Deep Learning and Machine Learning Techniques of Diagnosis Dermoscopy Images for Early Detection of Skin Diseases

Round 1

Reviewer 1 Report

1.The topic does not match the scope of the journal.

2. The experimental results should compare with exiting methods.

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 2 Report

Authors should address the following comments: 

1) " In the early stages of the disease, it is not detected due to the similarity of the cancer cells to other skin cells. Abnormal cancerous cells divide rapidly, penetrating the lower skin layers and becoming incurable malignant melanomas. "- add the following reference for this statement:  i) Melanoma detection and classification using computerized analysis of dermoscopic systems: a review

2) "According to dermoscopy images, there are many types of skin diseases. The two main types of skin disease can be identified as melanocytic and nonmelanocytic. Melanocytic diseases have two types: melanoma and 56
melanocytic nevi. "- add the reference for this statement: i) Region extraction and classification of skin cancer: A heterogeneous framework of deep CNN features fusion and reduction; ii) An improved strategy for skin lesion detection and classification using uniform segmentation and feature selection based approach.

3) "Artificial intelligence techniques have been applied to classify many types of images and medical records, such as dermatoscopy, magnetic resonance imaging, computed tomography (CT) scans, and medical records."- this statement is written without any reference. Authors should add the following references for this statement: i) Multi-Class Skin Lesion Detection and Classification via Teledermatology; ii) Pixels to classes: intelligent learning framework for multiclass skin lesion localization and classification

4) The related work need to be update. Add the following latest articles in the related work and summarize at the end of this section. i) Computer decision support system for skin cancer localization and classification; ii) Intelligent Fusion-Assisted Skin Lesion Localization and Classification for Smart Healthcare; iii) A two‐stream deep neural network‐based intelligent system for complex skin cancer types classification; iv) Skin lesion segmentation and multiclass classification using deep learning features and improved moth flame optimization.

5) The first bullet is not a contribution. Dull razor is a well-known tool for removing hairs. refine the major contributions. Also, what do you mean by the crucial features? do not use such type of horrible words in the research. 

6) "Balance the ISIC 2018 and PH2 datasets to achieve highly efficient diagnostic accuracy"- this is not a contribution. remove it from the list. 

7) How fusion is performed? what is the size of initial vectors and after the fusion?

8) Why 80:20 approach is opted? Authors should add results for 50:50 and standard 70:30. 

9) What are the parameters of CBB during the training? add in the manuscript. 

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 3 Report

Dear Authors,

The present manuscript is an interesting study deciphering the use of medical imaging data and machine learning techniques to detect skin diseases. 

The research design is mature and significant machine learning methods have been utilized to perform the tests on choosen datasets. At this stage I wouold only recommend to address the following finding.

Please also include the year of publication along with the authors names e.g., Line 142,143, 147 etc. 

Good luck

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 4 Report

In this article, the authors apply ML and DL algorithms on two datasets [the International Skin Imaging Collaboration (ISIC 2018) and Pedro Hispano (PH2)] for the early detection of skin lesions.
Early diagnosis can prevent unpleasant consequences in human life.
The topic of the article is engaging and stimulates the reader's interest.
It has the proper organization and structure. The technical contribution is limited. The references used by the authors are up to date but, their number is relatively limited. Several points in the text raise questions and, generally, the article needs considerable improvement. Specifically, I have to point out the following.
1)I suggest the authors rewrite the abstract. In its current state, it has many unnecessary details. It is not helpful to point out the methodology you followed regarding the data sets.
2)The introduction section needs enrichment. Ιt provides insufficient background. My suggestion is to increase the number of references for deeper discussion and comparisons with the state-of-the-art. What are the main challenges in this domain?
3)The Related Work section needs improvement. The authors cite researchers who have published papers on the diagnosis of skin lesions using artificial intelligence techniques. Evaluate how your study is different from others? Αlso, highlight the motivation of this research and summarize the challenges.
4)The links with the data sets could be found in the Appendix and not in the main body of the text.
5)The quality of Figure 1,2,3,5,6 is too poor.
6)What is the Dullrazor tool?
7)Subsection 3.5.1. ANN and FFNN need references.
8)Subsection 3.5.2. Convolutional neural networks (CNNs) need references(CNN, ResNet50 model, AlexNet model)
9)What criteria were used to select the ML algorithms (ANN and FFNN)?
Similar why ResNet-50 and AlexNet?
10)Figure 7 should be aligned in the centre.
11)The label "Dataset" of table 4 is wrong.
12)The results section is very tedious. The authors list too many figures derived from the algorithms they simulated. Is it useful to have all this information? Try to make a more coherent, accurate and focused presentation
13) Figure 17 is copy-paste and, its quality is too poor.
14)In the discussion section, the authors need to perform comparisons of their results with similar studies. How your approach is better in terms of accuracy?
15)For discussion, I would like to know the limitations and the potential issues of this study.
16)Conclusions can discuss future research directions and extensions of the study.
17)The authors miss the experiment setup. Please, demonstrate the environment of the experiment in detail.

18)The last question I have concerns the purpose of this work. The purpose was to compare the results of the ML and DL models you used? To make a simple application of well-known models? This work would make some sense if compared with research works that have been published. In its current form, it is not allowed for publication.

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

No comments.

Author Response

There are no comments here
Thanks a lot.

Reviewer 4 Report

The authors made a significant effort and improve the content of the article. A few points need improvement.
1)The description you make in the cover letter about the Dullrazor tool should be included in the body of the text and accompanied by the appropriate reference.
2)What criteria were used to select the ML algorithms (ANN and FFNN)?
Similar why ResNet-50 and AlexNet?
Also, the description you make in the cover letter should be included in the body of the text and accompanied by the appropriate references.
3)The label "Dataset" of tables 4 and 6 is wrong. Please correct it.
4)Please check for grammatical, syntactic and punctuation issues in the text.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf