ASD-GANNet: A Generative Adversarial Network-Inspired Deep Learning Approach for the Classification of Autism Brain Disorder

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

ASD-GANNet: A GAN inspired Deep Learning Approach for the Classification of Humans Autism Brain Disorder

Review

A brief summary

The study introduces "ASD-GANNet," an end-to-end approach for classifying pre-processed fMRI datasets using functional connectivity (FC) features. It employs a GAN-based dataset augmenter trained on the NYU Connectivity Features dataset to generate synthetic features, enhancing classification accuracy to 82% in 10-fold cross-validation, demonstrating superiority over existing state-of-the-art methods, particularly on the NYU dataset.

General concept comments

Article:

The paper is interesting and well-written. It contributes to both deep learning techniques and ASD diagnostics.

Review:

v  In section 1.3 GAN first should be described and only after is the contributions of this method should be mentioned.

For example, GAN stands for Generative Adversarial Network. It's a type of artificial intelligence framework that consists of two neural networks: a generator and a discriminator. These two networks are trained simultaneously in a competitive manner:

1.      Generator: This network generates new data instances, such as images, based on random input (often called latent variables). Its goal is to create realistic data that can potentially pass as real data.

2.      Discriminator: This network acts like a detective, attempting to distinguish between real data from the training set and fake data generated by the generator. Its goal is to correctly classify as many real and fake instances as possible.

During training, the generator improves its ability to generate realistic data by receiving feedback from the discriminator. Conversely, the discriminator gets better at distinguishing real from fake data as it sees more examples from both categories. This adversarial process continues until the generator produces data that is indistinguishable from real data according to the discriminator.

GANs have been successfully applied to various tasks such as image generation, video generation, text-to-image synthesis, and even tasks involving medical data augmentation, such as fMRI data augmentation.

v  Lines 157-158 "The imbalance nature of dataset…" it is not clear what is the imbalance since 408 ASD and 476 HC considered balanced or is there a difference in the definitions of ASD in the different sites? Why is it important that each site will be represented in the training and testing ? what is the reason for the separation between the sites ?  this issue should be further explained.

v  In Table 2, AUC (Area under Curve -ROC curve ) should be added to the comparison.  

Specific comments:

NA

Comments on the Quality of English Language

NA

Author Response

"Please see the attachment."

Author Response File: Author Response.docx

Reviewer 2 Report

Comments and Suggestions for Authors

ASD-GANNet: A GAN inspired Deep Learning Approach for the Classification of Human’s Autism Brain Disorder

I have read the manuscript with interest, and the authors can find my suggestions, commentaries, and concerns as follows:

According to me, the title needs to be revised since it contains many acronyms that could be avoided to be more informative. 

Introduction: DSM-5, not DSM V (Roman number, that was used for the previous editions of DSM). Please, revise.

In the first paragraph, you mentioned the definition, epidemiology, and prevalent concerns with ASD. However, you need to highlight the fact that ASD is a spectrum disorder. For this reason, you are advised to add more info about the definition and the worldwide prevalence.

In the “motivation,” the link between the fMRI (BOLD) and the fact that is an imaging technique is vague. Indeed, you mentioned that allows the quantification of the amplitude of the (BOLD) signal, and I advise improving this definition.  It is not easy, since it is quite obvious what fMRI is, but please, improve it. Moreover, fMRI has been widely used to study several neuropsychiatric and other pathological conditions. This paragraph needs to be revised, since fMRI is not promising, but was widely used.

In Related State-of-the-Art, the acronyms need to be explicated in the text, and then you could use it. What do you mean by “naive contaentated approach”? In the contribution, the Multi head needs to be specified in a better way.

Line 127-The challenge due to the noise is not clear, please clarify, since the ABIDE was a preprocessed data set.

Line 151: “ASD and HC is not much imbalanced but when observing the distribution site wise , there are drastic variations” I advise assessing the presence of statistical differences. However, you used a stratified approach that is very interesting and rigorously explained.

The methodology is written in detail. However, I advise reformulating several sentences, not because of the content, but the formal writing.

Line 189-190: The dataset is rs fMRI, but you are writing about a task, and this is not clear.

Line 218: please reformulate or delete the sentence about the reader. The algorithm is easy to follow and well explained in the text, despite some sentences need to be reformulated.

From a stylistic point of view, I advise to remove the sentence at the beginning of each section with a summary (i.e. lines 232-233).

Is cGAN the same as CGAN?

 The results are interesting, but I advise to reformulate or rewrite the discussion because it is quite difficult to follow.

The conclusion is supported by the results, but I advise to highlight the limitations of the studies in a better way. 

Author Response

"Please see the attachment."

Author Response File: Author Response.docx

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

The manuscript has been improved.

Authors stated: “A BOLD signal is the deficit of the oxygen level in the blood when a neuronal activity has taken place in a brain region and the key idea of the fMRI is to measure this deficit.” This statement is completely incorrect. Do you mean that BOLD signal is a deficit?

Please revise this sentence. 

Author Response

Corrected:

A BOLD signal reflects the changes in blood oxygenation levels in response to neuronal activity in a brain region. The key idea of fMRI is to measure these changes, which are indicative of increased blood flow and oxygenation following neuronal activation