Abstract

Mammogram is the gold standard for early breast cancer screening, and its integration with deep learning-based computer-aided diagnosis (CAD) models has demonstrated significant advantages in improving the accuracy of breast cancer diagnosis. However, due to differences in mammography acquisition protocols and scanner models, significant inter-domain variations exist in images obtained from different mammography devices. As deep learning models tend to overfit to domain-specific feature representations during training, models trained on source domain often experience notable performance degradation when applied to cross-domain data, hindering their deployment in dynamic clinical settings. Therefore, this paper proposes a novel domain generalization approach for mammogram classification by suppressing domain-specific features (MC-SDS). MC-SDS first employs an adaptive channel filter to identify and drop channels that have a tendency to capture domain-specific features to suppress domain-specific features. Then, by perturbing the low-frequency components, the model is encouraged to learn from the high-frequency parts, further suppressing the domain-specific features present in the low-frequency components. Experiments conducted on a public dataset and two internal datasets demonstrate that MC-SDS outperforms other benchmark methods.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/3338_paper.pdf

SharedIt Link: Not yet available

SpringerLink (DOI): Not yet available

Supplementary Material: Not Submitted

Link to the Code Repository

N/A

Link to the Dataset(s)

N/A

BibTex

@InProceedings{CheJiq_Domain_MICCAI2025,
        author = { Chen, Jiqun and Sun, Luhao and Jiang, Wenzong and Liu, Weifeng and Li, Chao and Yu, Zhiyong and Liu, Baodi},
        title = { { Domain Generalization for Mammogram Classification by Suppressing Domain-Specific Features } },
        booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
        year = {2025},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 15966},
        month = {September},
        page = {175 -- 185}
}


Reviews

Review #1

  • Please describe the contribution of the paper

    The main contribution of the paper is the development of a deep learning-based approach that enhances the accuracy of breast cancer detection in mammograms. The method focuses on bridging the gap between different domains by leveraging domain adaptation techniques to improve the generalizability of the model across various conditions. Additionally, the paper presents an innovative framework that uses advanced neural network architectures to automatically extract key features from mammographic images, facilitating early detection and diagnosis of breast cancer while addressing the challenge of limited annotated medical data.

  • Please list the major strengths of the paper: you should highlight 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 major weaknesses of the paper are as follows: Domain Adaptation Approach: The paper introduces a unique domain adaptation framework tailored for medical imaging, particularly focusing on bridging the gap between source and target domains. This method addresses the common challenge of limited annotated medical data, making the model more adaptable to real-world clinical settings where labeled data is sparse.

    Advanced Neural Network Architecture: The proposed deep learning architecture for breast cancer detection in mammograms leverages state-of-the-art techniques, enhancing the model’s ability to automatically learn and extract features from the images. This improves detection accuracy and robustness, especially in complex and diverse datasets.

    Clinical Feasibility: The model is designed to be easily deployable in clinical environments, demonstrating the feasibility of applying cutting-edge AI techniques to real-world medical scenarios. The framework’s potential for early detection and diagnosis of breast cancer is a significant strength, contributing to more effective healthcare.

    Evaluation Across Multiple Datasets: The paper presents comprehensive evaluation results across multiple datasets, highlighting the robustness and generalizability of the proposed method. By testing on varied data sources, the authors show that the model can perform well even in the presence of data variability, a key factor for clinical adoption.

  • Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.

    The major weaknesses are as follows:

    Lack of Detailed Comparison with State-of-the-Art Methods: While the paper demonstrates the effectiveness of the proposed domain adaptation technique, there is a lack of in-depth comparison with existing state-of-the-art methods in breast cancer detection using mammograms like DMASTER etc. A more detailed comparison with other advanced models, including their strengths and weaknesses, would have strengthened the evaluation and helped readers understand the relative contribution of the proposed method.

    Limited Real-World Testing: The clinical feasibility of the model is only discussed in theory, with no direct evidence of its performance in real-world clinical settings. It would be beneficial to include more details or even preliminary results from clinical trials or collaborations with healthcare institutions to demonstrate the model’s practical impact and accuracy in diverse, uncontrolled environments.

    Limited Consideration of Class Imbalance: Medical image datasets, especially mammograms, often suffer from class imbalance (e.g., more negative cases than positive). The paper does not provide any specific strategies or techniques to handle this imbalance, which could affect the model’s ability to generalize to rare cases (e.g., small or early-stage tumors). Addressing this issue with methods like class weighting or oversampling could significantly improve the model’s robustness.

