Brain Tumor Classification Using Conditional Segmentation with Residual Network and Attention Approach by Extreme Gradient Boost

Round 1

Reviewer 1 Report

This topic is interesting, but several points needs to be addressed. Look at these points:

- Line 1: "A brain tumor is a tumor in the brain that has grown out of control." "A brain tumor is a dangerous condition" "Brain tumors are 33 classified as benign (noncancerous) or malignant (cancerous)"  .. as all tumors. Please removed these obvious phrases.

- Lines 45-45 "The brain tumor becomes the root cause of interruption in the normal circulation of cerebrospinal fluid (CSF)" This sentence is not correct, no all tumors stop normal circulation of CSF. Revise

- Lines 81-82 "1.2. MOTIVATION" What do authors mean with this subtitle? Is this the aim of the paper?

- Lines 42-43: "The World Health Organization (WHO) has classified benign tumors as grade 1 and 2 low-grade tumors" grade 1 and 2 are not benign! Revised. Discuss also about recurrent glioblastoma. look at refs: Impact of recurrence pattern in patients undergoing a second surgery for recurrent glioblastoma. Acta Neurol Belg. 2022 Apr;122(2):441-446. doi: 10.1007/s13760-021-01765-4. -- Surgical outcome and molecular pattern characterization of recurrent glioblastoma multiforme: A single-center retrospective series. Clin Neurol Neurosurg. 2021 Aug;207:106735. doi: 10.1016/j.clineuro.2021.106735. 

- Table 1. Is this a review of the literature? How the papers of table 1 were selected?

- Figure 3 legend should be improved

- Lines 351-356: "Segmentation is performed in existing approaches... and classification." Improve this point.

- Some points in the paper probably need to be discuss with a neurologist/neurosurgeon to be improved.

- In the conclusion section, please add what this paper add new to the literature and previous published papers. 

Author Response

Dear Sir

Greetings

Here is my response related to the queries received. I hope with this response we will be able to justify the comments and hopefully this article will be accepted.

Please see the attachment related to received comments.  

Please find the Response as  mentioned below

Line 1: "A brain tumor is a tumor in the brain that has grown out of control."

"A brain tumor is a dangerous condition”

Response from Author

A brain tumor is a tumor in the brain that has grown out of control and its dangerous condition for human body

 

Line-33      Please removed these obvious phrases

"Brain tumors are33 classified as benign (noncancerous) or malignant (cancerous)”.. as all tumors.

Response from Author

UPDATED BY AUTHOR

Brain tumor is a dangerous condition. The proliferation of aberrant cells within the brain affecting the nervous system and even the central spine, compromising brain function

Lines 45-45 "

The brain tumor becomes the root cause of interruption in the normal circulation of cerebrospinal fluid (CSF)" This sentence is not correct, no all tumors stop normal circulation of CSF.       

Response from Author

REVISED BY AUTHOR

The brain tumour causes an interruption in the normal circulation of fluid (CSF) around the brain, resulting in increased intracranial pressure and the symptom of edoema.

Lines 81-82 "1.2. MOTIVATION"

What do authors mean with this subtitle? Is this the aim of the paper?

Response from Author

This is motivation of selecting this problem and deep learning approach

Lines 42-43:

 "The World Health Organization (WHO) has classified benign tumors as grade 1 and 2 low-grade tumors" grade1 and 2 are not benign! 

Response from Author

Updated by author

Based on brain health behaviour, the World Health Organization (WHO) has classified benign tumors as grade 1 and 2 low-grade tumors and malignant tumors as grade 3 and 4 high-grade tumors.

Discuss also about recurrent glioblastoma.

look at refs: Impact of recurrence pattern in patients undergoing a second surgery for recurrent glioblastoma.

Response from Author

Added by author

Francesco Pasqualetti et al. (2021) sought to determine how different patterns of GBM failure affect the outcomes of second surgery. In a prospective cohort of recurrent GBM patients, overall survival (OS) and post-recurrence survival (PRS) were evaluated as per clinical features, such as pattern of recurrence. Using the Kaplan–Meier technique, survival curves were determined by calculating, and the log-rank test was used to assess the distinctions between curves. Patients with a local recurrence had a longer OS than those with a non-local recurrence, 24.1 months versus 18.2 months, respectively [P = 0.015, HR = 1.856 (1.130–3.010)]. This advantage was more pronounced in patients with local recurrence [P = 0.002 with HR 0.212 (95% CI: 0.081–0.552) and P = 0.029 with HR = 0.522 (95% CI: 0.289–0.942)]. The recurrence pattern can influence the outcome of patients with recurrent GBM suitable for a second surgery.

Hashmi and Osman (2022) propose Conditional Deep Learning for Brain Tumor Segmentation, 4 Residual Network-Based Classification, and Overall Survival Prediction Using Structural Multimodal Magnetic Resonance Images (MRI). First, they came up with Convolution network-based segmentation and Conditional Random Field, which found patches that didn't overlap. Because of the tumour, these patches need to be put on as soon as possible. If it overlaps, make the errors bigger. In the second section of this paper, Residual network-based feature mapping with XG-Boost-based learning was proposed. In the second section, the main focus is on mapping features in nonlinear space with residual features. Residual features make it less likely that information will be lost, and nonlinear space mapping gives accurate information about tumours. After XG-Boost learns how to map features, structural-based learning gets better and class-wise accuracy goes up. The study uses two datasets: the first is based on three classes, and the second is based on two classes. Both of these datasets are much better than another existing method.

Nicola Montemurro (2021) confirmed that the extent of research (EOR) at the time of the initial and subsequent recurrence is a strong predictor of outcome in patients with recurrent glioblastoma multiforme (GBM). As per the findings, at the time of tumour recurrence, all patients in strong performance status must be offered repeated craniotomy with the purpose of achieving a maximal resection when this is feasible and safe. Furthermore, the study proved that PFS, female sex, third surgery, MGMT promoter methylation, tumour volume at first surgery, and adjuvant chemotherapy at recurrence are prognostic factors that influence OS. Resistance and tumour recurrence in GBM are caused by significant changes in the tumour microenvironment. The research backed up the idea of dynamic evolution of the GBM genome by displaying adjustments in genetic profiles of 1p/19q co-deletion, p53 mutations at recurrence, and MGMT promoter methylation.

 

Table 1. Is this a review of the literature? How the papers of table1 were selected?

Response from Author

Paper is selected according to datset, deeplearning and machine learning approches because we give overview of previous research

 

Figure 3 legend should be improved

Response from Author

UPDATED

Lines 351-356: "Segmentation is performed in existing approaches... and classification."    Improve this point

Response from Author

UPDATED BY AUTHOR

Existing segmentation techniques disregard overlapping patch boundary overlaps. The overlap of patches increases the detection and localization error of tumours. The proposed CNN segmentation method maps patches, but condition-based Conditional Random Field extracts improve the overlap. This improves the accuracy of detection by enhancing segmentation, feature mapping, and classification.

Some points in the paper probably need to be discuss with a neurologist/neurosurgeon to be improved

Response from Author

Added by author

In a prospective cohort of recurrent GBM patients, overall survival (OS) and post-recurrence survival (PRS) were evaluated as per clinical features, such as pattern of recurrence. Using the Kaplan–Meier technique, survival curves were determined by calculating, and the log-rank test was used to assess the distinctions between curves

In the conclusion section, please add what this paper add new to the literature and previous published papers.

Response from Author

UPDATED BY AUTHOR

the proposed research aims to improve the overlap between patches and its effect on the mapping efficiency of features by residual network. In addition, the proposed work contributes learning and focus mechanisms for enhancing learning and features, respectively.

Author Response File: Author Response.pdf

Reviewer 2 Report

Review Report for the Manuscript “Brain tumor classification using Conditional Segmentation with Residual Network and Attention approach by XG-BOOST”

 

Rating the Manuscript

 

Originality/Novelty: Is the question original and well defined? Do the results provide an advance in current knowledge?

 

Yes, the authors have discussed a method for improving Conditional deep learning for brain tumor segmentation, residual network- based classification, and overall survival prediction using structural multimodal magnetic resonance images (MRI).

 

Significance: Are the results interpreted appropriately? Are they significant? Are all conclusions justified and supported by the results? Are hypotheses and speculations carefully identified as such?

 

Yes, the results are interpreted well.

 

Quality of Presentation: Is the article written in an appropriate way? Are the data and analyses presented appropriately? Are the highest standards for presentation of the results used?

 

The article is written well. 

 

Scientific Soundness: is the study correctly designed and technically sound? Are the analyses performed with the highest technical standards? Are the data robust enough to draw the conclusions? Are the methods, tools, software, and reagents described with sufficient details to allow another researcher to reproduce the results?

 

Yes, study design, methods and data analysis are explained well in this manuscript.

 

Interest to the Readers: Are the conclusions interesting for the readership of the Journal? Will the paper attract a wide readership, or be of interest only to a limited number of people? (please see the Aims and Scope of the journal)

 

Yes, this would be a great article for the not only for researchers in the brain tumor research field but also for researchers in MRI imaging field. 

 

Overall Merit: Is there an overall benefit to publishing this work? Does the work provide an advance towards the current knowledge? Do the authors have addressed an important longstanding question with smart experiments?

 

Yes. This study provides an advancement to the current knowledge. 

 

English Level: Is the English language appropriate and understandable?

 

Yes, English language in the manuscript is appropriate and understandable. 

