Applications of Artificial Intelligence in Neonatology

Round 1

Reviewer 1 Report

The overall structure of the paper is conventional. The paper is not a systematic review, it can be classified as a narrative review. I miss some data that should put the review in context

1- What is the prevalence of the diseases? Low prevalence implies that the cost/benefit analysis at social level need to be considered in order to push research money in this direction. As the after effects of the pandemic response are showing, resources are not infinite. Another effect of low prevalence is the difficulty of building cohorts. Small cohorts imply scarce data, while machine learning approaches are currently data hungry. 

2- Ethical issues are paramount. Parents of a new born are in an extremely delicate circumstance. Not taking this into account means that their are being treated as cattle. We have had enough of that in the past years.

3- Please make a clear difference between references to methods in general, like CNNs applied to radiomics, and references to specific studies related to the disease in order to clarify what is the actual evidence.

4- It is desirable to have some  characteristics of the referred studies, like the size of the cohorts, that may give hints on the generality of the results. 

5- I would apreciate a critical view of the limitations of the studies. I people will be subject to the results of black box machine learning processes, at least we need to know any bias that can be hidden. For instance, a common practice in machine learning is to discard "outiers" that worsen the results. Knowing which and why is quite relevant.

In summary, the experience of the last three years should call for prudence on the claims about technological miracles in health. Making the clear difference between persons and guinea pigs is urgent.

Author Response

The overall structure of the paper is conventional. The paper is not a systematic review, it can be classified as a narrative review. I miss some data that should put the review in context

• What is the prevalence of the diseases? Low prevalence implies that the cost/benefit analysis at social level need to be considered in order to push research money in this direction. As the after effects of the pandemic response are showing, resources are not infinite. Another effect of low prevalence is the difficulty of building cohorts. Small cohorts imply scarce data, while machine learning approaches are currently data hungry. 

Thanks for underlying this thoughtful point. The prevalence of the diseases along with the estimate of related (significant) direct and indirect costs have been previously reported: we added some references to the discussion to highlight these concepts. For example, it has been estimated that the impact of preterm birth in the US in 2005 accounted for USD 26.2 billion, and probably in low-income countries the relative impact of neonatal morbidities is even higher (Pinto 2019).

Alvarez-Fuente M, Arruza L, Muro M, Zozaya C, Avila A, López-Ortego P, et al. The economic impact of prematurity and bronchopulmonary dysplasia. Eur J Pediatr. 2017;176:1587–1593.

Pinto F, Fernandes E, Virella D, Abrantes A, Neto MT. Born Preterm: A Public Health Issue. Port J Public Health 2019;37:38–49.

Zejin Ou Z, Yu D, Liang Y, He H, He W, Li Y, et al. Global trends in incidence and death of neonatal disorders and its specific causes in 204 countries/territories during 1990–2019. BMC Public Health 2022;22:360.

 

• Ethical issues are paramount. Parents of a newborn are in an extremely delicate circumstance. Not taking this into account means that they are being treated as cattle. We have had enough of that in the past years.

We could not agree more with the reviewer and discussed this point in the manuscript.

• Please make a clear difference between references to methods in general, like CNNs applied to radiomics, and references to specific studies related to the disease in order to clarify what is the actual evidence.

Thanks for this comment. We briefly resumed the methods in general in the Introduction, whereas discussed specific studies related to the most significant diseases in the other sections of the manuscript.

• It is desirable to have some characteristics of the referred studies, like the size of the cohorts, that may give hints on the generality of the results. 

We thank the reviewer for this comment. We implemented this aspect of the discussion of the cited studies, especially in those cases in which a small study cohort could limit the generalizability or the accuracy of the proposed algorithm.

5- I would apreciate a critical view of the limitations of the studies. I people will be subject to the results of black box machine learning processes, at least we need to know any bias that can be hidden. For instance, a common practice in machine learning is to discard "outiers" that worsen the results. Knowing which and why is quite relevant. In summary, the experience of the last three years should call for prudence on the claims about technological miracles in health. Making the clear difference between persons and guinea pigs is urgent.

We thank the reviewer for raising this point and strongly agree with her/his vision. We only included well-conducted studies about significant neonatal diseases and made no claims about miracles of AI. We tried to further underline this point in the manuscript.

Author Response File: Author Response.docx

Reviewer 2 Report

This ms is novel, excellently written and of intense interest to medicine, AI and the world.

Author Response

We are greatly thankful to the Reviewer for these kind comments.

Reviewer 3 Report

I have reviewed the article "Applications of Artificial Intelligence in Neonatology".

Below are the comments and suggestions:

·       The abstract is just showing some introduction and applications of AI. Re-write the abstract, it should contain some important findings of the work as well.

·       What is your contribution? The authors need to highlight the significance of the study.

·       Write clear separate sections for methodology and results.

·       The discussion section is very short and needs much improvement.

·       The manuscript needs to be well structured with proper headings.

·       There is no conclusion section.

·       The above sections are important; without these, the article is incomplete. It should be considered after resubmission if the above sections are included properly.

·       There are grammatical issues; proofread the manuscript carefully.

 

Author Response

• The abstract is just showing some introduction and applications of AI. Re-write the abstract, it should contain some important findings of the work as well.

Thank you for your comment. We expanded the abstract though we could not overcome 200 words as per journal’s guidelines.

• What is your contribution? The authors need to highlight the significance of the study.

Given that this is a narrative review, our aim was to summarise previous studies about important aspects of neonatal care in order to inform clinicians and researchers on the potential applications of AI, their strengths and areas of improvement, and implications for research agenda. We clarified these points in the discussion.

• Write clear separate sections for methodology and results.

Since this is a narrative review, we felt not to use the typical structure of a systematic review. As done in these kinds of publications, we provided a more fluid and readable structure in order to discuss the findings of selected studies organised by systems and diseases.

• The discussion section is very short and needs much improvement.

We updated the discussion according to the reviewers’ suggestions.

• The manuscript needs to be well structured with proper headings.

We divided the manuscript into 3 chapters and then the second chapter into 6 sub-categories. Every chapter and subcategory have its own headings.                                                                                                           

• There is no conclusion section.

We added a conclusion section.

• The above sections are important; without these, the article is incomplete. It should be considered after resubmission if the above sections are included properly. There are grammatical issues; proofread the manuscript carefully.

The manuscript has been proof-read by a native English speaker physician.

Round 2

Reviewer 1 Report

thanks for answering my questions

Reviewer 3 Report

All my comments are addressed with suitable explanations. I have no further comments.