Three-Dimensional Multifaceted Attention Encoder–Decoder Networks for Pulmonary Nodule Detection

Round 1

Reviewer 1 Report

The authors proposed  Multifaceted Attention Encoder-Decoder Networks (MAED) for  Pulmonary Nodule Detection for lung cancer prevention and treatment, which is quite interesting. However there are few concerns to be addressed by the authors:

1. The authors need to rephrase some portion of the abstract to improve reader's interest, particularly from the lines 8 to 16 

          The authors stated in line 8 " this paper  incorporates the multifaceted attention block into an encoder-decoder structure." This line implies different meaning from proposing.

           In line 9 the authors stated that  "It combines it with a 9 region suggestion network to propose Multifaceted Attention Encoder-Decoder Networks (MAED)".  These it and it do not give proper meaning.

            And also in the last line 16, the author stated  that " The model’s overall performance was at the level of state-of-the-art methods"., The words "was at the level of" gives the meaning that they achieved the same performance like the state of art methods.

2.  In section 1, the authors mentioned that "According to data published by the World Health Organization’s Agency for Research on Cancer" , here the cite the necessary statistical data to improve the reader's interest.

The authors need to mention the organization of the paper at the end of the section 2.

3. In section 4.2, it would be more intersecting if the authors outlines the implementation details in a tabular form.

In line 309, the authors mentioned that " In this paper, we further experiment on the LUNA16 dataset to compare our method", so the authors need to mention about their further experimentation what they did on the dataset. 

 

4. The authors need to rephrase the results section in order to avoid the grammatical mistakes for instance in the lines 366. As the authors already did this experimentation, they need to address those words like that they added, combined but not to "add'' or "combine".

Finally the authors need to mention their obtained accuracies for the models they have developed, in the results section.

The authors need to do a thorough grammatical check through out the manuscript.

The authors need to do a thorough grammatical check through out the manuscript and make necessary changes to improve the quality of English in the manuscript.

Author Response

1. The authors need to rephrase some portion of the abstract to improve reader's interest, particularly from the lines 8 to 16

 

Thank you very much for your suggestion! We rewrote the relevant content of the abstract to make the content of the abstract more clear and reasonable.

 

Modified: Lung cancer is one of the most dangerous cancers in the world, and its early clinical manifestation is malignant nodules in the lungs, so nodule detection in the lungs can provide the basis for the prevention and treatment of lung cancer. In recent years, the development of neural networks has provided a new paradigm for the creating computer-aided systems for pulmonary nodule detection. Currently the mainstream pulmonary nodule detection models are based on Convolutional Neural Networks (CNN), however, as the output of a CNN is based on a fixed-size convolutional kernel, it can lead to a model that cannot establish an effective long-range dependence and can only model local features of CT images. The self-attention block in the traditional Transformer structures, although able to establish long-range dependence, are as ineffective as CNN structures in dealing with irregular lesions of nodules. To overcome these problems, this paper combines the self-attention block with the learnable regional attention block to form the multifaceted attention block, which enables the model to establish a more effective long-term dependence based on the characteristics of pulmonary nodules. And the multifaceted attention block is intermingled with the encoder-decoder structure in CNN to propose 3D Multifaceted Attention Encoder-Decoder Networks (MAED), which is able to model CT images locally while establishing effective long-term dependencies. In addition, we design a multiscale module to extract features of pulmonary nodules at different scales and use a focal loss function to reduce the false alarm rate. We evaluated the proposed model on the large-scale public dataset LUNA16, with an average sensitivity of 89.1% across the seven predefined FPs/scan criteria. The experimental results show that the MAED model is able to simultaneously achieve efficient detection of pulmonary nodules and filtering of false-positive nodules.

  

2. In section 1, the authors mentioned that "According to data published by the World Health Organization’s Agency for Research on Cancer" , here the cite the necessary statistical data to improve the reader's interest.

 

Thank you very much for your suggestion! We cited the necessary statistics.

 

Modified: According to data published by the World Health Organization’s Agency for Research 25 on Cancer, in 2020 alone, 2.21 million people suffered from lung cancer and 1.8 million 26 people died from lung cancer. Lung cancer is the second most common cancer globally and 27 the number one cause of death.