 

Overall Recommendation: Accept after Minor Revisions

 

This is a well written paper and I only have few comments. Here’s the detailed review of each section of the manuscript.

 

Abstract

 

Abstract is well written and summarizes the content of the manuscript.

 

Introduction

 

Introduction is well written and explains the current MRI imaging methods.

 

Line 35: “A primary brain tumor originates in the brain and seldom spreads to several other body areas. It grows slowly, has well-defined boundaries, and spreads only rarely. It affects an estimated 700,000 persons in the United States now. Secondary brain tumors are the most prevalent type of tumor. It is both dangerous and life threatening. when cancerous cells from other organs, like the breast or lung, move to the brain. A malignant brain tumor expands to neighboring brain areas, grows swiftly, and has uneven borders. It contains cancerous cells.”

 

Line 44: “The brain tumor becomes the root cause of interruption in the normal circulation of cerebrospinal fluid (CSF) around the brain and generates higher intracranial pressure and causing the symptom of Edema.”

 

Authors need to references to these sentences in the introduction.

 

1.1  Brain MRI images

 

Authors could briefly discuss about the drawbacks of currently available methods for image processing and explain the need for improving these methods.

 

Are these methods helpful for the imaging of brain for other diseases as well? For example MRI imaging is a widely used method for stroke diagnosis. 

 

 

Proposed Methodology 

 

Methods section is written well by dividing it into subtopics so that it’s easy to follow. Given below are few comments on the methods section.

 

3.1 Dataset

 

Line 132: The pixels of the photographs are 0.49 by 0.49 mm2 in size, with an in-plane resolution of 512 by 512.

 

Please correct the typo. In mm2, 2 should be superscript. 

 

3.3 Segmentation

 

Line 161: “The method is commonly used in medical image processing”.

 

What types of medical images are processed? Give few examples here.

 

4. Experimental Results and Discussions

 

4.1 Performance Metrics

 

Line 266: “The following section has the five-efficiency metrics used to evaluate the suggested method’s performance Eq.”

 

Are there any other metrics that can be used to evaluate the performance? What are the reasons to select these 5 metrics?

 

4.5. Comparsion with existing

 

Line 235: “Table 5 and Figure 5 compares the accuracy of our technique to that of seven other published methods, including two feature-driven approaches and six deep learning approaches, all of which are used to classify tumors.”

 

Have they analyzed other efficiency metrics that you have mentioned in the earlier section. If so how those metrics compare with your method?

 

Tables, Figures and Figure legends

 

Figure captions need to be improved. They need to be more informative and must explain the content of the figure.

 

Figure 4: The graph in the figure needs a better title. “Comparison” doesn’t explain what’s shown in the graph.

 

Table 5 and Figure 5 represents the same results. Authors could remove one of them.

 

Author Response

Dear Sir

Greetings

Here is my response related to the queries received. I hope with this response we will be able to justify the comments and hopefully this article will be accepted.

Please see the attachment. 

Please find the response related to the queries  as mentioned below

Line 35:

“A primary brain tumor originates in the brain and seldom spreads to several other body areas. It grows slowly, has well-defined boundaries, and spreads only rarely. It affects an estimated700,000 persons in the United States now. Secondary brain tumors are the most prevalent type of tumor. It is both dangerous and life threatening. when cancerous cells from other organs, like the breast or lung, move to the brain. A malignant brain tumor expands to neighboring brain areas, grows swiftly, and has uneven borders. It contains cancerous cells.”

Authors need to references to this  sentences in the introduction

Response from Author

Reference given by author

Line 44:

“The brain tumor becomes the root cause of interruption in the normal circulation of cerebrospinal fluid (CSF) around the brain and generates higher intracranial pressure and causing the symptom of Edema.”

Authors need to references to this sentences in the introduction

Response from Author

Reference given by author

• Brain MRI images

Authors could briefly discuss about the drawbacks of currently available methods for image processing and explain the need for improving these methods.

Are these methods helpful for the imaging of brain for other diseases as well? For example MRI imaging is a widely used method for stroke diagnosis.

Response from Author

UPDATED BY AUTHOR

In image processing, segmentation is a challenging task due to the inefficiency of previous research and approaches, which increase overlapping and reduce feature efficiency.

Proposed Methodology

Methods section is written well by dividing it into subtopics so that it’s easy to follow. Given below are few comments on the methods section.

 3.1 Dataset

Line 132: The pixels of the photographs are 0.49 by 0.49 mm2 in size, with an in-plane resolution of 512 by 512.  Please correct the typo. In mm2, 2 should be superscript.

Response from Author

UPDATED BY AUTHOR

3.3 Segmentation

Line 161:

“The method is commonly used in medical image processing”.    What types of medical images are processed? Give few examples here.

