A Modified Robust FCM Model with Spatial Constraints for Brain MR Image Segmentation

Round 1

Reviewer 1 Report

This paper proposes a modified fuzzy c-means (FCM) algorithm for brain MR image segmentation, which is robust to noise. The method is based on utilizing a new measure to incorporate weighted intensity information from a local neighborhood of a pixel. The measure tends to suppress image noise while maintaining details and edges. The proposed method is tested with synthesized images, simulated brain MR images and real brain images and is compared with other similar algorithms.

I have following concerns:

1. The authors used too much content to describe related algorithms, which should be put into the main algorithm section.

2. How the size of the sliding window will affect the denoising/segmentation result?

3. I do not see why this method cannot be applied to 3D MR images directly. However, this paper applied the method to 2D frames that are extracted from a 3D MRI. Please provide more details about how to apply this method to segment 3D images.

Author Response

Response to Reviewer 1 Comments

 Point 1: The authors used too much content to describe related algorithms, which should be put into the main algorithm section. 


 Response 1: In section 2, two classical algorithms: EnFCM and FGFCM are described. The principle and mechanism of two algorithms is the basis of our improved scheme, so we made a more detailed explanation. Considering the reviewer’s suggestion, we have sorted out and improved the content, and the title of section 2 is replaced by “Basic concepts and related algorithms”.

 Point 2: How the size of the sliding window will affect the denoising/segmentation result?

 Response 2: We have added this content in subsection 4.4 according to the reviewer’s suggestion. With the change of the radius size of the window, the segmentation result of MRFCM algorithm is illustrated in detail.

 Point 3: I do not see why this method cannot be applied to 3D MR images directly. However, this paper applied the method to 2D frames that are extracted from a 3D MRI. Please provide more details about how to apply this method to segment 3D images.

 Response 3: Generally, 3D image data is acquired in magnetic resonance imaging, but in clinical medical imaging applications, the extracted 2D slices can more conveniently view and analyse the internal tissue and structure of brain MR image from 3D image, and then the location of the organ and tissue is described more specifically. Therefore, in the experiments in this paper, our experimental samples used 2D slice images. In future research work, we will focus on the extension of the existing algorithm to 3D data and multi-modal image segmentation. 

Reviewer 2 Report

In this paper, the authors proposed an image segmentation technique for brain magnetic resonance imaging (MRI) scans. Their proposed technique is based on a modified version of the fuzzy c-means (FCM) technique. They tried to suppress the impulse and Gaussian noise. In addition, they tried to preserve the details and edge information in the brain images. The authors should consider the following issues:

Major Comments:

1.       In the abstract section, the authors should represent the overall performance accuracy of their proposed brain segmentation technique to show its effectiveness.

2.       The authors should write a section that discusses the current related work. At the end of this section, they should discuss the current limitations of the related work and how did they overcome these limitations in their proposed technique.

3.       It is recommended to write the proposed segmentation technique as an algorithm (Psuedo code) to follow it easily (instead of the subsection program flow). Also, the authors should draw a block diagram of their proposed segmentation technique.

4.       Using the average segmentation accuracy is not enough to evaluate the proposed segmentation technique. The authors should calculate other performance metrics, such as sensitivity, specificity, Hausdorff coefficient, and the area under the ROC curve.

5.       The authors should discuss how did they overcome the overfitting to the training of their proposed technique. Did they use any cross-validation method?

6.       What is the number of images used from BrainWeb in subsections 4.2 and 4.3?

7.       The experimental results section is very concise and needs more details.

8.       Where is the discussion section?

9.       The authors should discuss their future work directions at the end of the conclusion section.

Minor Comments:

1.       The paper needs English proofreading as it contains many grammar errors and typos, such as:

a.       In the abstract section, change “to accurately segment and extract brain tissue” to “to segment and extract brain tissue accurately”

b.       There are many missing articles in the manuscript, such as “of FCM algorithm” in the abstract section and “tissue in physiological” in the introduction section.

c.       In subsection 2.1, write “local spatial neighborhood” instead of “spatial local neighborhood”

2.       The authors should discuss the structure of their paper at the end of the introduction section.

3.       The title of section 2 should be changed to “Basic Concepts.”, which will be clearer to the reader.

4.       In section 2, speak about the previous techniques in the simple past tense.

5.       It is recommended to cite the reference directly after the author’s names, i.e. change “Szilagyi et al.” to Szilagyi et al. [9]”

Author Response

Response to Reviewer 2 Comments

 Major Comments:

 Point 1: In the abstract section, the authors should represent the overall performance accuracy of their proposed brain segmentation technique to show its effectiveness. 


 Response 1: We have added corresponding content in abstract according to the reviewer’s suggestion, and the overall performance of our algorithm is described at the end of the paragraph.

 Point 2: The authors should write a section that discusses the current related work. At the end of this section, they should discuss the current limitations of the related work and how did they overcome these limitations in their proposed technique.

 Response 2: Considering the reviewer’s suggestion, we have added this part to discuss the current related work at the second paragraph in “1.Introduction”, and at the end of this section we analyze the limitations of the related work and introduced our improved scheme for these problems.

 Point 3: It is recommended to write the proposed segmentation technique as an algorithm (Psuedo code) to follow it easily (instead of the subsection program flow). Also, the authors should draw a block diagram of their proposed segmentation technique.

 Response 3: Psuedo code of the proposed segmentation technique is illustrated in subsection 3.4, and the flowchart for MRFCM is shown in Figure 2.

 Point 4: Using the average segmentation accuracy is not enough to evaluate the proposed segmentation technique. The authors should calculate other performance metrics, such as sensitivity, specificity, Hausdorff coefficient, and the area under the ROC curve.

 Response 4: In medical image segmentation, the main evaluation indicators of the proposed segmentation algorithm include average segmentation accuracy (SA), Dice coefficient (DC), Jaccard similarity (JS), Hausdorff coefficient, etc. In order to avoid confusion between subjective and objective evaluation criteria, we select two typical evaluation indicators: SA and DC, and SA and DC are used in subsection 4.1 (Table 1), 4.2 (Figure 6) and 4.3 (Table 2), respectively.

 Point 5: The authors should discuss how did they overcome the overfitting to the training of their proposed technique. Did they use any cross-validation method?

 Response 5: In the segmentation technique proposed in this paper, we design an improved algorithm based on fuzzy c-means clustering, which belongs to an unsupervised learning algorithm. The data category of experimental sample is unknown, which needs to divide the sample set according to the similarity between samples to achieve the purpose of minimizing the distance within the class and maximizing the distance between classes. The segmentation process does not require training of the sample image, so no cross-validation method is used.

 Point 6: What is the number of images used from BrainWeb in subsections 4.2 and 4.3?

 Response 6: We selected 9 simulated brain MR images from BrainWeb in subsections 4.2, and 9 real brain MR images from IBSR database are used in subsections 4.3.

 Point 7: The experimental results section is very concise and needs more details.

 Response 7: Considering the reviewer’s suggestion, we have revised section 4 and analyzed the experimental results in detail. At the same time, we also added a subsection 4.4 to discuss the segmentation results with the change of neighborhood window size.

 Point 8: Where is the discussion section?

 Response 8: According to the reviewer's comments, we specifically added a subsection 4.4 to discuss the impact of the change in window radius on the segmentation results in addition to using some objective evaluation indicators.

 Point 9: The authors should discuss their future work directions at the end of the conclusion section.

 Response 9: We have added corresponding content in section 5, and the future work directions are described at the end of the conclusion section.

Minor Comments:

 Point 1: The paper needs English proofreading as it contains many grammar errors and typos, such as:

       a.  In the abstract section, change “to accurately segment and extract brain tissue” to “to segment and extract brain tissue accurately”

       b. There are many missing articles in the manuscript, such as “of FCM algorithm” in the abstract section and “tissue in physiological” in the introduction section.

       c. In subsection 2.1, write “local spatial neighborhood” instead of “spatial local neighborhood”

 Response 1: Considering the reviewer’s suggestion, the full paper has been revised and polished up in the grammar and expression, especially the point pointed out by the reviewer. For many missing articles in the manuscript, we have supplemented in the references.

 Point 2: The authors should discuss the structure of their paper at the end of the introduction section.

 Response 2: The structure of our paper has been described at the end of the introduction section.

 Point 3: The title of section 2 should be changed to “Basic Concepts.”, which will be clearer to the reader.

 Response 3: The title of section 2 has been changed to “Basic concepts and related algorithms”.

 Point 4: In section 2, speak about the previous techniques in the simple past tense.

 Response 4: In section 2, the corresponding tense has been revised.

 Point 5: It is recommended to cite the reference directly after the author’s names, i.e. change “Szilagyi et al.” to Szilagyi et al. [9]”

 Response 5: According to the reviewer’s suggestion, the cite form of the reference has been revised.

Round 2

Reviewer 1 Report

I do not have additional comments.

Reviewer 2 Report

The authors satisfied most of my comments. I have no more comments for them.