    Model Interpretability: While deep learning models have shown promise in breast cancer detection, they often suffer from a lack of interpretability, which can hinder clinical adoption. The paper does not discuss interpretability techniques, such as saliency maps or attention mechanisms, to explain the model’s decisions. In the context of medical diagnostics, having an interpretable model is crucial to gain clinicians’ trust and ensure that the AI system’s decisions align with medical expertise.

    Training Data Generalization: Although the model demonstrates good performance on multiple datasets, there is limited discussion on how the model generalizes to datasets outside the ones used for training. Given the variability in mammogram quality, patient demographics, and imaging equipment, it would be helpful to discuss how the model might perform on datasets from different sources or regions.

  • Please rate the clarity and organization of this paper

    Poor

  • 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.

  • Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2025/en/REVIEWER-GUIDELINES.html

    I would like to commend the authors for their approach to addressing domain adaptation in breast cancer detection. The proposed method for learning domain-invariant features holds great potential for improving the generalizability of models trained on medical images, which is a critical challenge in the field.

    A few additional suggestions for enhancing the work:

    Real-World Clinical Collaboration: As medical applications require high levels of clinical validation, I recommend exploring collaborations with healthcare institutions to test the proposed method in clinical settings. This could provide a valuable bridge between theoretical research and practical application.

    Data Augmentation and Handling Imbalance: Consider implementing or discussing strategies such as data augmentation or class balancing techniques to address the class imbalance often seen in medical imaging datasets. This could further improve model performance, especially in detecting rare instances like small tumors.

    Interpretability and Explainability: For clinical adoption, interpretability of deep learning models is crucial. Including a discussion on explainability techniques could enhance the model’s trustworthiness and make it more acceptable for use in real-world healthcare environments.

    Overall, this work is promising and has the potential to make a significant impact in the domain of medical image analysis. I look forward to seeing how the research progresses and how these potential improvements could further strengthen its application in clinical practice.

  • 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.

    (2) Reject — should be rejected, independent of rebuttal

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

    My decision to reject this paper stems from several key factors that I believe need to be addressed for it to meet the standards required for acceptance:

    Limited Novelty: While the proposed method for domain-invariant feature learning is an interesting direction, it does not introduce sufficient novelty when compared to existing approaches. The paper does not clearly establish how it significantly advances the current state of the art in domain adaptation for medical imaging, especially in breast cancer detection. For example, previous work such as [Reference] and [Reference] has explored similar techniques for domain adaptation, making the contribution feel incremental rather than groundbreaking.

    Evaluation: The evaluation of the method, while decent, lacks depth in terms of real-world clinical validation. The experiments conducted are confined to controlled datasets, and there is little discussion about how the method would perform in more challenging, real-world conditions where noise and variability in imaging are more pronounced. Additionally, the performance metrics used are standard, but a more detailed error analysis or failure cases could have strengthened the paper’s impact.

    Limited Clinical Relevance: Given the application of the research to medical imaging, the lack of collaboration with healthcare professionals or testing on diverse clinical datasets makes the work less clinically feasible. The authors could have demonstrated how the model would integrate into real-world healthcare workflows or discussed potential challenges in adopting this method in clinical practice.

    Methodology Clarity: The methodology section lacks some clarity, particularly regarding the implementation details of the domain-invariant feature learning process. This leaves questions about how generalizable the method truly is across different datasets and imaging modalities.

    In summary, while the paper is promising in some aspects, particularly in the direction of domain adaptation, it falls short in terms of novelty, clinical validation, and thorough evaluation. These weaknesses led me to conclude that the paper does not yet meet the criteria for acceptance at this stage.

  • Reviewer confidence

    Very confident (4)

  • [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.

    Accept

  • [Post rebuttal] Please justify your final decision from above.

    I am satisfied with the rebuttal, however the class imbalanced data can be a bottleneck for the proposed framework which authors have mentioned to include in the future work.



Review #2

  • Please describe the contribution of the paper

    The paper introduces MC-SDS, a novel domain generalization approach for mammogram classification that suppresses domain-specific features using an adaptive channel filter and perturbing low-frequency components. This method effectively improves model performance across different mammography devices, as demonstrated by superior results compared to benchmark methods on multiple datasets.

  • Please list the major strengths of the paper: you should highlight 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 paper’s main strengths include: Effective Feature Suppression: The combination of adaptive filtering and frequency perturbation effectively reduces domain-specific biases, enhancing model robustness across different mammography devices. Superior Performance: Experiments on multiple datasets demonstrate that MC-SDS outperforms benchmark methods, proving its effectiveness in improving classification accuracy and generalizability.

  • Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.

    The paper’s main shortcomings are:

    1. Insufficient Novelty**: The method (MC-SDS) appears to be a straightforward combination of DomainDrop and Aloft, lacking significant innovation beyond integrating these existing techniques.
    2. Inadequate Comparative Experiments**: The paper fails to directly compare MC-SDS with DomainDrop and Aloft individually, making it unclear whether the performance improvements are due to the combination itself or other factors.
    3. Lack of In-depth Explanation**: The paper does not provide detailed analysis on how the combination of these methods works synergistically or why it outperforms the individual methods, leaving gaps in understanding the underlying mechanisms.
  • 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 mention open access to source code or data but provides a clear and detailed description of the algorithm to ensure reproducibility.

  • Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2025/en/REVIEWER-GUIDELINES.html

    N/A

  • 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.

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

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

    The paper is well-written with good experimental design, but its overall rating is limited by insufficient novelty and inadequate comparative analysis. The method combines existing techniques without introducing groundbreaking innovations, and the lack of direct comparisons with the original methods makes it hard to assess the added value.

  • Reviewer confidence

    Confident but not absolutely certain (3)

  • [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.

    N/A

  • [Post rebuttal] Please justify your final decision from above.

    N/A



Review #3

  • Please describe the contribution of the paper

    The main contribution of the paper is the proposal of a novel domain generalization approach for mammogram classification, termed MC-SDS (Mammogram Classification by Suppressing Domain-Specific Features). MC-SDS addresses the challenge of performance degradation in deep learning models when applied to mammograms from different devices due to inter-domain variations. The approach introduces two key components: (1) a dropout-based Adaptive Channel Filter (ACF) that identifies and suppresses channels capturing domain-specific features and (2) a Low-Frequency Perturbation Module (LFPM) that further suppresses domain-specific features within the low-frequency components of the image. Through experiments on a public dataset and two internal datasets, the authors demonstrate that MC-SDS outperforms existing benchmark methods, indicating its effectiveness in improving the generalization capability of mammogram classification models across diverse imaging domains. The study provides a method to reduce the impact of device-specific artifacts in mammogram analysis, which could lead to more robust and reliable CAD systems for breast cancer screening.

  • Please list the major strengths of the paper: you should highlight 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.

    Novelty of the MC-SDS Framework:

    The MC-SDS framework introduces a unique combination of techniques to address domain generalization in mammogram classification. The integration of Adaptive Channel Filtering (ACF) and Low-Frequency Perturbation Module (LFPM) is a novel approach that hasn’t been previously explored.

    This is interesting because it directly tackles a significant challenge in medical imaging: the variability in image characteristics across different devices and institutions, which can severely impact the performance of deep learning models. By explicitly targeting domain-specific features, the method aims to learn more robust and generalizable representations.

    Adaptive Channel Filter (ACF):

    The dropout-based ACF is a novel component that adaptively identifies and suppresses channels that capture domain-specific features. By using a gradient reversal layer (GRL) and channel scoring based on domain classification performance, the ACF dynamically adjusts its filtering behavior.

    This is interesting because it allows the model to focus on learning domain-invariant features relevant to the actual pathology, rather than being influenced by device-specific artifacts or acquisition protocols. The multi-layer stochastic activation strategy is also an interesting way to control the amount of dropout and prevent excessive feature loss.

    Low-Frequency Perturbation Module (LFPM):

    The LFPM is another novel component that further suppresses domain-specific features by perturbing the low-frequency components of the image. The rationale behind this is that low-frequency components often contain stylistic information and device-specific artifacts, while high-frequency components capture essential anatomical structures.

    This is interesting because it provides a complementary approach to ACF by explicitly targeting domain-specific features in the frequency domain. The use of a Gaussian model to resample low-frequency spectra allows the model to explore different variations and learn more robust representations.

    Comprehensive Evaluation:

    The authors evaluated MC-SDS on a combination of a public dataset (INbreast) and two internal datasets (InH1 and InH2), which provides a good balance between reproducibility and real-world relevance.

    The use of multiple datasets with different characteristics demonstrates the generalizability of the proposed method across diverse imaging domains. The comparison with several domain generalization methods based on ResNet-50 further strengthens the validity of the results.

