MSCF: Multi-Scale Canny Filter to Recognize Cells in Microscopic Images

Round 1

Reviewer 1 Report

The manuacript proposed a new segmentation technique named MSCF to recognize the boundaries of cells. There are some suggestions for the authors to consider:

1.      Figure 1 is not neccesary, and it is suggested to be deleted.

2.      In the section “3. Methods”, the multi-scale canny filter is the most important novelty of this manuscript, so Algorithm 1 can be shown in detail in page 5. While the description for procesures of the algorithms 2 and 3 should be deleted or listed in the supplementary materials, and the related content need to be refined.

3.      In the section “4. Results”, the comparison between different segmentation methods need to be simplified, and just retain the most parameters or in brief.

4.      The segmentation results of different methods should be shown and compared directly, taking one kind of the specimen for example.

Minor editing of English language required.

Author Response

Thank you for your comments to improve the manuscript. Below we include the detail answer for each suggestion.

 

The manuscript proposed a new segmentation technique named MSCF to recognize the boundaries of cells. There are some suggestions for the authors to consider:

1. Figure 1 is not neccesary, and it is suggested to be deleted.

RESPONSE: We think that Figure 1 is absolutely required for the reader to understand the complexity and variability of the cells in the images for the different fish species. We do not work with images from a standard benchmark image database, well known for all the researchers, but with very specific images such as in Figure 1. This is the reason why the reader needs Figure 1 in order to know how the images are.

2. In the section 3. Methods, the multi-scale canny filter is the most important novelty of this manuscript, so Algorithm 1 can be shown in detail in page 5. While the description for procesures of the algorithms 2 and 3 should be deleted or listed in the supplementary materials, and the related content need to be refined.

RESPONSE: Following the reviewer suggestion, algorithm 3 and its related content was moved to the supplementary material. The figure 2 was also moved to the supplementary material.

3.  In the section 4. Results, the comparison between different segmentation methods need to be simplified, and just retain the most parameters or in brief.

RESPONSE: The results section is the most relevant part of the paper and the other reviewers did not recommend to reduce it.

4.  The segmentation results of different methods should be shown and compared directly, taking one kind of the specimen for example.

RESPONSE: Tables 2 and 4, and Figure 3, already report the results for each fish species, so they already compare directly the segmentation results of the different methods. 

 

Reviewer 2 Report

Please find my comments as follows:

1.       The novelty of the proposed work is not significant. It is just showing the comparison of the works with older methods.

2.       The literature review of the paper is quite old. Only 3-4 papers from the last 6 years. It could not establish the current research scope.

3.       The research gaps for the proposed work are not clear.

4.       The state-of-the-art comparison with earlier works on given datasets is missing to establish the significance of the work.                                                                                                                              

5.       The caption of the table should be written above the Tables.

6.       Many deep learning-based segmentation approaches are also available such as U-Net+, SegNet, Deeplab etc. In the era of deep learning, the current work is not looking significant. Why authors have not considered those architectures in your work? It should be clarified.

7.       Authors can check for typos errors and grammatical mistakes in the work.

8.       The number of images is less.

9.       Authors should highlight the contributions of the proposed work in the second last paragraph of the Introduction section.

10.   The authors should elaborate on all the variables used in Algorithm 1 and 2.

Minor editing of English language required.

Author Response

Thank you for your comments to improve the manuscript. Below we include the detail answer for each suggestion.

Please find my comments as follows:

1. The novelty of the proposed work is not significant. It is just showing the comparison of the works with older methods.

RESPONSE: The current paper proposes a novel approach based on the Canny filter and compares it with existing segmentation methods that are appliable for this problem, over several fish species.

2. The literature review of the paper is quite old. Only 3-4 papers from the last 6 years. It could not establish the current research scope.

RESPONSE: Following the reviewer suggestion, we updated the bibliography with other 7 recent related papers.

3. The research gaps for the proposed work are not clear.

RESPONSE: The solutions proposed in the literature to the difficulties issued by the current problem do not provide a fully automatic cell recognition. The current paper goes one step ahead by evaluating several classical segmentation techniques on this problem, and proposing a multi-scale Canny filter that performs unsupervised segmentation. This method is computationally efficient, outperforms the existing approaches and saves work to the experts in the fishing research laboratories to manage marine resources. This is explained in the paper introduction.

4. The state-of-the-art comparison with earlier works on given datasets is missing to establish the significance of the work.

RESPONSE: The current paper already compares the proposed method to existing state-of-the-art approaches, outperforming them on the given datasets (see Table 4 and Figure 3, for the comparison at pixel and region level, respectively) and proving the significance of the work. 

