A Novel Hybridoma Cell Segmentation Method Based on Multi-Scale Feature Fusion and Dual Attention Network

Round 1

Reviewer 1 Report

Hi Authors,

Thank you so much for submitting this manuscript for review!

Please find the below comments which I would recommend we address to determine the necessary next steps,

 

1. Please address the typos and the grammatical errors in this manuscript.
2. How is the performance of the proposed determined here? Have we introduced approach obtained outstanding performance with the measures of accuracy, sensitivity, and specificity and if yes, what are they
3. Which database was the data stored for analyzing and then coming to a conclusion in this manuscript ?

MySQL or something else ? What techniques/tools if any were used for ETL process if any ?

4. Can we please add "Future Directions" as a the final section so that we can address the below questions,
1. Future directions/recommendations to mitigate the risks ?
2. How much cost (monetary impact) will it happen after the issue has been resolved ?

5. Mechanistic modeling and data-driven modeling constitute two approaches which are different in their traits however while the development of a mechanistic model is many times cumbersome/laborious and requires detailed knowledge about the process, data-driven approaches are rather quickly applicable and require less knowledge. Can you please throw some light on the various techniques pertaining to the modeling used in this manuscript and how were they successful in coming up with a solution that you are trying to address as a part of this manuscript?
6. Are we approaching the deep learning techniques of image segmentation from an analytical perspective? We should provide an intuitive understanding of the major techniques that have made a significant contribution to the image segmentation domain i.e. Starting from some of the traditional image segmentation approaches we should further progress by describing the effect that deep learning has had on the image segmentation domain. We should also back it up with the necessary supporting data.
7. We should also be mention thresholding techniques i.e. Mean method, P-tile method, Histogram dependent technique, Edge Maximization technique, and visual technique in this manuscript and back it up with the necessary dataset.
8. We have not included a key concept 'PDE Based Image Segmentation' i.e. PDE (Partial Differential Equations) equations or PDE models which are used widely in image processing, and specifically in image segmentation. They use active contour model for segmentation purpose. Can we please make sure to add it ?
9. Can we throw some light on Markov Random Field (MRF) based segmentation which is known as Model based segmentation. An inbuilt region smoothness constraint is presented in MRF which is used for color segmentation.
10. The CAM (Channel attention mechanism) follows the mode of encoder–decoder. In the encoder, two identical deep residual networks are both divided into multiple levels and acted on spectral images and auxiliary data, respectively. Additionally, in the decoder, the channel attention mechanism (CAM) should introduced to automatically weigh the channels of feature maps to perform feature selection. I do not see any details around this line in Section 3.2.2. Can we please add them?
11. As the complexity of neural network models has grown, it has become increasingly important to optimize their design automatically through metalearning. Methods for discovering hyperparameters, topologies, and learning rate schedules have led to significant increases in performance. In this manuscript we should highlight that loss functions can be optimized with metalearning as well, and result in similar improvements in Section 3.3 in addition to that the Loss function optimization provides a new dimension of metalearning, and constitutes an important step towards AutoML.
12. In Section 4.5, we do not have any information around SSD (Single Shot Multibox Detector) which is one of the best object detection algorithms with both high accuracy and fast speed. Can we please make sure to include it?
13. Overall the content of this paper is well articulated and written. Good Job !!!
14. For clarity sake, it would be very helpful to add more details about the sample data set. Also what's the model of the graph used in this manuscript?
15. The methods evaluated in this manuscript are not significantly different in their performance. Can you please throw some light on the performance KPI's or indicators?
16. In computer vision, image segmentation is the process of dividing a digital image into multiple segments and the objective of segmentation is to simplify or modify the representation of an image into somewhat that is more significant and easier to analyze. Image segmentation is normally used to trace objects and boundaries (lines, dots, curves, etc.) that can occur in images. Additionally, image segmentation is the process of allocating a label to each pixel in an image such that pixels with the same

label share some pictorial characteristics. Can we please build on this thought and include it in this manuscript?

Best Regards,

Dr. Shahani

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

In this paper, the authors tackled an interesting problem of automated segmentation of hybridoma cells using deep learning. The topic is worthy of investigation, and the ideas presented in the manuscript are sound, but the manuscript suffers from the following shortcomings which should be, in my opinion, thoroughly addressed before it could be considered for publication:

1.       Although the English is acceptable, the manuscript would benefit from careful proofreading, preferably with a help of a native speaking colleague. I suggest reworking the abstract to make it easier to follow (e.g., please rephrase such sentences as “Finally, the loss function Focal Loss is used to…”).

2.       It would be useful to slightly rework the related literature part of the manuscript, especially related to the segmentation of hybridoma cells. Instead of enumerating selected cell segmentation algorithms, I encourage the authors to provide more critical overview of the state of the art, to clearly present the open questions and research gaps which are being addressed in this work.

3.       Please improve the quality of the figures (they should be in a vector format).

4.       It would be useful to expand Figure 8 and to include the images with the masks overlaid on the color images as the third row for each example.

5.       The authors claim that one of their contribution is the dataset – to make this contribution valid, please provide a link to the repository containing this dataset.

6.       Similarly, please provide a link to the repository containing the implementation of the proposed framework to ensure its full reproducibility.

7.       Are the differences across the investigated variants of the model (and across different algorithms) statistically significant? Please report p values.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

In this paper multi-scale feature fusion and dual attention, the network is introduced for cell segmentation. The paper is interesting and written well. However, there are a few major issues that need to be resolved before this manuscript can be accepted.

1)      The abstract has to be changed because it doesn't adequately explain the study topic and research question.

2)      Motivation is not clear.

3)      Research gap is not defined.

4)      I suggest the authors to write their main contributions in bullet form in the introduction section.

5)      Paper structure paragraph is missing in the introduction section.

6)      Very deep neural networks face different challenges. However, a discussion on it is missing in the introduction and literature section. Following are some closely related to the topic, discussion on them should be included in the revised version: (https://doi.org/10.1007/s11554-020-01020-8, https://doi.org/10.1155/2022/9580991 )

7)      Implementation and system details are missing, which is very important for the reader to produce similar results.

8)      It is crucial to explain how the machine learning algorithms' hyper-parameters are configured. How can we be certain that parameter tuning won't impact the techniques' accuracy?

9)      I suggest the authors to provide the comparative complexity analysis of the proposed method with the other methods.

10)  The language is poor and needs polishing.

 

 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Hi Authors,

Hope you are doing well!

I would like to take this opportunity to thank you for adding updates/revisions to the manuscript as per my last comments and I will provide my feedback to the editorial team as you have my blessings.

Best Regards,

Dr. Shahani

Reviewer 2 Report

I am happy to see that the authors have addressed the majority of my concerns.

Reviewer 3 Report

All my suggestions and comments are addressed.