EEG Signals Based Internet Addiction Diagnosis Using Convolutional Neural Networks

Round 1

Reviewer 1 Report

At the outset, before proceeding with the review process of this article, I would like to point out that the reviewed paper is titled as " Electroencephalogram (EEG) Signals based Internet Addiction Diagnosis Using Convolutional Neural Networks (CNN)", posted at MDPI system (details in email dated 22/04/2022).

After an in-depth review of the present work, it is important to emphasize that the work has been developed at a high scientific level, thus contributing not only to the knowledge covering the field of IT, in particular Internet addiction, based on the use of convolutional neural networks, but also presenting predictions in the application of the proposed approach (EEG, CNN), to other potential addictions, including the application of the developed algorithm in some real-time health monitoring systems.

This is evidenced by many arguments, including considered by the authors of this study in the aspect of analysis of the literature on the subject of the research, the process of analysis of various types of addictions based both on the use of electroencephalogram (EEG) and the created model of neural networks CNN, and above all on the research conducted in this area through the development of CNN network architecture (model, experimental results), including the presented discussion on the proposed solution to the scientific problem (analysis of computational and experimental results).

However, it should be noted that the main idea/ essence of this paper is the technical approach/methods proposed by the authors necessary to make an accurate analysis in the field of Internet addiction, using the proposed approach (EEG, CNN) based on deep learning. For this purpose, the area of knowledge covering convolutional neural networks was used, which, together with other fields of science, e.g. knowledge of algorithms, expert systems or fuzzy logic, is included in the field defined in the literature on the subject of research as the so-called artificial intelligence.

Confirming the above assessment, it should be noted that the authors of this article, apart from a graphic presentation, discussion and proper analysis of their research and simulation tests in the aspect of the scientific problem that concerns them, including the experiment performed, highlighted their research findings through graphic illustration (Figs. 1-5) and tabulation (Tables 1-2) and the simulations performed, together with a thorough analysis of the results obtained, which allowed them to formulate important observations and final conclusions, reflected in in practical applications.

In addition, both in terms of methodology (structure, layout of the work), and mainly in terms of content, based on the aptly conducted analysis of the current state of knowledge in the analysis of the process of diagnosing Internet addiction on the basis of EEG signals and the created CNN network model, based on the use of neural networks, the authors have made the necessary research, testing and simulation experiments.

This is very important from the point of view of existing knowledge, as evidenced by the research results obtained by the authors in terms of the strategy adopted to solve the research problem.

During the review of this work, except for minor editing errors and some inaccuracies occurring in the abstract section and in the conclusion of the reviewed work in the methodological aspect, I did not find any other shortcomings that significantly affect the level and quality of this study.

Abstract:

In accordance with the recommendations of reputable publishing houses and journals, e.g. IEEE TTE, IEEE Access, Wiley and Sons, or MDPI, part of the abstract should contain the following basic elements: introduction (reference to the subject matter of the study), clear definition of the aim of the work, approximation/addressing of the potential solution to the problem/methods, and response to on the basis of the research, test, experience, developed mathematical model, to formulate relevant observations and final conclusions.

The abstract should not exceed 200 words, in this article there are 165 of them. Apart from inaccuracies such as explanation of abbreviations and designations, as well as their duplication, in my opinion the summary section of this article lacks both an explicit statement of the purpose of the article and a reference to the observations and formulated conclusions (part of the conclusion of the work) in the context of the applied solution/method and results obtained.

Inaccuracies observed in part of the abstract of this manuscript:

1. No explicit statement of the purpose of the research in this manuscript.

2. In the abstract, it is not recommended to explain abbreviations and designations as done in this article, e.g. CNN, FFT especially since some of them are explained both in the title, keywords, or later in the article.
3. Lack of reference to both expected and predicted research results (analysis, model, simulations), as well as reference to important observations and final conclusions reflected in practical applications.

