Paper Info Reviews Author Feedback Meta-Review Back to top
List of Papers Browse by Subject Areas Author List

Abstract

Accurate brain tissue segmentation is a vital prerequisite for charting infant brain development and for diagnosing early brain disorders. However, due to inherently ongoing myelination and maturation, the intensity distributions of gray matter (GM) and white matter (WM) on T1-weighted (T1w) data undergo substantial variations in intensity from neonatal to 24 months. Especially at the ages around 6 months, the intensity distributions of GM and WM are highly overlapped. These physiological phenomena pose great challenges for automatic infant brain tissue segmentation, even for expert radiologists. To address these issues, in this study, we present a unified infant brain tissue segmentation (UinTSeg) framework to accurately segment brain tissues of infants aged 0-24 months using a single model. UinTSeg comprises two stages: 1) boundary extraction and 2) tissue segmentation. In the first stage, to alleviate the difficulty of tissue segmentation caused by variations in intensity, we extract the intensity-invariant tissue boundaries from T1w data driven by edge maps extracted from the Sobel filter. In the second stage, the Sobel edge maps and extracted boundaries of GM, WM, and cerebrospinal fluid (CSF) are utilized as intensity-invariant anatomy information to ensure unified and accurate tissue segmentation in infants age period of 0-24 months. Both stages are built upon an attention-based surrounding-aware segmentation network (ASNet), which exploits the contextual information from multi-scale patches to improve the segmentation performance. Extensive experiments on the baby connectome project dataset demonstrate the superiority of our proposed framework over five state-of-the-art methods.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2024/paper/0074_paper.pdf

SharedIt Link: pending

SpringerLink (DOI): pending

Supplementary Material: N/A

Link to the Code Repository

N/A

Link to the Dataset(s)

https://www.humanconnectome.org/study/lifespan-baby-connectome-project

BibTex

@InProceedings{Liu_UinTSeg_MICCAI2024,
        author = { Liu, Jiameng and Liu, Feihong and Sun, Kaicong and Sun, Yuhang and Huang, Jiawei and Jiang, Caiwen and Rekik, Islem and Shen, Dinggang},
        title = { { UinTSeg: Unified Infant Brain Tissue Segmentation with Anatomy Delineation } },
        booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2024},
        year = {2024},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 15002},
        month = {October},
        page = {pending}
}

Reviews

Review #1

  • Please describe the contribution of the paper

    While in the literature segmentation of infant brain tissue is based upon registration and learning, the authors propose a unified infant brain tissue segmentation, from neonatal to 24 months in a single model called UinTSeg.

  • Please list the main strengths of the paper; you should write about a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.

    The strengh of this paper is the creation of a single model to segment infant brain tissues. They propose an original architecture not yet provided.

  • Please list the main weaknesses of the paper. Please provide details, for instance, if you think a method is not novel, explain why and provide a reference to prior work.

    This paper presents several weaknesses in the way the novelty of the research is presented (do they create the CBF and the loss?), in the citation of previous works, in the reproducibility of their framework (some details are not given such as parameters, dimensions and formula), in main comparisons (some are not shown in all the figures and tables specifically with TMSN) while the architecture and results seems important in this field.

  • Please rate the clarity and organization of this paper

    Satisfactory

  • Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.

    The submission does not provide sufficient information for reproducibility.

  • Do you have any additional comments regarding the paper’s reproducibility?

    As described in 6 and 10, there is a lack of details concerning the architecture (what are the skip connection used, the composition of encoder/decoder, the formula of the hybrid dice-ce loss, or the training for the second part of the network for instance). A link of the code is given but does not work.

  • Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review. Pay specific attention to the different assessment criteria for the different paper categories (MIC, CAI, Clinical Translation of Methodology, Health Equity): https://conferences.miccai.org/2024/en/REVIEWER-GUIDELINES.html

    To improve the paper, such as described in 6 and 10, I suggest :

    • The authors may say the CSF is equivalent to the cerebrospinal fluid in the abstract (as done for GM and WM). This acronym is not defined in all the text.
    • “Variations in intensity” in the introduction may be written in the same syntax as other words. The same may be done for Sobel (once in italic, once not) and other groups of words.
    • In figure 1 the authors may describe the axes in the three graphics shown. The authors may add a legend for the acronyms too.
    • “…which is proven effective in enhancing segmentation performance.” is mentioned in the introduction. Authors may cite a publication to prove it.
    • Does the attention-based cross-branch fusion (CBF) is a creation from the authors or taken from literature? The authors may mention in the introduction a citation, or explain the origin and interest of using this module if they created it.
    • The authors may add LP and SP in the Figure 2 as it is described in the text for a quickest understanding of the architecture. Should it be possible to add the dimensions/resolution of the patches in the figure or the text?
    • Which kind of skip-connections has been used in this four stage encoder decoder, mentioned page 4?
    • At the beginning of page 5, same as previously, may it be possible to add the original dimensions and the dimension of cropped images?
    • The authors may add the formula of their hybrid loss and the interest of this loss. Do the authors create or do they apply a hybrid loss taken from the literature?
    • How is the second part of the network trained?
    • The link of the github code is not given (blank page).
    • The TSMN line should be given in table 1 for a fair comparison (in the second table this architecture renders a best result and it could be interesting to see the results for the first table too with this architecture).
    • “We can find that UinTSeg consistently shows superior performance over the five SOTA methods and even achieves better performance as the method TMSN [10], which is specifically tailored for the isointense phase.” This sentence is almost true for table 2 but not for table 1, where the TMSN line is missing… The authors must give the results for this architecture.
    • Authors should add the results of the TMSN architecture in Figure 3 too.
    • Table 3 : 6.76±2.4 for HD, column GM, line ASNet should be in bold not the UInTSeg line.
  • Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making

    Weak Reject — could be rejected, dependent on rebuttal (3)

  • Please justify your recommendation. What were the major factors that led you to your overall score for this paper?

    This paper could be of a great interest in the field and could be accepted if some main issues are corrected. The main issues of this paper are linked with the fact some comparisons are cited in the paper and not shown in the tables and figures. Then, the reproducibility is in my opinion compromised with the lack of details concerning the architecture.

  • Reviewer confidence

    Very confident (4)

  • [Post rebuttal] After reading the author’s rebuttal, state your overall opinion of the paper if it has been changed

    N/A

  • [Post rebuttal] Please justify your decision

    N/A



Review #2

  • Please describe the contribution of the paper

    This paper presents a unified framework called UinTSeg for segmenting brain tissues (gray matter, white matter, and CSF) in infant MRI scans from 0 to 24 months of age using a single model. The key novelties are:

    1. A two-stage approach with the first stage extracting tissue boundaries using Sobel edge maps, and the second stage incorporating those boundaries to guide tissue segmentation.
    2. An attention-based surrounding-aware segmentation network (ASNet) that leverages multi-scale contextual information.
    3. Ability to handle the “isointense phase” around 6-9 months where GM and WM intensities are highly overlapped by synthesizing adult-like data first.
  • Please list the main strengths of the paper; you should write about a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.

    Main strengths of the paper: 1) Tackling an important and challenging problem of infant brain segmentation across the full 0-24 month age range with a unified model.

    2)Novel idea of extracting intensity-invariant boundaries to help mitigate intensity distribution shifts with development.

    3)Promising quantitative and qualitative results showing improvements over several state-of-the-art methods.

    4)Technically sound approach with good experimental evaluation.

  • Please list the main weaknesses of the paper. Please provide details, for instance, if you think a method is not novel, explain why and provide a reference to prior work.

    Weaknesses of the paper: 1) The boundary extraction idea is interesting but seems quite dependent on the Sobel filter perrformance which has limitations.

