Feature Detection Based on Imaging and Genetic Data Using Multi-Kernel Support Vector Machine–Apriori Model

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The article addresses the problem of improving Alzheimer disease diagnosis.  While the content presents valuable insights, there are several areas that need attention for improvement. Please consider the following points:

 

1.       Content Issues:

 

The title is lengthy; consider focusing on the main original contribution and avoiding acronyms.

 

Clearly present the goals in the abstract, specifying the original contribution.

 

Ensure consistency in statements; prove statements like the use of Multi-Kernel SVM with Apriori algorithm for rapid and effective detection.

 

Align the state of the art in Section 1 with the established goals.

 

Provide references for unproven statements, such as the use of Apriori algorithms.

 

Explain figures in detail; avoid vague descriptions like "Figure 1 looks more like an extended abstract."

 

Detail and mathematically explain the Apriori algorithm and tables without repeating their titles.

 

Clarify references, and carefully select and detail citations in Section 2.1.

 

2.       Technical (Editing Issues):

 

Write out all abbreviations upon their first appearance.

 

Maintain consistency in notations, e.g., Fig. 2 should use consistent capitalization.

 

Clearly define terms, especially when introducing equations and figures.

 

3.       General Suggestions:

 

Clearly articulate the essence and novelty of the Apriori algorithm and Multi-Kernel SVM.

 

Provide a rigorous but concise presentation of the Apriori algorithm and Multi-Kernel SVM, explaining the choice of kernels.

 

Consider presenting the Multi-Kernel SVM-Apriori Model (lines 195-242) in an algorithmic format.

 

Justify model choices and kernel selection for comparison in lines 246-249.

 

Avoid conclusions based on limited examples, e.g., line 278.

 

Ensure Fig. 7 specifies the reported accuracy.

 

Clarify what "our model" means in line 331.

 

Relate your results to other papers discussed in the discussion section (lines 371-389).

 

4.       Specific observations

 

The goals of the article should be clearly presented in the Abstract, as well as the original contribution:

 

·        First, you state that you want to identify new biomarkers for AD. Later you present the accuracy of the classification problem, but you did not define any classification problem. You missed the connection between biomarkers and classification.

 

 You should be consequent in your statements.

 

·        The statement from line 15-16:  ”However, most machine learning methods used to detect AD biomarkers require lengthy training and are unable to rapidly and effectively detect AD biomarkers. To address this issue, we propose a novel approach using the Multi-Kernel Support Vector Machine (SVM) with Apriori algorithm …” You have to prove how you detect more rapide and effective biomarkers, but get lost in other details.

 

·        In lines 66-67, it stated  “The SVM[1-3] has been widely used in AD classification due to its ease of use and understanding” , while in lines 115-116, is written that “While SVM can accurately classify patients with AD from healthy controls, the decision-making process behind the classification can be difficult to interpret.”

 

Any statement that is not proved by the author in the paper should have at least a reference item to be based on.

 

In lines 122-123 is stated that “The use of Apriori algorithms, has been shown  to improve the classification accuracy of machine learning algorithms in medical imaging data analysis”. Give referece citations to prove this affirmation.

 

Give detailed explanations for the figures, not general and ambiguous ones (e.g. The overall description of the model is shown in Figure 1.). Figure 1 looks more like an extended abstract and does not have any relevance without explanation at this point of the article.

 

The Apriori algorihm in the form of frequent items should be detailed and complete and mathematical rigorous explained.

 

Do not give explanation about a table rewriting the title of the table (e.g.  the name of the Table 1  is The details of the participants. And the explanation is “The details of these participants are shown in Table 1.”)

 

In Section  2.1, in lines 148 – 150 is written “The participants' T1 and fMRI images were preprocessed by skull stripping, head  motion correction, and normalization. The present invention utilized a smoothing process  with a Gaussian kernel size of 4mm full width at half maximum (FWHM) and a frequency range of 0.01Hz to 0.08Hz to remove noise from the images.”

 

·        What invention are you talking about? Reformulate

 

In line 142, authors probably refer to DPARSF and not to DPRASF. The references [12] and [13] do not seem to be relevant. However, the entire subsection 2.1 should be totally redesign, to give a complete and clear image to the reader on the preprocessing steps. The citations should be carefully chosen and the authors must give more details about what exactly they taken from these reference items.

 

In section 2.2, the feature fusion process must be explained rigorous using an appropriate mathematical notation.

 

Equations (1), (2) and (4) should be reformulated. They just display a matrix and could be expressed in a condensed form:

 

M_gene = (bp_{ij} ),  with 1<=i<= 280, 1<=j<=130   (Obs. I used the Latex notation for subscript and superscript )

 

The same for equation (2)

 

When defining the Pearson correlation coefficient, the author can guide on

 

Zhang X, Xie Y, Tang J, Qin W, Liu F, Ding H, Ji Y, Yang B, Zhang P, Li W, Ye Z, Yu C. Dissect Relationships Between Gene Co-expression and Functional Connectivity in Human Brain. Front Neurosci. 2021 Dec 9;15:797849. doi: 10.3389/fnins.2021.797849. PMID: 34955741; PMCID: PMC8696273.

 

In equation (3), you should specify what E represent. More, specify that M_geneM_roi represent the matrix product of the matrices M_gene and M_roi. Also specify what does M_gene^2 and M_roi^2.