 

The authors need to mention the organization of the paper at the end of the section 2.

 

       Thank you very much for your suggestion! we mention the organization of the paper at the end of the section 2.

3. In section 4.2, it would be more intersecting if the authors outlines the implementation details in a tabular form.

 

Thank you very much for your suggestion! We listed all the hyperparameter lists in Section 4.2.

 

In line 309, the authors mentioned that " In this paper, we further experiment on the LUNA16 dataset to compare our method", so the authors need to mention about their further experimentation what they did on the dataset.

 

 We apologize for the misdirection caused by our use of the wrong words. We changed the word "experiment" to "analysis".

 

4. The authors need to rephrase the results section in order to avoid the grammatical mistakes for instance in the lines 366. As the authors already did this experimentation, they need to address those words like that they added, combined but not to "add'' or "combine".

 

Finally the authors need to mention their obtained accuracies for the models they have developed, in the results section.

 

Thank you very much for your suggestion! We rewrote the conclusion to make it more reasonable.

 

Modified: This paper proposes a multi-task deep learning algorithm, the MAED model, for pulmonary nodule detection and false positive nodule screening. We combined the self- attention module with the LRA block to form the multi-faceted attention block, which enabled the model to establish irregular long-range dependencies based on the features of the nodules, making the model better able to deal with the irregular lesions of pulmonary nodules. We integrated the multifaceted self-attention block with the encoder-decoder structure in CNN with each other, which enabled the proposed model to model the local texture features of nodules while establishing an effective long-range dependency, making the model better able to identify the fine nodules. The performance of the model was further improved using methods such as multi-scale modules and focal loss functions. We conducted extensive experiments on the LUNA16 dataset to validate the MAED model had the effectiveness of each component, and the overall performance of the model had an average sensitivity of 89.1% at 7 predefined FPs/scan. The model proposed in this paper has better performance compared to state-of-the-art single- or multi-stage detection methods.

Representativeness/racial bias will occur when the datasets used to develop AI algorithms are not sufficiently diverse to represent many different population groups and/or characteristics. Therefore, in future work, we will collect more clinical data from different regions/country to avoid over-concentration of data leading to representativeness/ethnicity bias of the model. In addition, we will further expand the pulmonary nodule detection model into a pulmonary nodule segmentation model to provide more help to radiologists in their diagnostic work.

 

The authors need to do a thorough grammatical check through out the manuscript.

 

Thank you very much for your suggestion! We re-proofed all the descriptions in the manuscript and made extensive changes.

Author Response File: Author Response.docx

Reviewer 2 Report

The paper is indeed well written and easy to follow. However, I have some suggestions for improvement. Firstly, the result section appears to be too concise and lacks a proper discussion and conclusion. Additionally, certain aspects of the results and discussion seem to have been incorporated into the method section, which may confuse readers. To enhance the paper, I recommend expanding the result section to include a thorough discussion and a clear conclusion. This adjustment will likely enhance the overall quality of the research.

Author Response

Thank you very much for your suggestion! We rewrote the conclusion to make it more reasonable.

Modified: This paper proposes a multi-task deep learning algorithm, the MAED model, for pulmonary nodule detection and false positive nodule screening. We combined the self- attention module with the LRA block to form the multi-faceted attention block, which enabled the model to establish irregular long-range dependencies based on the features of the nodules, making the model better able to deal with the irregular lesions of pulmonary nodules. We integrated the multifaceted self-attention block with the encoder-decoder structure in CNN with each other, which enabled the proposed model to model the local texture features of nodules while establishing an effective long-range dependency, making the model better able to identify the fine nodules. The performance of the model was further improved using methods such as multi-scale modules and focal loss functions. We conducted extensive experiments on the LUNA16 dataset to validate the MAED model had the effectiveness of each component, and the overall performance of the model had an average sensitivity of 89.1% at 7 predefined FPs/scan. The model proposed in this paper has better performance compared to state-of-the-art single- or multi-stage detection methods.

Representativeness/racial bias will occur when the datasets used to develop AI algorithms are not sufficiently diverse to represent many different population groups and/or characteristics. Therefore, in future work, we will collect more clinical data from different regions/country to avoid over-concentration of data leading to representativeness/ethnicity bias of the model. In addition, we will further expand the pulmonary nodule detection model into a pulmonary nodule segmentation model to provide more help to radiologists in their diagnostic work.

Reviewer 3 Report

 

In this work, the authors highlight the need for the study of MAED: 3D Multifaceted Attention Encoder-Decoder Networks for Pulmonary Nodule Detection. The manuscript is mostly well-written. I have some major comments and suggestions (s) below.

 

1. Avoid abbreviations in the title and it needs to be revised accordingly

2.  List of abbreviations should be provided before the Introduction.

3. Avoid using direct abbreviations, which is quite confusing following the manuscript.

4. The abstract seems to be incoherent and can be re-structured for proper understanding among readers.

5. The introduction is too short and should provide a more elaborative aspect of the need for study with as many as citations from relevant literature.

6. All figures should be of high quality and source images to be submitted in original while submitting revisions.

7. It is quite absurd that authors have carried out our work but cannot conclude what they have achieved, as the conclusion section is missing.

8. I suggest to the authors that there should be a separate section after the conclusion on Future perspectives of Multifaceted Attention Encoder-Decoder Networks for Pulmonary Nodule Detection on present findings, although it is present at Line 371-373.

9. Please follow the below-recommended references may be helpful in improving the manuscript

a)      Utilization of Artificial Intelligence in Disease Prevention: Diagnosis, Treatment, and Implications for the Healthcare Workforce. https://doi.org/10.3390/healthcare10040608

b)      Pulmonary nodules detection based on multi-scale attention networks. https://doi.org/10.1038/s41598-022-05372-y

c)      DeepSEED: 3D Squeeze-and-Excitation Encoder-Decoder Convolutional Neural Networks for Pulmonary Nodule Detection. https://doi.org/10.1109/ISBI45749.2020.9098317

d)      3D multi-scale deep convolutional neural networks for pulmonary nodule detection. https://doi.org/10.1371/journal.pone.0244406 

 

Minor editing of English language required

Author Response

1. Avoid abbreviations in the title and it needs to be revised accordingly

      

Thank you very much for your suggestion! We have revised the title of the article.

 

Modified:  << 3D Multifaceted Attention Encoder-Decoder Networks for Pulmonary Nodule Detection >>

      

2. List of abbreviations should be provided before the Introduction.

      

       Thank you very much for your suggestion! We added a list of all the acronyms for this article before the model representation section in Section 3.

 

      

3. Avoid using direct abbreviations, which is quite confusing following the manuscript.

 

       Thank you very much for your suggestion! We went through each of the abbreviations, explaining each one in detail when it first appeared.

 

4. The abstract seems to be incoherent and can be re-structured for proper understanding among readers.

 

Thank you very much for your suggestion! We rewrote the relevant content of the abstract to make the content of the abstract more clear and reasonable.

 

Modified: Lung cancer is one of the most dangerous cancers in the world, and its early clinical manifestation is malignant nodules in the lungs, so nodule detection in the lungs can provide the basis for the prevention and treatment of lung cancer. In recent years, the development of neural networks has provided a new paradigm for the creating computer-aided systems for pulmonary nodule detection. Currently the mainstream pulmonary nodule detection models are based on Convolutional Neural Networks (CNN), however, as the output of a CNN is based on a fixed-size convolutional kernel, it can lead to a model that cannot establish an effective long-range dependence and can only model local features of CT images. The self-attention block in the traditional Transformer structures, although able to establish long-range dependence, are as ineffective as CNN structures in dealing with irregular lesions of nodules. To overcome these problems, this paper combines the self-attention block with the learnable regional attention block to form the multifaceted attention block, which enables the model to establish a more effective long-term dependence based on the characteristics of pulmonary nodules. And the multifaceted attention block is intermingled with the encoder-decoder structure in CNN to propose 3D Multifaceted Attention Encoder-Decoder Networks (MAED), which is able to model CT images locally while establishing effective long-term dependencies. In addition, we design a multiscale module to extract features of pulmonary nodules at different scales and use a focal loss function to reduce the false alarm rate. We evaluated the proposed model on the large-scale public dataset LUNA16, with an average sensitivity of 89.1% across the seven predefined FPs/scan criteria. The experimental results show that the MAED model is able to simultaneously achieve efficient detection of pulmonary nodules and filtering of false-positive nodules.