    2) No comparison to registration/atlas-based methods

    3)Some lack of clarity in writing and figures (e.g. Fig 2 is quite dense).

    4)The dataset size of 672 scans is relatively modest for a deep learning method, especially when divided into training/validation/test sets. There is a risk of overfitting or lack of generalization that should be discussed.

    5)No external validation on data from different sites/scanners, which is important to ensure the method generalizes beyond the single dataset used.

  • Please rate the clarity and organization of this paper

    Good

  • Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.

    The authors claimed to release the source code and/or dataset upon acceptance of the submission.

  • Do you have any additional comments regarding the paper’s reproducibility?

    No

  • Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review. Pay specific attention to the different assessment criteria for the different paper categories (MIC, CAI, Clinical Translation of Methodology, Health Equity): https://conferences.miccai.org/2024/en/REVIEWER-GUIDELINES.html

    1) Why have you use Sobel edge detector? This is a very basic hand-crafted edge detector that may struggle with the complexity and variability of infant brain MRI data. More advanced or learned edge detectors may provide better boundary maps.

    2) Method is focused on segmenting tissues into just 3 classes (GM/WM/CSF). Are these classes enough? In some infant studies they require segmenting more detailed structures.

    3) It seems there is no comparison to registration/atlas-based methods in your SATA analysis. Registration/atlas-based methods are still commonly used in clinical practice for infant brain segmentation, so a comparison to these more classical techniques it could be useful to understand the contribution of this work.

    4) Figure 2 is very dense with a lot of detail crammed into one figure, making it difficult to parse.

    5) According to what you choose the selected 672 scans? With a dataset of only 672 scans split into training/validation/test, there is a risk that the model is overfitting to the limited data. Deep learning methods often require very large datasets to properly learn generalizable representations, especially for complex tasks like medical imaging. The authors should provide insights into steps taken to mitigate overfitting risks.

    6)Evaluating only on the BCP dataset from which the training/test data was drawn provides an overly optimistic assessment of performance. Variations in scanner models, imaging protocols, patient demographics etc. across sites can significantly impact model generalization. Lack of external validation raises concerns about the clinical applicability and generalization ability of the proposed approach.

  • Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making

    Weak Accept — could be accepted, dependent on rebuttal (4)

  • Please justify your recommendation. What were the major factors that led you to your overall score for this paper?

    The paper is technically sound and makes useful contributions, but has some weaknesses in the approach should be addressed in the rebuttal.

  • Reviewer confidence

    Confident but not absolutely certain (3)

  • [Post rebuttal] After reading the author’s rebuttal, state your overall opinion of the paper if it has been changed

    Weak Accept — could be accepted, dependent on rebuttal (4)

  • [Post rebuttal] Please justify your decision

    The author has clarified some of my doubts; however, I am still not convinced by some of the answers provided on the following points:

    1. The Sobel filter used has limitations, and it is not clear to me why it was the right choice.

    2. The dataset size of 672 scans is relatively small, raising concerns about the risk of overfitting or lack of generalization.

    For these reasons, I would like to maintain my initial score.



Review #3

  • Please describe the contribution of the paper

    In this paper, the authors propose UniTSeg, which is a segmentation method that divides T1-weighted images of infant brains into GM, WM, and CSF. UniTSeg consists of a boundary extraction stage and a tissue segmentation stage using ASNet to extract global and local features. Experiments on a public dataset demonstrate the effectiveness of UniTSeg.

  • Please list the main strengths of the paper; you should write about a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.

    The strengths of this paper are as follows.

    • The edge maps obtained by Sobel filtering are combined with T1-weighted images to enhance the differences between tissues.
    • ASNet was proposed to improve the segmentation accuracy by connecting 3D U-Nets, which handles both global and local patches, with a cross-branch fusion module.
    • A two-stage approach was introduced, where the boundaries are extracted and then tissue segmentation is performed.
  • Please list the main weaknesses of the paper. Please provide details, for instance, if you think a method is not novel, explain why and provide a reference to prior work.

    The weaknesses of this paper are as follows.

    • This paper is targeted at infant, but it does not explain how UniTSeg is specialized for infant.
    • UniTSeg uses edge maps obtained with the Sobel filter, but since there are other edge detection methods besides the Sobel filter, it is necessary to explain why the Sobel filter was chosen.
    • The structure of UniTSeg is unclear: it does not explain why the edge map obtained with the Sobel filter is also used in Stage 2, even though the boundary was extracted in Boundary Extraction.
    • The three phases of the infant are shown in Fig. 1. In the experiment, no evaluation was done for each phase.
    • As shown in Fig. 1, the head size of the infant changes significantly with the age of the child. Therefore, the segmentation results may vary depending on the head size. In general, head size is normalized before segmentation of T1-weighted images, but UniTSeg does not explain whether head size is normalized or not.
  • Please rate the clarity and organization of this paper

    Satisfactory

  • Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.

    The authors claimed to release the source code and/or dataset upon acceptance of the submission.

  • Do you have any additional comments regarding the paper’s reproducibility?

    N/A

  • Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review. Pay specific attention to the different assessment criteria for the different paper categories (MIC, CAI, Clinical Translation of Methodology, Health Equity): https://conferences.miccai.org/2024/en/REVIEWER-GUIDELINES.html

    • Although it is an age-specific model, a comparison with [19] and others would emphasize the effectiveness of UniTSeg.

    • It would be good to consider how much accuracy depends on the edge map of the Sobel filter, and a comparison with other edge detection methods besides the Sobel filter would be helpful. Although the edge map is combined with the T1-weighted image, an alternative approach would be to sharpen the T1-weighted image.

    • In Fig. 2, the input/output arrows to the CFM module are incorrect: the CBF Module is a process for each layer, but it is not shown as such in Fig. 2 (a) and (b).

    • Although the split ratio of the dataset is described, the number of subjects for each is not indicated. Not only the number of data but also the number of subjects should be described.

    • As shown in the infant phases in Fig. 1, the brain morphology and size are different in each phase. Therefore, it is necessary to describe the number of data corresponding to each phase as well as the evaluation in each phase.

    • It is easier to understand the ablation study in Table 3 if the notation is based on UniTSeg. It would also be helpful to evaluate the accuracy of adding Sobel filters and boundary extraction to SegNet.

  • Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making

    Weak Accept — could be accepted, dependent on rebuttal (4)

  • Please justify your recommendation. What were the major factors that led you to your overall score for this paper?

    There is novelty as a brain segmentation method, and the effectiveness of the proposed method has been demonstrated experimentally. On the other hand, it is unclear whether UniTSeg is infant-specific and has not been evaluated according to the infant phase. Based on the quality of the paper and with the expectation that additional explanations will be provided in the rebuttal, I have rated the paper as weak accept.

  • Reviewer confidence

    Very confident (4)

  • [Post rebuttal] After reading the author’s rebuttal, state your overall opinion of the paper if it has been changed

    Weak Accept — could be accepted, dependent on rebuttal (4)

  • [Post rebuttal] Please justify your decision

    Thank you for providing additional explanations for the reviewers’ comments. I understand the claims made by the authors. On the other hand, I would have liked the authors to have emphasized the infant-specific aspects of the method in their initial submission. Since the novelty of the method is recognized, I decided that this paper is weak accept as well as the initial rate.



Review #4

  • Please describe the contribution of the paper

    This paper addresses the overlapping and confusion caused by intensity distributions of Gray Matter and white Matter and cerebrospinal fluid in infants from 0-24 months; by proposing a two stage model for boundary extraction and tissue segmentation using ASNet as baseline for comprehensive contextual information.

  • Please list the main strengths of the paper; you should write about a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.
    • Addressing a critical matter in segmenting infant brain tissue with special focus on isointense phase.
    • Smart use of Sobel filter for boundary extraction based on intensity-invariant anatomy information.
    • Using an Attention based surrounding aware segmentation network as a baseline for contextual information.
    • Capturing fine details from local patches with consideration of contextual information from surrounding patches.
    • A critical but successful use of a synthesis model to transform isointense into adult like samples.
  • Please list the main weaknesses of the paper. Please provide details, for instance, if you think a method is not novel, explain why and provide a reference to prior work.
    • Not proving the efficiency (Or Not efficiency) of weight sharing between the two stages, choice not justified.
  • Please rate the clarity and organization of this paper

    Very Good

  • Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.

    The authors claimed to release the source code and/or dataset upon acceptance of the submission.

  • Do you have any additional comments regarding the paper’s reproducibility?

    N/A

  • Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review. Pay specific attention to the different assessment criteria for the different paper categories (MIC, CAI, Clinical Translation of Methodology, Health Equity): https://conferences.miccai.org/2024/en/REVIEWER-GUIDELINES.html

    Consider providing the CSF Cerebrospinal fluid refrence before using abbreviation.

  • Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making

    Accept — should be accepted, independent of rebuttal (5)

  • Please justify your recommendation. What were the major factors that led you to your overall score for this paper?

    An intuitive use of classical methods over new architectures to improve infant brain tissue segmentation, Proving it’s efficacy over SOTA architectures, and good results over different infant phases, remarkable on the isointense phase.

  • Reviewer confidence

    Confident but not absolutely certain (3)

  • [Post rebuttal] After reading the author’s rebuttal, state your overall opinion of the paper if it has been changed

    Strong Accept — must be accepted due to excellence (6)

  • [Post rebuttal] Please justify your decision

    Authors have provided enough justifications for techniques choices along with including more benchmark results in final paper.



Author Feedback

We thank reviewers for their constructive comments. We address concerns one-by-one below.

-Clarification of contributions of UinTSeg (R1,R3,R5,&R6) Thanks for pointing out. To our knowledge, previous works mainly designed age-specific methods to deal with dramatic intensity-variation of infant in the first two postnatal years, which used to introduce longitudinally inconsistent segmentations. In our work, we proposed an integrated framework, UinTSeg, to achieve longitudinally-consistent and accurate infant brain tissue segmentation for infants aged from 0 to 24 months.

-Clarification of choice of Sobel filter (R1&R3) In our UinTSeg framework, tissue edges serve as intensity-invariant anatomical information to guide longitudinally-consistent tissue segmentation for 0-24 month old infants. To this end, we mainly employ a Sobel edge filter to extract initial edges in the first and second stages. We agree with reviewers that different edge filters would affect the segmentation performance, for which we will provide comparison results in the final paper. Nevertheless, current results already demonstrate the effectiveness of intensity-invariant anatomy information in segmentation.

-Clarifications of the results (R1,R3,&R5) 1) Except for comparing with Infant FreeSurfer (the latest and most advanced registration-based infant brain segmentation method) with results shown in Tables 1 & 2, we did not include results from other registration-based methods since their results are not good. 2) Table 2 further illustrates the results obtained with TMSN, which is the leading infant brain segmentation method tailored for isointense phase. All these results show that UinTSeg has comparable or even superior segmentation performance to all age-specific infant brain segmentation methods.

-Clarifications of BCP dataset and the generalizability issue (R1,R3,&R5) The generalizability of UinTSeg is twofold: 1) UinTSeg is trained and evaluated on BCP data, gathered from diverse sites, scanners, and demographics (10.1016/j.neuroimage.2018.03.049), enhancing its robustness; 2) UinTSeg leverages intensity-invariant anatomical information, to deal with variations from different datasets, scanners, and even ages. In fact, the generalizability of UinTSeg was validated through further experiments: 1) segmenting tissues from the dHCP and NDAR datasets, and 2) segmenting tissues across lifespan datasets covering ages 0 to 100 years. Due to the extensive results, we plan to include them in our journal paper.

-Clarifications of the CBF module (R3&R5) Inspired by HF-UNet (10.1109/TMI.2021.3072956), the CBF module employs dual-attention mechanisms to integrate features from the two segmentation branches in ASNet, thus enhancing tissue segmentation with large receptive field patches. We will revise Fig. 2 for better clarity.

-Clarification of specificity for infants (R3) Due to rapid maturation and myelination of the brain during infancy, infants experience significant changes in brain tissue contrast. We thus proposed intensity-invariant anatomical features to guide segmentation using a single model.

-Clarification of data and brain size (R3&R5) We mainly employed BCP data, consisting of 264 subjects with 672 scans aged 0-24 months. To avoid introducing artifacts that could reduce segmentation accuracy, we did not normalize head sizes and only normalized brain sizes for training the synthesizing model. The results demonstrate that UniTSeg effectively handle such variations.

-Figure readability and description clarity (R1,R3,R5,&R6) We appreciate feedback and will address these issues pointed out by all the reviewers, including, descriptions of figures and proper citations.

-Model reproducibility (R1,R3,&R5) We will make the code publicly available upon acceptance.

-Hybrid loss selection (R5) Inspired by 10.1016/j.media.2021.102035, we utilize a hybrid loss, Dice and cross-entropy loss, which have been widely adopt for medical image segmentation tasks.



Meta-Review

Meta-review #1

  • After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.

    Accept

  • Please justify your recommendation. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’

    The majority of reviewers see value in this work and vote for acceptance. The main issue identified is the Sobel filtering, but the authors will provide more explanation on this in their camera-ready version. The rebuttal also addresses most of the concerns raised. The final version should include detailed information on the loss function and the CBF module, as provided in the rebuttal.

  • What is the rank of this paper among all your rebuttal papers? Use a number between 1/n (best paper in your stack) and n/n (worst paper in your stack of n papers). If this paper is among the bottom 30% of your stack, feel free to use NR (not ranked).

    The majority of reviewers see value in this work and vote for acceptance. The main issue identified is the Sobel filtering, but the authors will provide more explanation on this in their camera-ready version. The rebuttal also addresses most of the concerns raised. The final version should include detailed information on the loss function and the CBF module, as provided in the rebuttal.



Meta-review #2

  • After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.

    Accept

  • Please justify your recommendation. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’

    Agree with all reviewers.

  • What is the rank of this paper among all your rebuttal papers? Use a number between 1/n (best paper in your stack) and n/n (worst paper in your stack of n papers). If this paper is among the bottom 30% of your stack, feel free to use NR (not ranked).

    Agree with all reviewers.



Meta-review #3

  • After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.

    Reject

  • Please justify your recommendation. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’

    After much thought, I decided to reject the paper. The reviewers were mixed for this paper and I believe there are certain aspects of the paper which are indeed interesting especially with respect to the problem statement. In terms of the method, I see issues with generalization because of the delineation and reliance of robustness of Sobel filters. This also shows in the less difference in terms of performance when compared to other methods. So, I do not feel there is much new findings interesting to the community imparted by this paper.

  • What is the rank of this paper among all your rebuttal papers? Use a number between 1/n (best paper in your stack) and n/n (worst paper in your stack of n papers). If this paper is among the bottom 30% of your stack, feel free to use NR (not ranked).

    After much thought, I decided to reject the paper. The reviewers were mixed for this paper and I believe there are certain aspects of the paper which are indeed interesting especially with respect to the problem statement. In terms of the method, I see issues with generalization because of the delineation and reliance of robustness of Sobel filters. This also shows in the less difference in terms of performance when compared to other methods. So, I do not feel there is much new findings interesting to the community imparted by this paper.



back to top