5. The caption of the table should be written above the Tables.

RESPONSE: Following the reviewer suggestion, the caption of all tables have been written above the Tables in the new manuscript.

6.  Many deep learning-based segmentation approaches are also available such as U-Net+, SegNet, Deeplab etc. In the era of deep learning, the current work is not looking significant. Why authors have not considered those architectures in your work? It should be clarified.

RESPONSE: Thank you for your question. The use of deep learning (DL) for image classification or segmentation task are growing exponentional in the last years. But, it is known that DL techniques require powerful computational resources and they are time consuming. The microscopic images are normaly high resolution images, so it is necessary to apply the DL on image patches or down-sampled images. None of both options are acceptable in our case, because: 1) image patches should be large enough to include a few number of whole cells, and therefore patches would be too large for DL setting. And 2) image down-sampling until the size required by DL networks would significantly reduce the image information to perform cell recognition. On the other hand, our objective is the design of efficient segmentation techniques to run in general purpose computers, in order to be used in all the institutions around the world to manage their marine resources, without the need of specific computing resources. As well, we want to design techniques to be included in STERapp software and to work interactively, without huge training times as happens with DL. Finally, all the techniques used in our comparison are unsupervised state-of-art segmentation techniques, but, DL is normally used in a supervised manner. Due to this causes, deep learning tecniques are not included in our comparison. 

We incorporated two paragraphs explaining this in the sections 1 and 4, marked in red color, of the new manuscript.

7. Authors can check for typos errors and grammatical mistakes in the work.

RESPONSE: The new manuscript has been carefully reviewed to correct all the typos all around the text.

8.  The number of images is less.

RESPONSE: Although the number of images for each fish specie is low, the images have been adquired using different acquisition systems in two sample processing labs, and each image has about several tens cells to be segmented. In addition, the specimens were selected in such a way that all the possible variability within each fish species is contemplated in the images.

9.  Authors should highlight the contributions of the proposed work in the second last paragraph of the Introduction section.

RESPONSE: We incorporate the following paragraph in the new manuscript:

The aim of this work is to design and evaluate algorithms to segment cells in histological images of fish gonads requiring low computational time and resources in order to: 1) be executed on general purpose computers in order to be used in all institutions around the world without any specific computing equipment in order to manage its marine resources, e.g. to be included in the STERapp software; and 2) be used in interactive systems, because the problem is too complex to provide a totally automatic recognition and it may be necessary the expert supervision before the image quantification.

10.  The authors should elaborate on all the variables used in Algorithm 1 and 2.

RESPONSE: We carefully reviewed section 3, alongside with algorithms 1-3, in order to ensure that the meaning of all the variables is fully explained.

 

Reviewer 3 Report

This manuscript proposed a multi-scale canny filter based image segmentation algorithm in the application of fish fecundity. The proposed algorithm has the flexibility to use various scales and thresholds. By applying the algorithm to different sample datasets and comparing to other common algorithms, the authors showed varying degrees of improvement even though the improvement was not across the board. As stated by the authors, there are no single algorithm to solve all image segmentation problems. In this manuscript, the improvement claimed by the authors is supported by the metrics the authors adopted to compare with other well established algorithmns. Thus the manuscript can be accepted with minor revisions:

1. Needs minor grammar check.

2. Needs consistency check, e.g., Table 4 pouting dataset TH doesn't match that in Table 2.

None

Author Response

Thank you for your comments to improve the manuscript. Below we include the detail answer for each suggestion.

This manuscript proposed a multi-scale canny filter based image segmentation algorithm in the application of fish fecundity. The proposed algorithm has the flexibility to use various scales and thresholds. By applying the algorithm to different sample datasets and comparing to other common algorithms, the authors showed varying degrees of improvement even though the improvement was not across the board. As stated by the authors, there are no single algorithm to solve all image segmentation problems. In this manuscript, the improvement claimed by the authors is supported by the metrics the authors adopted to compare with other well established algorithmns. Thus the manuscript can be accepted with minor revisions:

1. Needs minor grammar check.

RESPONSE: We reviewed carefully the whole paper and solved all the grammar issues in the new manuscript.

2. Needs consistency check, e.g., Table 4 pouting dataset TH doesn't match that in Table 2.

RESPONSE: Thank you for your comment. Both tables have check and modify in the new manuscript.

 

Round 2

Reviewer 1 Report

The authors have addressed some of my concerns. It is suggested to be accepted.

Minor editing of English language required.

Reviewer 2 Report

After a thorough review and consideration of the author's responses, I am satisfied with the revisions made and confidently recommend the well-written and relevant manuscript for publication. The author's efforts in revising the manuscript are appreciated

Minor editing of English language required