An Approach to Automatic Hard Exudate Detection in Retina Color Images by a Telemedicine System Based on the d-Eye Sensor and Image Processing Algorithms

Line 363 “To sum up   this chapter” – this is not a thesis chapter!

We are thankful for   your comment. You are of course right, we changed this sentence to “To sum up   this section…”.

Frequent use of “not   only … but also” need to be avoided.

Thank you for your   valuable comment. We made a language inspection of our paper. In most of the   cases we changed “not only” and “but also” to other different language forms.

The literature review   was performed in an ad-hoc manner, with the authors discussing a few selected   papers and comparing them to their approach. Since this is a well-established   research field, that has been around for some time, I would expect to a more   formal, systematic literature review.

Thank you for your comment.   We changed literature review. At the beginning we present two, most important   general approaches (the one that is based on image processing and the second   artificial intelligence based). Then we present the most interesting algorithms   of each group. Both of them were also summarized. Another improvement was   additional literature review about telemedicine. We present two groups of   this idea – stationary and mobile. Moreover, we present a few interesting   approaches connected with both of them.

The authors do not   systematically compare their approach with existing approaches in the   literature. For the same reasons as above, I would expect (a) a table   comparing techniques used in their approach with techniques in other   approaches and (b) a table comparing the accuracy (including positives and   false negatives) of these approaches.

We are thankful for   your comment. We added two tables (Table 3 and Table 4) in section number 4   (Discussion). The first of them presents a comparison between our approach   and other approaches (we take into consideration – accuracy, sensitivity and   specificity). The second table presents summary of the used techniques. In   both of the cases we chose four most interesting approaches (to which our   idea was compared).

The authors claim   that they have employed three different acquisition devices in order to   capture the 150 images used in this study: d-Eye overlay for smartphone (a)   Kowa VX-10 Fundus 131 Camera (b) and Digitial Eye Center Microclear Handheld   Ophthalmic Camera HNF (c). However, they do not cite anywhere, how many of   these images were gained with each of these devices, they only cite that they   employed 50 “low quality healthy retina images”. They should explicit the source   of the images.

Thank you for your   valuable comment. We described each of three used devices in Table number 1   (section 2.1). We added information about: price, quality, weight, refractive   compensation and image resolution. Moreover, we added information about how   many images were gained with each device (and with which quality) in section   3. We also provided information that all images were obtained in medical   conditions. We also pointed out that we used all these devices because we   would like to check whether it will be possible to properly mark pathological   changes in images from different devices.

And where are the   “low quality retina images with pathological changes”? Only high quality   images present pathologies. The lack of one diagnostic category of images is   a one of the image sets is also a flaw.

Thank you very much,   indeed. Right now, we have 10 samples with pathological changes that were   obtained with low quality devices. We added this information in the table.   Moreover, the experiments have shown that for new 10 samples, pathological   changes were properly detected.

Authors describe in   the abstract the data handled and presented in the results “68 patients – 39   females and 28 males with an age ranging between 50 and 64 were examined” but   there’s no explanation about the selection procedure in methodology section.   In the results a hundred images are mentioned (50 with pathology and 50   without it). This question should be cleared.

Thank you very much   for your valuable comment. In the paper we added information that we   collected 100 images from 67 patients because some of them were represented   by two samples (one that was obtained at the beginning of treatment and the   second one after some time of treatment). Moreover, we added information that   all images were obtained in medical conditions.

There’s no detailed   description of d-eye, only a reference, even though it was described in   articles 21 and 22. I believe a short description would fit the article   between than just a reference. The camera, sensor and algorithm are the main   key of the article but none of them are well described.

We are thankful for   your comment. We added information about d-Eye (especially how the retina   image is projected and what the conditions are for the best quality of images   and how it has to be used). Moreover, we provided a short description about   the problem that is solved by d-Eye in comparison to standard ophthalmoscopes   (corneal glare).

In the methodology   the pre-processing and processing methodology is mentioned, but nothing about   software implementation (language, app or cloud based) is said. Everything   seems blurry. It should be considered to add this information so readers may   picture how the system works physically in a more precise way.

Thank you for your   comment. We added information about languages used for implementation.   Moreover, we presented how the system was composed (REST Services) and we   also provided information about two cloud platforms on which our application   was deployed. Additional information about user identification was also   described. This information is presented in section 2.1.

At the start three   systems of image capture are discussed: d-Eye overlay for smartphone (a),   Kowa VX-10 Fundus 131 Camera (b) and Digital Eye Center Microclear Handheld   Ophthalmic Camera HNF (c), but no further information is exposed. Adding   extra information like a price range, quality of the images captured etc.   could improve the section.

Thank you very much   for your comment. We provided additional information about each device:   price, quality, weight, refractive compensation and image resolution. All of   them are presented in Table number 1 (in section 2.1).

In results section a   hundred images’ result is presented but there’s no information about how they   were captured and with which system, neither it is mentioned the criteria   used to differentiate between high- and low-quality images. It’s all quite   blurry. I do not understand either why the three capturing system have been   used to compare results. It is not enough saying that low quality images may   induce false positives, detecting pathology where there isn’t. I think it’s   way interesting to mention when its detected in images that has. Overall this   whole section is poor in its presentation and quite imprecise, which make for   a conclusion that don’t support the data. Authors should perform more   experiments and present them appropriately, so their conclusions are backed   up by the data.

We are very thankful   for your valuable comment. At the beginning we have added information that we   used three devices because we would like to check whether it will be possible   to observe pathological changes also in low-quality images. We also described   in which conditions each of them was obtained and which devices were used to   obtain each group. We also provided this information in Table 2 which was   added only for this purpose. Another improvement was connected with the analysis   of low-quality images of pathological changes. Moreover, we provided a   comparison between our approach (in terms of accuracy, specificity and   sensitivity) to other algorithms. We also made a short comparison between our   idea (methodology) and other solutions. This information is provided in the   new Table 3 and Table 4 (section 4) prepare in accordance with your   recommendations.

The discussion   section is more of a reflection by the authors, even when presenting metrics   of time of diagnostic must be acknowledged. Still, for that to be done   properly, more information about the communications system and the way the   software is executed should be presented. It is important to perform a   detailed discussion in this section, where authors compare their results with   others achieved by other authors. That way readers may be able to build a   better picture of the findings revealed by this paper. It may also help to   reinforce the conclusions.

Thank you for your   comment. We added information in section 2.1 and 4 about communications   system and the way in which software is executed. Moreover, we presented   comparison between our approach and four selected algorithms (in terms of   accuracy, sensitivity and specificity). We also made a short comparison   connected with used techniques in our idea and four selected.

References should be   updated, only 14 or 32 range from 2015 to 2019. There are many works in the   same line since the topic researched is of current interest.

Thank you for your   comment. We changed a few source papers and added some new articles (published   between 2015 and 2018).

After passing the   document through the anti-plagiarism software, the percentage is shown as 9%.   I see that this doesn’t affect the content of the document. I will attach the   file.

Thank you very much   for this information. We have analyzed the report you attached and indeed we   are happy this would not affect the content of the document.