    Strong Results:

    The experimental results demonstrate that MC-SDS outperforms other benchmark methods. This indicates that MC-SDS is effective in suppressing domain-specific features and improving the generalization capability of mammogram classification models across diverse imaging domains.

    In summary, the key strengths of this paper lie in the novelty of the MC-SDS framework, the adaptiveness and effectiveness of the ACF and LFPM components, and the comprehensive evaluation on multiple datasets. The results suggest that MC-SDS has the potential to improve the robustness and reliability of CAD systems for breast cancer screening in real-world clinical settings.

  • Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.

    Limited Novelty of Individual Components: While the integration of ACF and LFPM is novel, the individual components have some overlap with existing techniques:

    Channel dropout: Dropout-based channel pruning or feature selection has been explored in prior work for model compression and robustness. For example, work by Li et al. (2016) [Filter Pruning for Accelerating Deep Convolutional Neural Networks] used L1 regularization to prune unimportant channels. The difference here is the gradient reversal layer for learning domain-specific features.

    Frequency domain manipulation: Modifying frequency components of images for various tasks (e.g., style transfer, image enhancement) is also not entirely new. The approach here, though, seems to be novel to mammogram classification.

    Gradient Reversal Layer: GRL is a known technique in domain adaptation.

    Lack of Theoretical Justification for LFPM: While the paper provides an intuitive explanation for perturbing low-frequency components, there’s a lack of theoretical justification or analysis to support this approach. For example, it’s not clear why a Gaussian distribution is the most appropriate model for resampling low-frequency spectra. Also, one might think that removing domain-specific information is better than perturbing it.

    Limited Ablation Studies: The paper lacks comprehensive ablation studies to evaluate the individual contributions of ACF and LFPM. While the authors evaluate different values of r, they don’t provide results for models trained with only ACF or only LFPM. This makes it difficult to assess the effectiveness of each component and to determine whether their combination is truly synergistic.

    Dataset Limitations: The paper relies on a combination of a public dataset (INbreast) and two internal datasets (InH1 and InH2). While this provides a reasonable evaluation, the limited size of the INbreast dataset (410 images) may raise concerns about the generalizability of the results. The two internal datasets are not publicly available which poses a reproducibility problem.

    Lack of Clinical Validation: Like many CAD papers, there is no demonstration of clinical feasibility, i.e., whether the improved cross-domain performance translates to better clinical outcomes.

    Reproducibility: It may be difficult to reproduce this paper’s finding given that two of the three datasets are private. Also, the source code is not provided.

  • 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 mention open access to source code or data but provides a clear and detailed description of the algorithm to ensure reproducibility.

  • Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2025/en/REVIEWER-GUIDELINES.html

    N/A

  • 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.

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

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

    The decision to accept this paper stems from its novel approach to a significant challenge in mammogram analysis: domain generalization. The paper addresses the real-world problem of performance degradation in deep learning models when applied to mammograms from different devices, a consequence of inherent inter-domain variations. The proposed MC-SDS (Mammogram Classification by Suppressing Domain-Specific Features) framework presents a promising solution by explicitly targeting and suppressing domain-specific features, thereby enhancing the generalization capability of mammogram classification models. MC-SDS introduces two key modules: an Adaptive Channel Filter (ACF) and a Low-Frequency Perturbation Module (LFPM), which work in tandem to reduce the model’s reliance on domain-specific cues and promote learning of more robust, clinically relevant features. The comprehensive and well-written introduction clearly explains the limitations of current domain generalization methods, especially the issue that previous methods impose constraints on the prediction layer, potentially disregarding the effect of the intermediate layers to learn excessive irrelevant information.

    The strength of the MC-SDS framework lies in its novel integration of these techniques. The ACF, which leverages a gradient reversal layer (GRL) and a multi-layer stochastic activation strategy, adaptively identifies and suppresses channels that contribute to domain classification, effectively pruning domain-specific information. The LFPM complements the ACF by perturbing the low-frequency components of the feature maps, based on the premise that these components often encode stylistic information and device-specific artifacts. The combination of these modules encourages the model to focus on domain-invariant features, such as the edges, shapes, and curvatures of lesions, which are crucial for breast cancer diagnosis. The experimental validation conducted on a public dataset (INbreast) and two internal datasets demonstrates that MC-SDS outperforms several benchmark methods, indicating its effectiveness in suppressing domain-specific features and improving generalization.

