An Adaptive Generative Adversarial Network for Cardiac Segmentation from X-ray Chest Radiographs

Round 1

Reviewer 1 Report

Dear Authors,

 

The paper is well written and the content is interesting for both scientific and application fields.

 

In my opinion, giving some detail more, will help readers of any extraction to go trough the paper with confidence.

Please detail a little bit more the caption of the pictures and specify when it is possible some details like, for example, Dice_coeff, ROC in the training curve (Figure 8), etc.

It seems that the caption of the figure 8 is not the right one.

 

Regards 

Author Response

Dear Reviewers:

Thank you very much for your confirmation and for giving me the opportunity to revise the paper. Your opinion is very insightful and very pertinent. My paper has been revised according to your comments. specifically:

1. The evaluation criterias in the paper and the title of the figure, such as the details of Dice_coeff, etc., have been explained in more detail. The evaluation indicators of the training process have been compared to ensure readers of any extraction to understand this paper.
2. Figure 8 was not a ROC curve.. Based on your opinion, The ROC curve may better illustrate the network performance. I replaced it with a ROC curve.
3. For time reasons, the "methods described" part is too late to make too many changes, please forgive me.

Special thanks to you for your good comments.

Reviewer 2 Report

Specific comments

 

Introduction

P2: ‘However, due to the low resolution of X-ray chest radiography and serious tissue interference, the task of image segmentation is difficult’. Please refer to the resolution limits in digital radiography with appropriate references.

P3: ‘At present, good results have been achieved in applying GANs for medical image segmentation tasks, but the more accurate the medical image segmentation results are, the better.’ Please revise this sentence.

 

P3: ‘Adaptive frameworks are learning frameworks with great potential that have achieved good results in the context of control algorithms [22,23]. In this paper, an adaptive framework is introduced into the GAN approach.’ Please provide more information regarding the selection of this approach. What are the benefits compared to other approaches? Furthermore, it would be beneficial to add a comparative table of the various approaches.

 

P3: ‘dynamically selects features extracted by the feature extractor’ Please revise.

 

P3: ‘An artificial neural network is an algorithm that simulates the human visual nervous system. A convolutional neural network is a deep learning algorithm built on traditional artificial neural networks’ Please use appropriate references.

 

P4: ‘first layer is H0 = X0. Under the assumption that Hi corresponds to a convolutional layer, Hi’ Please use subscripts.

 

 

Results and discussion

 

P10, Table: The present results should be justified better. It is impressive that AGAN shows better results in every case except for Specificity. Authors could discuss on the specific characteristics of the selected sample of chest radiographs. First of all, there were only 247 images. Some more information on the sample characteristics? If the sample were greater or with different characteristics, AGAN would provide these results?

Author Response

Thank you very much for giving me the opportunity to revise the paper. My reply has been written in the attached file, please consult, thank you again.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Dear authors,

 

Thank you for your manuscript.