduces a self-supervised framework for automating cilia segmentation, crucial for diagnosing ciliopathies. Traditional methods rely on labor-intensive, manually labeled data. The new approach leverages optical flow to generate motion-based pseudolabels from healthy cilia videos, creating a robust training set for a semi-supervised neural network. This model effectively segments both motile and immotile cilia, overcoming inconsistencies in ciliary motion. Using a two-stage process involving optical flow and autoregressive modeling, the method achieves high accuracy without expert-drawn masks, significantly enhancing automated cilia analysis and potentially accelerating research and diagnostics for ciliopathies.

My general comments

I have listed few general comments as below,

1. All the figures have two copies in png and webp formats. I suggest removing one of them.
2. The paper could be further enhanced by including a detailed discussion section. This section could compare the proposed method with existing approaches, elaborate on potential limitations, and suggest future research directions. Furthermore, it would be nice to explain about the generalization of the model. Additionally, a deeper analysis of the pseudolabeling accuracy and its impact on the model’s performance would provide more comprehensive insights into the framework’s efficacy.

[chongshenng]

Hello @salirezav 👋 , thanks for your submission! I'm Chong Shen, one of the reviewers of your paper. I'll add my comments shortly.

[chongshenng]

General comments

The authors propose a two-stage process to improve image segmentation models for cilia detection. The first step of the process first applies a series of image processing techniques, followed by auto regressive modelling. Finally, they apply a FPN architecture to train their image segmentation model.

The results from this work is promising especially since the proposed method appears to be quite novel. I've provided a few comments in the manuscript. However, there're still a few points below that I'd encourage the authors to address before accepting it:

• There's no description of the cilia video/image dataset in this paper. Can you provide a brief description of it in the Methodology section? Is it an open source dataset? What were the training data sizes?
• It'd be instructive to readers if a simple diagram can be included to show the workflow from raw video/image to a trained model.
• The results refer to IoU, sensitivity, and specifity, which are fine. It would be more instructive to readers if there's an indication of a baseline for comparison of performance.

[chongshenng]

Please refer to my commment here.

[salirezav]

Hi @salirezav, I am Tek Kshetri, A graduate student at the University of Calgary, Canada. and one of the reviewers of your paper. I am currently reading your paper and will provide you with some comments here.

Hello @iamtekson,
Thank you for your comments! I will be addressing your suggestions and comments in the following days.

[salirezav]

Hello @salirezav 👋 , thanks for your submission! I'm Chong Shen, one of the reviewers of your paper. I'll add my comments shortly.

Hello @chongshenng,
Thank you for your comments and suggestions! I will be addressing the suggestions throughout the following days.

[sanhitamj]

Hi @salirezav Just a reminder that 2nd Sept is the final deadline for the author(s) to make edits to the submission.

[salirezav]

Hi @salirezav Just a reminder that 2nd Sept is the final deadline for the author(s) to make edits to the submission.

Hi @sanhitamj ! Thank you for the reminder! All of the changes suggested by the reviewers have been already made and pushed. I only need to make a minor change and then everything will be ready way before the deadline.

Thank you for all the help!

[sanhitamj]

@salirezav thanks for the submission and the subsequent edits as well.

@iamtekson and @chongshenng thank you so much for the careful review. Will you please get back with your final decision about the paper by 9th Sept?

[iamtekson]

Now, it looks much more improved. I vote for accepting it. Thanks @sanhitamj and @salirezav for providing me the opportunity to review it.