Response from Author

UPDATED BY AUTHOR

The method is commonly used inbio medical image processing like skin cancer images mammography images

4. Experimental Results and Discussions

4.1 Performance Metrics

Line 266:

 “The following section has the five-efficiency metrics used to evaluate the suggested method’s performance Eq.”

Are there any other metrics that can be used to evaluate the performance? What are the reasons to select these 5 metrics?

Response from Author

UPDATED BY AUTHOR

The five-efficiency metrics used to evaluate the performance of the suggested method are listed in the following section. Other available performance metrics are all variants of the given performance metrics. One of the five main performance metrics is accuracy, which provides an overall picture of model performance. Other performance metrics, such as precision, sensitivity, and specificity, are used to analyse the true positive and true negative rate of the proposed model.

 4.5. Comparison with existing

Line 235:

 “Table 5 and Figure 5 compares the accuracy of our technique to that of seven other published methods, including two feature-driven approaches and six deep learning approaches, all of which are used to classify  tumors.”

Have they analyzed other efficiency metrics that you have mentioned in the earlier section? If so how those metrics compare with your method?

Response from Author

UPDATED BY AUTHOR

In comparison, only accuracy is used because other parameters are not analysed by the given research, and accuracy is an important parameter to analyze for the proposed model in our research.

Tables, Figures and Figure legends

Figure captions need to be improved. They need to be more informative and must explain the content of the figure.

Figure 4: The graph in the figure needs a better title. “Comparison doesn’t explain what’s shown in the graph. Table 5 and Figure 5 represents the same results. Authors could remove one of them.

Response from Author

UPDATED BY AUTHOR

Updated caption Figure 4. Analysis of different performance metrics on Proposed approach and delete Figure 5

Author Response File: Author Response.pdf

Reviewer 3 Report

This paper explores techniques that can be used to aid machine learning technologies to detect

tumors. The abstract and introduction are a bit confusing, but the sections detailing the

algorithms and results are much easier to understand.

This paper requires some heavy reworking with regard to the information that it presents.

Throughout the paper, it feels that there is too much emphasis on existing literature, and

there is quite a bit of irrelevant information that isn’t useful for understanding the models or to

understanding the results.

Specific Suggestions to improve the paper:

1. Information regarding the results of other studies should be made more concise in the

introduction.

2. In the motivation section of the introduction, there should be a description of why the

research being done within this paper is novel compared to prior studies.

3. Table 1 should be split up or shortened.

4. Spelling of tumor should be consistent throughout the paper (Line 124 uses “tumor”).

5. Lines 152-155 regarding the method of augmentation are unclear.

6. Unnecessary to restate the statistics from Table 5/Figure 5 in Section 4.5. The explanation of

upper bound accuracies of each process is difficult to understand.

Author Response

Dear Sir

Greetings

Here is my response related to the queries received. I hope with this response we will be able to justify the comments and hopefully this article will be accepted.

Please see the attachment.

I am providing the response related to the queries as mentioned below

1. Information regarding the results of other studies should be made more concise in the introduction.

Response from Author

Updated by author

In prior research, machine learning and deep learning techniques were used to classify brain tumours. In machine learning, the use of support vector machine (SVM) [29] and fisher kernel [37] improves accuracy in comparison to deep learning techniques such as convolution neural network (CNN) [33], Multi scale CNN [36], and CNN with GA [38]. In-depth learning Comparatively more improve the accuracy of classification. Consequently, these outcomes encourage the adoption of a deep learning strategy.

2. In the motivation section of the introduction, there should be a description of why the research being done within this paper is novel compared to prior studies.

Response from Author

Response from Author

Updated by author

So, in the proposed approach, an efficient deep learning model is used to improve segmentation with a conditional random field, which smooths the segmentation boundary edge and improves feature mapping.

3. Table 1 should be split up or shortened.

Response from Author

Already split

4. Spelling of tumor should be consistent throughout the paper (Line 124 uses “tumor”).

Response from Author

Updated by author

5. Lines 152-155 regarding the method of augmentation are unclear.

Response from Author

Updated by author

By augmenting images, we can increase the number of images in the dataset and improve the class imbalance. This was accomplished through the use of two transformation methods. The first image had a 90-degree angle. In both datasets, the second transformation was used three times more than the original to flip images vertically

6. Unnecessary to restate the statistics from Table 5/Figure 5 in Section 4.5. The explanation of upper bound accuracies of each process is difficult to understand.

Response from Author

Updated by author

To validate our proposed technique and its impact on detection accuracy, we compared it to the work of other researchers, who used SVM and Fisher Kernel machine learning based on detection accuracy ranging from 91 to 94%. The accuracy of CNN and CNN-based approaches ranges from 81% to 97%. Aside from that, the proposed method CRF-Resnet50 has a maximum accuracy of 99.56%.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Authors solved all my criticisms.