Minor inaccuracies noted in the rest of the work:

1. The content contained in a study (article, paper, manuscript) should be written in the impersonal form or in the 3rd person, not as presented in this paper both in the abstract and in the rest of the paper, e.g. the expression we adopt ..., p. 1, point 14 (abstract) and we found ..., p. 2, point 58, or p. 5, points 185 and 187 respectively in terms of we run ..., our model ... and we choose ... (the rest of the paper), etc. Furthermore, please note that these types of inaccuracies can be multiplied. Please check the entire paper in this regard and make the appropriate correction.
2. Duplication of many abbreviations and designations (title, abstract, keywords, and the rest of the article). Please check the entire paper in this regard and make the appropriate correction.
3. Editorial errors in both the abstract and the rest of the paper in terms of lack of clarity in the writing, for example, page 1, point 15, in Internet addiction (abstract) and internet addiction, page 1, point 33 (rest of the paper). Please refer to the entire article in this regard.
4. Failure to maintain both unambiguous writing, e.g. [3,4] on page 1, point 35 and [5, 6] on page 2, point 52, as well as proper sequencing, and duplication of literature references pertaining to the subject of the study, mainly in the use of punctuation marks. Please check the entire paper in this regard and make appropriate corrections if possible.
5. From the methodological aspect, in my opinion, it is not recommended to present a figure in the concluding part of the chapter/subchapter, as it was done in this article, e.g. Figure 3 in subsection 2.2, or tables, e.g. tab. 1-2 on p. 7 in subsection 3.2. Moreover, there was an error in the numbering of the formulas on this page, it should also be added that the mathematical analysis in this paper boils down to placing only 3 formulas, is it not enough?
6. Minor inaccuracies in the explanation of abbreviations and designations, e.g. in the case of the abbreviation Internet addiction (IA), where its explanation is only given on page 2, point 55, despite the fact that the subject of this paper's research is this term, moreover, its reference is cited both in the keywords and in the rest of the paper, so it is a key term and should have been explained earlier, e.g. in the keywords.
7. Sequences/phrases that are too short, as well as the interchangeable use of punctuation marks, resulting in formal errors (e.g., editing and punctuation) that do not affect the merit of the work, however, make the work appear careless and unprofessionally prepared, e.g., page 1, point 45; pages 2-3, points 84-85; page 6, points 213 and 226, or page 7, point 239 regarding algorism. Please check the entire paper in this regard.
8. Duplicate references to figures, e.g., relative to Figure 2 on page 4, points 151 and 159, respectively, in the text of this paper. Please make the appropriate correction in this regard.
9. No explanation of some abbreviations cited in this paper, e.g., SVM, ReLU, and others, e.g., cited in Table 2 (CNN-BN, fNIRS). Please check the entire paper in this regard.
10. The final conclusions should be supported by the obtained research results. This is all the more incomprehensible, since the reviewed manuscript was developed by the authors at a high scientific level, especially in terms of the performed research (analysis, tests, simulations, experiment), as well as in terms of in-depth analysis of the obtained results.

Strong aspects:

The technical approach/methods used (CNN model, SVM algorithm, EEG signal, etc.), the idea of solving the problem and its explanation, the analysis of the research results obtained, supported by the analysis and the formulation of the final conclusions, the relevance in terms of the methods used and the ability to use them. 

Weak aspects:

Weak aspects concern mainly editing errors, use of punctuation marks and some inaccuracies of methodological character, namely these are minor shortcomings that do not significantly affect the quality of the reviewed work, i.e. editing errors and poorer quality of the abstract and conclusion parts in terms of methodology.  

Recommended changes:

Regardless of the Editor's decision, at this stage of the work I would recommend that the authors of this paper make the corrections contained in the above review and in weak aspects. 

In conclusion, the solution proposed by the authors of this manuscript in the form of the created CNN model together with the use of Fast Fourier Transform (FTT), based on the solution using neural networks based on EEG signals, is characterized by a high level, and thus contributes to scientific development.

Author Response

Please see the attachment (response letter) for details. 

Author Response File: Author Response.pdf

Reviewer 2 Report

The authors performed a diagnostic study on Internet addiction using CNN and FFT features of EEG recordings.

The paper lacks novelty for both preprocessing and classification. The use of frequency-domain features is frequently investigated in other studies as well as there is no technical improvement in CNN.

The authors did not discuss the obtained results and the provided table consisted of comparison with drug addiction studies that are not related to this study.

As a result, more comprehensive experiments, detailed comparisons, and discussions are required.

Author Response

Please see the attachment (response letter) for details. 