 

Equation (4) is given by

 

M_{gene-roi}= (Gene_i Roy_k), with 1<=i<=280 and 1<=k<=90

 

Gene_i Roy_k denotes the matrix product between the row vector Gene_i from M_gene and column vector Roy_k from M_roi

 

To make more clear the originality of the article it would be recommended that the authors briefly but rigorous present

 

·        The essence of the Apriori algorithm and what is the novelty in applying this algorithm in the present article

 

·        Multi-kernel SVM and the form of this algoritm applied in this paper. In fact the authors just sum three simple kernels to obtain the multi-kernel. They should explain why they have chosen these kernels, while do they decide to make a simple sum and not a weighted sum. A discussion about potential optimizations of the multi-kernel would be benefic.  (the article Boram Jeong et. al. Multiple-Kernel Support Vector Machine for Predicting Internet Gaming Disorder Using Multimodal Fusion of PET, EEG, and Clinical Features, Front. Neurosci., 30 June 2022, Sec. Neuropharmacology, Volume 16 - 2022 | https://doi.org/10.3389/fnins.2022.856510, should serve as possible inspiration for this task)

 

In Section 2.3, the construction of Multi-kernel SVM-Apriori Model (lines 195-242) should be described in an algorithmic way (Algorithm Multi-kernel SVM-Apriori Model). All terms used in equations should be explained  (e.g. in equation (8))

 

Lines 246-249 – give a justification of the reasons that lead you to choose the models for comparation. Specify how did you chose the kernels for single and dual-SVM used for comparation.

 

In line 278 is stated that “This indicates that our method can obtain the optimal  feature set after filtering in one of the frequent itemsets” You cannot conclude this from only one example.

 

In lines 295-299, and 302-303 the comments are just simple transposing in words what we can see in figures. The comments should be explicative and constructive.

 

It is not relevant to compare a model with multi kernel and apriori algorithm with a model with simple kernel without apriori algorithm.

 

In Fig. 7 must be stated clearly what accuracy was reported.

 

In line 331 is stated that “The results showed that our model exhibited excellent classification performance  and stability.” The authors must clearly specify what “our model means”. In Multi-kernel SVM they used only one simple sum of three kernels, without optimization. Is this considered the model? The problem of choosing the kernel in Multi-kernel SVM is a problem in itself.

 

The discussions from lines 371-389 refers to other papers. It would be interesting to underline with what extend your results confirm these conclusions, not to just enumerate these others’ conclusions.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

1- It was not very clear what information resulting from Apriori was used. In section 2.4 we talk about decision tree-Apriori model. It was not clear how he arrived at this decision tree. You could give an example with a characteristic or illustrate the steps of this part with a flowchart.

 

2- On Line 108: "However, while Apriori algorithms and support vector machine (SVM) have shown promising results in Alzheimer's disease (AD) research, there are also some disadvantages to using these algorithms". Only works using SVM were presented, so I could add a bibliographic review of studies, related to AD or not, using a priori as pre-processing.

 

3- There is an imbalance between the HC and AD classes. This is something that can interfere with the performance of the method. It would be interesting to add a confusion matrix to get an idea of how the method behaves between classes.

 

4- Add information about programming language, libraries, operating system, processor.

 

5- Make it clear in the figures and in Table 2 when the accuracy presented is in the training set and when it is in the test set.

 

6- Create a table to summarize the differences between the sets L_i.

 

7- What are the values of the other SVM parameters, such as C, gamma? Was default used in the other methods?

 

8- What do the acronyms HC, EMCI, LMCI and P mean in Table 1?

 

9- Check the spelling, an example: in line 123 “to improv” appears.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

The authors aimed to predict AD using a multimodal approach with SVM. Topic and approach is interesting but there are some major issues that need to be solved. Here are my comments:

·         In abstract, dataset can be mentioned.

·         In introduction, references can be given in the first paragraph.

·         Background of the problem can be extended, it seems to be a bit short, the authors jumped right to ML perspective.

·         It would be better if the authors create a related work section.

·         Are these references used the same dataset? If not, please cite studies that work on the same dataset.

·         What about deep learning approaches in related works?

·         Figures should come after they are referenced in the text.

·         An organizational paragraph is needed at the end of introduction.

·         Figure 1 should be described in a separate section called “Proposed framework”.

·         A preprocessing figure could be better for understanding.

·         Overall, article lacks references.

·         Models should be described briefly, and references should be added.

·         Since deep learning approaches are popular on the AD imaging domain, authors should compare their approach with other DL models on the same dataset, otherwise validity of proposed approach is questionable.

·         Authors need to add a comparison table at the end of results section. This table should include studies that used the same dataset.

·         Text of figure 8 is hard to read.

·         Limitations and future studies can be extended.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

Based on my review, I suggest minor revisions to the manuscript. Here are the observations and corresponding suggestions:

1.       Equations Notations: I still have concerns regarding the notations used in equations (1), (2), and (4) for specifying the range of the indices. It would be beneficial to reconsider and potentially revise these notations for clarity and consistency.

2.       Algorithm 1:

-        Clarify the purpose of the numbers (labels) assigned to each line. Are they intended to represent the count of lines or the algorithm steps? If the latter is the case, steps 2 to 6 seem to be explanations and could be differentiated visually (e.g., using italics) and tabulation. They should not be counted as steps.

-          Use capital letters at the beginning of each step to enhance readability and consistency.

-          Ensure correctness from an algorithmic perspective, particularly regarding the "repeat" structure. Consider revising as follows:

Repeat 

   12: .....

Until ...

-          The presence of "DO" before line 11 seems unnecessary.

I recommend the removal of the "DO" before line 11 to improve the clarity and correctness of Algorithm

 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

The authors answered my concerns and revised accordingly.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf