A Semi-Unsupervised Segmentation Methodology Based on Texture Recognition for Radiomics: A Preliminary Study on Brain Tumours

Round 1

Reviewer 1 Report

This paper proposes a brain tumor image segmentation framework. The image segmentation method adopts the traditional machine learning algorithm to replace the manual segmentation, which can assist the medical diagnosis and has a particular application value. However, there are the following defects:
1.The LBP algorithm used in this paper is similar to the census algorithm, and the author does not analyze the difference between them.
2.There is no description of the dataset used, and there is no motivation to use the SVM algorithm, which may be due to the small amount of dataset, but the author does not fully explain it.
3.The experiment is not compared with other algorithms, such as the census algorithm. The article lacks the motivation to use the LBP algorithm.

Author Response

Thank you very much for your valuable suggestions aimed at improve our work. In the following we'll tried to clarify your concerns.

1.The LBP algorithm used in this paper is similar to the census algorithm, and the author does not analyze the difference between them.

Thank you for your valuable suggestion. Local binary patterns and Census are quite similar. They encode the local region by establishing the relationship between neighbor pixels to obtain information concerning the features. Recently, LBP and its variants have been successfully applied in various applications, such as texture classification, segmentation, face recognition, object detection, and other interesting practical application. For the knowledge of the authors Census transform has only been used to investigate stereo correspondence problem and some recent works [1] demonstrate that LBP is superior to census.  

[1] Nguyen, V. D., Nguyen, P. H. and , N. C., "Local Binary Pattern and Census, Which One is Better in Stereo Matching," 2020 7th NAFOSTED Conference on Information and Computer Science (NICS), 2020, pp. 244-249, doi: 10.1109/NICS51282.2020.9335907.

2.There is no description of the dataset used, and there is no motivation to use the SVM algorithm, which may be due to the small amount of dataset, but the author does not fully explain it.

A detailed description of the considered dataset has been introduced in the body of the manuscript. For sure SVM is one of numerous machine learning algorithms. We used SVM because they are quite effective to solve complex classification problem with an high degree of accuracy. However in the future our research group can try to apply other machine learning algorithms in order to identify the one which work better for this application. Thank you for your suggestion.

3.The experiment is not compared with other algorithms, such as the census algorithm. The article lacks the motivation to use the LBP algorithm.

The considered experiments have been compared with other segmentation algorithms, in particular we used the semi unsupervised segmentation tools provided by Slicer 3D which consist of different growing area algorithm starting from a single or multiple seeds and with the indication of boundaries of the suspected area, moreover a threshold tool algorithm (Slicer 3D) has been applied trying to segmentate the pathological area. All these attempts failed because the contrast between the pathological and healthy area was very low. The considered LBP algorithm seems to work well and we agree with the reviewer, the considered LBP algorithm  is quite similar to the census algorithm, they belong to the same family, the motivation of the LBP with respect to census are reported in the answer to your point number 1. We provided to insert a better description of the comparisons made with the other segmentation algorithms in the experimental validation section. Thank you again for your valuable suggestion.    

Reviewer 2 Report

General Comments

Reviewed is the manuscript “A semi-unsupervised segmentation methodology based on texture recognition for Radiomics: a preliminary study on brain tumours.” submitted by Massimo Donelli, et, al. This manuscript proposed a novel approach for the segmentation of brain pathologies by using a growing area algorithm and a SVM based method. The article is well organized with a smooth flow of information during the explanation of each method. It is well-written, with very few clerical errors, and the style and layout are very well articulated. Overall, the authors clearly demonstrate their approach and detail the performance gain in this research field and the article meets the required standards for publication after minor edits.

Specific Comments

1. A related paper section is recommended, to further clarify the novelty compared with previous papers.
2. There were still some typos in the paper, I hope the authors to check the full text carefully.
3. It's important to share the code of the experiments to guarantee the reproducibility of the experiments and to assure the veracity of the results.
4. Check figure 2, consider adding more details to the description and legends for to explain it.

Author Response

The authors want to thank you for your valuable comments aimed at improve the quality of our work. In the following we tried to clarify your concerns.

Specific Comments

1. A related paper section is recommended, to further clarify the novelty compared with previous papers.

Thank you for your suggestion, we provided to insert a better description of the innovation introduced in this work with respect to the other state of the art methodologies in the introduction section. Thank you.

2. There were still some typos in the paper, I hope the authors to check the full text carefully.

The body of the manuscript has been accurately revised and the minor typos and grammar errors corrected. Thank you.

3. It's important to share the code of the experiments to guarantee the reproducibility of the experiments and to assure the veracity of the results.

We are certainly open to share the data and the related code with anyone. We are happy to collaborate with other research group that can help us of validate our methodology, the dataset and the code can be shared if requested. The goal is to improve this tool since we think that it could be of great interest for the collectivity involved in this important research area. Thank you for your interest.

4. Check figure 2, consider adding more details to the description and legends for to explain it-

Following your kind suggestion we provided to improve the description and the details of Fig. 2.