Author Response File: Author Response.pdf

Reviewer 3 Report

The authors proposed a Convolutional Neural Network (CNN) model for supporting the diagnosis of Internet addiction. Three types of input signals were considered: raw EEG data, full spectral power after FFT, and filtered alpha-beta-gamma spectral power. The comparison of the results showed a better performance of the full spectral power. The topic is interesting but I have some doubts, in particular about the method.

1. The dominant brain rhythm during closed-eyes condition is alpha. Why did you perform your analysis considering alpha, beta, and gamma bands?
2. You have a dataset consisting of EEG recordings in opened-eyes and closed-eyes conditions. Why did you consider just the opened-eyes EEG signals? A comparison between the results deriving from these two conditions could be interesting.  
3. In the introduction, you do not mention that you used three different input signals and it is not clear which of the results the performance refers to (lines 94-96). Explain it.

Minor comments:

1. You labeled Equation 3 as Equation 2.
2. Argumentation process -> Augmentation process (line 193).

Author Response

Please see the attachment (response letter) for details. 

Author Response File: Author Response.pdf

Reviewer 4 Report

This paper presented a deep learning method for the diagnosis of internet addiction. Although the topic is interesting I have some comments that must be addressed before publication, therefore, I suggest major revision.

1. Abstract must be rewritten. It should contain objective, method, results, and significance. Currently, it is too general. I expect to see more about the method and why is it better than the state of the art algorithms.
2. I suggest authors to start the introduction with few sentences about EEG and its different applications for the real-world scenarios such as epilepsy detection ( https://doi.org/10.3390/s22093318,  https://doi.org/10.3390/s22083066), driver drowsiness detection (10.1109/JBHI.2021.3096984,  https://doi.org/10.3390/s18124477), and sleep analysis (10.1109/JSEN.2022.3155345, https://doi.org/10.3390/s22083079). Then, authors can state the importance and applicability of EEG for detection of addiction.
3. I could not find enough motivation for this study. Authors are expected to clearly state why it is important to conduce this study and why CNN.
4. This paper is full of typos. Authors must reread the paper and fix them, e.g., page 2, line 54 Internet addiction (IA), Internet Addition (IA), page 2 line 57 Correlation Verification Results, correlation verification results.
5. I do believe that EEG pre-processing plays an important role for the accurate classification, in particular, for the eye open epochs where eye blinks can seriously adverse the classification results. Yet, authors used a simple linear filter. Authors are expected to add one paragraph to discussion section stating that using advanced EEG de-noising methods (e.g., 10.1109/TIM.2021.3115586, https://doi.org/10.3390/s22082948) may further improve the classification results.   I do not believe that comparison in Table 2 about drug and internet addiction is appropriate. These are two different things which obviously have different influence on the brain activity.  

Author Response

Please see the attachment (response letter) for details. 

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Thanks to the authors for their great improvements to the manuscript. However, there are major things in the paper that cause confusion for the readers.

-          “Figure 2. Basic concepts of CNN” is the representation of an artificial neuron. The figure does not include the convolutional layers, pooling layers, etc.

-          Line 211. “The pooling layer translates the output of the neuron clusters in one layer into a single neuron in the next layer, and this process can reduce the feature map resolution.”

The pooling operation is used to reduce the feature map dimensions as mentioned by the authors; however, the definition of the pooling layer is confusing if the authors are not defining the global pooling. As is shown in Figure 3, the authors used max-pooling, which is applied to the extracted features and reduces the dimensions of the feature maps but does not convert/translate/map them into a single layer.

-          Equation 8 is not for ReLU but the Softplus activation function. They are similar, but not the same.

 

-          It would be better if the authors add a block diagram for preprocessing phase that demonstrates each step in the frequency domain.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

In my opinion the paper improved and can be accepted.

Author Response

Thanks a lot for the reviewer's help, and we believe the quality of the manuscript has been enhanced a lot following the reviewer's constructive comments. 

Reviewer 4 Report

I would like to thank the authors for the revised paper. It seems much better now. 

Author Response

Thanks a lot for the reviewer's help, and we believe the quality of the manuscript has been enhanced a lot following the reviewer's constructive comments. 

Round 3

Reviewer 2 Report

Thanks to the authors for the corrections. All my concerns are addressed in this revised version.