Advancements in Healthcare: Development of a Comprehensive Medical Information System with Automated Classification for Ocular and Skin Pathologies—Structure, Functionalities, and Innovative Development Methods

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

My comments are the following:

The introduction should highlight the most important points of the research.
The classification model in 3.3 should be analyzed in more detail.

The experiments can be presented in a better and more comprehensive way.

Comments on the Quality of English Language

-

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

This is a well written expository paper that displays and discusses various components of a machine learning based platform for the development of machine learning/AI architectures focused on the early detection of eye disease.

-The use of the Google GT tool is nicely explained and reviewed.

-The various components of the system and possible data types are carefully listed and discussed.

-In reality, however, users of this platform will still have challenging issues to deal with, apart from integrating so many different types of data. For example, the CNN "pre-trained" machine learning model that is either assumed or part of the process here (using Google GT?), raises issues for any particular application affecting its potential accuracy. 

(i) How can you guarantee the training data is comparable to what will typically be encountered by the system. 

(ii) As the system described here is applied, it will begin to alter and reflect the properties of the database in question. Machine-learning methods and algorithms are very data-centric and over time may alter. Will the model be evaluated and possibly re-calibrated over time? How stable is the algorithm as it "learns" and modifies its parameters? 

(iii) Why do you think the multi-nodal network model underlying a CNN is a good template for understanding eye related issues and the onset of eye related conditions? CNN is good for cluster analysis, but as a predictive model can have issues. 

-What sample size is necessary for the platform here to converge and give useful predictions? 

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

Thank you for allowing me to review this innovative and interesting manuscript.

I consider that it addresses a topic of interest for various readers. Nevertheless, assumptions are broad and need to be delimited, no ethical considerations are provided, no limitations are declared, conclusions cannot be supported given the nature of the study and unjustified use of adjectives are the main areas of improvement.

Additionally, the rationale underling the precision of the proposed means has an important limitation, which is assuming that patients have unique pathologies, pathognomonic and classical phenotypes. Ideally, 95% of all cases would have a classical phenotype, but in reality, humans are complex, according to Hickman dictum on medical diagnosis, patients can (and will) have all conditions they can have, and those conditions would superimpose and coexist. In that sense, what is the accuracy of digital classification presented in the manuscript?

What is the accuracy in patients with chronic conditions causing multiple complications in a same organ? 

One of the assumptions in the introduction states that interrater variability would be reduced by using image classification, but, what if even having perfect concordance with a specific diagnosis, such diagnosis is wrong? what I mean is that clinical diagnosis is not a matter of opinion or consensus, but of an evidence based diagnostic reasoning.

Between 278-316 lines, theoretical benefits are listed, but there should be variables, no theory, theory should not be in this section.

Lines 333-341 present skewed, unscientific text such as "significant" and "advantages" when none have been proven. Methods should be clear, complete, straightforward without eloquent exaggerated compliments. Description of the materials and methods should not include unnecessarily positive adjectives.

Another bias underlies in assuming that the studied models of organs in the image classifications are uni-bidimensional, but in reality, retinopathy may be inspected only as mage, as no clinician can touch the retina or smell it without surgery. But skin conditions are not to be diagnosed only based on images, but also touched. Other conditions may need the five senses to be suspected.

Line 832 mentions a predictive model. Please provide the prediction rate. Does prediction estimate adjust by the prevalence of the studied condition in different populations?

For example, 

In conclusion, the key findings and contributions of this article unveils a pioneering leap in healthcare technology with the introduction of a revolutionary medical information system developed within the SF environment.

without unnecessary adjectives should state something like:

This article has provided new insights on healthcare technology with the introduction of a medical information system developed within the SF environment.

Lines 874-876 are not sustained with the study and cannot be concluded, no likelihood of early diagnosis was tested in the study. It cannot be concluded.

Lines 878-880 are also not sustained, cannot be concluded.

lines 882-884 cannot be concluded. Human errors can also occur from wrong assumptions.

Lines 901-903 includes a misleading, unstained assumption, as a patient grows older and has a chronic disease, you may start treating him/her for a reason, but many coexisting pathologies ad-up with time and what seemed to be something easy to monitor will change, but not necessarily due to that specific problem. Patients are dynamic, they are not a single organ, if you follow them intime, changes may not mean a progression.

Please state ethical correlates of the proposed strategy, both bioethical and social/psychological.

Please include limitations.

Revise references, many of them are wrong/incomplete. Web pages for general audiences are used instead of scientific or academic sources.

 

 

 

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Round 2

Reviewer 3 Report

Comments and Suggestions for Authors

 

 

The text after bullet points in lines 235-277 in some cases start with capital letters and in others without, please homologate. 

Tables 2 & 3. could it be simplified with two decimals? 

I have reviewed the corrected version of the manuscript.

It has improved significantly.

Points to be corrected:

lines 413-414, please add "may" be advantageous

Line 443, Finally, upon successful validation, the trained model can be deployed for use in..., the word "successful" is not needed, let the reader arrive to that impression.

 

 

Author Response

Please see the attachment

Author Response File: Author Response.pdf