 

5. The introduction is too short and should provide a more elaborative aspect of the need for study with as many as citations from relevant literature.

 

       Thanks for your suggestions! We cited more documents in the Introduction section.

 

6. All figures should be of high quality and source images to be submitted in original while submitting revisions.

 

       Thanks for your suggestions! We replaced the low definition image with a higher definition image.

 

7. It is quite absurd that authors have carried out our work but cannot conclude what they have achieved, as the conclusion section is missing.

 

       Thanks for your suggestions! We rewrote the conclusion.

 

Modified: This paper proposes a multi-task deep learning algorithm, the MAED model, for pulmonary nodule detection and false positive nodule screening. We combined the self- attention module with the LRA block to form the multi-faceted attention block, which enabled the model to establish irregular long-range dependencies based on the features of the nodules, making the model better able to deal with the irregular lesions of pulmonary nodules. We integrated the multifaceted self-attention block with the encoder-decoder structure in CNN with each other, which enabled the proposed model to model the local texture features of nodules while establishing an effective long-range dependency, making the model better able to identify the fine nodules. The performance of the model was further improved using methods such as multi-scale modules and focal loss functions. We conducted extensive experiments on the LUNA16 dataset to validate the MAED model had the effectiveness of each component, and the overall performance of the model had an average sensitivity of 89.1% at 7 predefined FPs/scan. The model proposed in this paper has better performance compared to state-of-the-art single- or multi-stage detection methods.

 

 

8. I suggest to the authors that there should be a separate section after the conclusion on Future perspectives of Multifaceted Attention Encoder-Decoder Networks for Pulmonary Nodule Detection on present findings, although it is present at Line 371-373.

 

       Thanks for your suggestions! We have another paragraph on the future perspectives of MAED model.

 

Modified: Representativeness/racial bias will occur when the datasets used to develop AI algorithms are not sufficiently diverse to represent many different population groups and/or characteristics. Therefore, in future work, we will collect more clinical data from different regions/country to avoid over-concentration of data leading to representativeness/ethnicity bias of the model. In addition, we will further expand the pulmonary nodule detection model into a pulmonary nodule segmentation model to provide more help to radiologists in their diagnostic work.

 

 

9. Please follow the below-recommended references may be helpful in improving the manuscript

 

Thanks for your suggestions! We've added all the references in the introduction.

Author Response File: Author Response.docx

Reviewer 4 Report

Congratulations to the authors for this well-crafted article. The proposed method is quite effective. Only with the addition of a more comprehensive literature review can the article be published.

Author Response

Thanks for your suggestions! We cited more documents in the Introduction section.

Round 2

Reviewer 1 Report

From the revised manuscript, it is observed that the authors have addressed most of the concerns but there are some minor concerns that need to be done.

1. Figure 2 needs to be replaced without grammatical mistakes. 

2. The authors need to add the conclusion section to the manuscript.

3. It is advised to authors to refer to the journal template once to identify the missing sections in their manuscript such as author contributions, Funding, data availability statement, etc..,

4. The references are to be organized according to the journal guidelines. 

Check the grammar in the entire manuscript. 

Author Response

1. Figure 2 needs to be replaced without grammatical mistakes. 

           We are very sorry for the error caused by our negligence. We fixed the                 error in Figure 2.

2. The authors need to add the conclusion section to the manuscript.

          Thank you very much for your suggestion! We revised the title to make a              further distinction between the results and conclusions of the experiment.

3. It is advised to authors to refer to the journal template once to identify the missing sections in their manuscript such as author contributions, Funding, data availability statement, etc..,

         Thank you very much for your suggestion! We filled in the missing parts of            the manuscript.

4. The references are to be organized according to the journal guidelines. 

          Thank you very much for your suggestion! We reformatted the references.

Reviewer 3 Report

The authors have addressed all my queries. I recommend the manuscript for publication in its present form.

Meanwhile, I would also suggest that the title in SuSy should also be rewritten.

Author Response

Thank you very much for your suggestion! We revised the title to make a further distinction between the results and conclusions of the experiment.