MSRL-Net: An Automatic Segmentation of Intracranial Hemorrhage for CT Images Based on the U-Net Framework

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Intracranial hemorrhage (ICH) occurs when blood leaks inside the skull as a result of trauma to the skull or due to medical conditions. ICH usually requires immediate medical and surgical attention because the disease has a high mortality rate, long-term disability potential, and other potentially life-threatening complications. To diagnose ICH, several imaging modalities are commonly used, including computed tomography (CT) and magnetic resonance imaging (MRI). In this paper the authors proposed MSRL to accurately segment the lesion regions in CT images of intra-14 cranial hemorrhage. They designed MSPool module to replace traditional max-pooling, reducing information loss during network down sampling and enhancing segmentation accuracy. They also compared with the existing methods and proposed method achieves better outcome.

Comments on the Quality of English Language

Minor editing of English language required.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

The study significantly contributes to our understanding of Intracranial Hemorrhage (ICH) and presents the development of the MSRL-Net.

The research acknowledges the severe impact of intracranial hemorrhage, a hemorrhagic condition affecting the brain and ventricles. It particularly highlights a pressing challenge in the medical image analysis domain: the subpar segmentation performance of the widely employed U-Net network when it comes to accurately identifying and outlining small lesion areas in CT images of intracranial hemorrhage.

In response to this challenge, the paper introduces the MSRL-Net, a novel convolutional neural network explicitly tailored for precise lesion region segmentation in CT images of intracranial hemorrhage. This network represents an innovative solution to enhance segmentation accuracy in this critical medical application.

The authors propose an ingenious strategy that amalgamates max-pooling and soft-pooling techniques to address the issue of information loss during the downsampling process. This novel approach substantially aids in retaining crucial image details, strengthening the network's ability to discern small lesion areas.

The research acknowledges the imbalanced distributions of healthy and lesion regions in medical images. 

The authors introduce the MRHDC (Multi-Receptive-Field High-Dimensional Context) module at the network's bottleneck layer. This module is purpose-built to capture intricate spatial information within image features and acquire multi-scale features with distinct receptive fields. This feature is indispensable for precisely segmenting intracranial hemorrhage lesions, especially the small and intricate ones.

The study provides substantial empirical evidence of the MSRL-Net's performance, reporting an average Dice coefficient of 0.712 on a comprehensive dataset. The Dice coefficient, a widely recognized metric in image segmentation, quantifies the degree of overlap between predicted and ground truth segmentation masks. An average Dice coefficient of 0.712 indicates a remarkably high level of accuracy in segmenting intracranial hemorrhage lesions, affirming the model's effectiveness.

The experimental findings presented in the study underscore the substantial clinical promise of the MSRL-Net. Accurate segmentation of intracranial hemorrhage lesions in CT images is pivotal for medical diagnosis and treatment planning, and the MSRL-Net's performance suggests its potential as a valuable asset in the medical field.

However, the Crucial point is that precision is significantly lacking in the retrieval phase. The authors should replant the experiments and results.

Comments on the Quality of English Language

The sentences in the text are long and complex, making it difficult to follow the intended meaning. It would be beneficial to break these long sentences into shorter, more concise ones for clarity and ease of understanding.

There needs to be more consistency in verb tenses. The text starts with present tense ("we propose") but then switches to past tense ("the implementation shows"). To maintain consistency, consider saying, "In this study, we proposed a new Unet network structure called MRSL-Net, which was used to segment intracranial hemorrhage. The implementation showed that the MRSL-Net method was superior to other methods."

The sentence "Segmentation results can directly help the doctor to observe the bleeding area effectively" is somewhat awkward. It could be made more concise and clear: "The segmentation results directly assist doctors in 

The phrase "our method" is used twice quickly, making the text seem repetitive. Consider rephrasing to avoid redundancy. For instance, "Although our approach demonstrates good segmentation results, it still has several limitations."

 Phrases like "it still has many shortcomings" and "more features of focal areas" are vague. Specifying the shortcomings and elaborating on these "features of focal areas" would be helpful.

 The text has minor punctuation and grammatical issues, such as the need for more commas to separate clauses in long sentences and inconsistent capitalization. Proofreading for such problems is necessary.

Author Response

Sorry, I don't understand the meaning of the replanting experiment and the results. I have changed the comparison table of experimental results, please check it. I also corrected all the other English issues you mentioned    Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

Why author used U-Net architecture instead of other architecture?

Please mention all the hyperparameter used in your proposed work.

Why the author used ADAM optimizer?

What is the significance of binary cross-entropy (BCE)loss function in your article?

What is the significance of MRHDC module in your article?

Discuss more about the findings of your work, why your work is more accurate as compare to the existing work using different parameters.

Please enhance the resolution of figures for better readability.

Please mention the limitations of your proposed approach.

How your work is useful for the society? Please discuss it in conclusion section.

Why the author have not used transfer learning or fuzzy ensemble techniques for better accuracy?

More recent references can be added for more up to date work.

Altameem, A., Mahanty, C., Poonia, R. C., Saudagar, A. K. J., & Kumar, R. (2022). Breast Cancer Detection in Mammography Images Using Deep Convolutional Neural Networks and Fuzzy Ensemble Modeling Techniques. Diagnostics, 12(8), 1812.

 

Comments on the Quality of English Language

Minor editing of English language required

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

The paper has been conducted with excellence. However, it does require some minor refinements in terms of English and editing. Congratulations on your work!

Comments on the Quality of English Language

The paper has been conducted with excellence. However, it does require some minor refinements in terms of English and editing. Congratulations on your work!

Reviewer 3 Report

Comments and Suggestions for Authors

Authors have modified the article as per the review comments.