    While there is room for improvement by providing a more rigorous theoretical justification for the Gaussian model in LFPM and making the code public for reproducibility, the paper’s strengths in addressing a relevant problem, proposing a novel and well-justified solution, and demonstrating promising results justify its acceptance. The explanation of the domain shift problem in the introduction is well-written, and the description of the proposed method is clear and well-structured. The MC-SDS framework offers a valuable contribution to the field by enhancing the robustness and reliability of CAD systems for breast cancer screening in real-world clinical settings.

  • Reviewer confidence

    Very confident (4)

  • [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.

    N/A

  • [Post rebuttal] Please justify your final decision from above.

    N/A




Author Feedback

For Reviewer 1, 2, and 3:

  1. Regarding the innovation and methodology of MC_SDS: MC_SDS is a model specifically designed for domain generalization in mammogram classification, aiming to maximize the suppression of domain-specific features that may mislead the model. ACF is specifically designed for domain generalization in mammogram classification, including a gradient reversal layer, domain classification-based channel scoring, dynamic drop, and a multi-layer random activation strategy. The motivation for LFPM comes from our observations of the frequency-domain characteristics of mammogram. In MC_SDS, ACF suppresses more obvious domain-specific features by leveraging a domain classification task. LFPM complements ACF by perturbing the low-frequency part after ACF has suppressed some domain-specific features, guiding the model to further suppress domain-specific features in the low-frequency information. This innovative structure enables MC_SDS to achieve better performance in the experiments.
  2. Regarding the dataset, clinical validation, and code availability: Our internal dataset comprises real clinical data from a leading provincial cancer hospital, professionally annotated by physicians. We have engaged in multiple rounds of communication with the hospital. We plan further validation and clinical integration, with public release of data and code upon manuscript acceptance.

For Reviewer 1:

  1. Regarding the theoretical basis of LFPM: Low-frequency components in mammograms, caused by varying imaging protocols, show random, smooth statistics, well-modeled by a Gaussian distribution. Moreover, since the low-frequency part retains most of the image’s energy (i.e., the location information of the target), we believe it is inappropriate to directly remove the low-frequency components.
  2. Regarding ablation experiments: We have provided the results of models trained with only ACF and only LFPM in Table 2 (labeled as ACF and LFPM in the table).

For Reviewer 2:

  1. Regarding novelty: DomainDrop and ACF share some similarities, but considering the differences between each mammogram sample, ACF performs adaptive drop for each sample. In contrast, DomainDrop sets a fixed drop ratio without considering the differences between samples. LFPM complements ACF, motivated by our observations of the frequency-domain characteristics of mammogram.
  2. Regarding comparative experiments: MC_SDS is a model specifically designed for domain generalization in mammogram classification and is not a combination of DomainDrop and Aloft. Therefore, we believe that the ablation experiments in Table 2 (using ACF alone and LFPM alone) are sufficient to demonstrate the effectiveness of MC_SDS.

For Reviewer 3:

  1. Regarding comparison with SOTA methods: You mentioned that we did not compare MC-SDS with SOTA methods (such as DMASTER). DMASTER is a domain adaptation method, whereas our focus is domain generalization. While related, these are distinct areas - domain adaptation uses target domain data during training, while domain generalization prohibits this. We compared our method with SOTA domain generalization methods.
  2. Regarding consideration of class imbalance: We acknowledge that class imbalance may affect the model’s generalization ability for small tumors/early-stage tumors. However, the dataset used in this study contains many such cases, and MC_SDS achieves excellent performance. Addressing class imbalance will be a direction for future improvement.
  3. Regarding model interpretability: We used Grad-CAM technology to demonstrate the interpretability of the model. The relevant results are shown in Fig. 3.
  4. Regarding the generalization ability: The domain generalization we studied is precisely aimed at enabling the model to generalize to datasets outside the training dataset. Our experiments strictly followed the experimental settings of domain generalization. We believe this sufficiently demonstrates the generalization ability of MC_SDS.




Meta-Review

Meta-review #1

  • Your recommendation

    Invite for Rebuttal

  • If your recommendation is “Provisional Reject”, then summarize the factors that went into this decision. In case you deviate from the reviewers’ recommendations, explain in detail the reasons why. You do not need to provide a justification for a recommendation of “Provisional Accept” or “Invite for Rebuttal”.

    N/A

  • 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’

    N/A



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’

    This paper is recommended for acceptance, as all reviewers have reached a unanimous positive consensus.



Meta-review #3

  • 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’

    N/A



back to top