List of Papers Browse by Subject Areas Author List
Abstract
Although the utilization of multi-stacks can solve fetal MRI motion correction and artifact removal problems, there are still problems of regional intensity heterogeneity, and global consistency discrimination in 3D space. To this end, we propose a novel coarse-to-fine self-supervised fetal brain MRI Radiation Diffusion Generation Model (RDGM). Firstly, we propose a novel self-supervised regionally Consistent Implicit Neural Representation (CINR) network with a double-spatial voxel association consistency mechanism to solve regional intensity heterogeneity. CINR enhances regional 3D voxel association and complementarity by two-voxel mapping spaces to generate coarse MRI. We also fine-tune the weighted slice reconstruction loss to improve the network reconstruction performance. Moreover, we propose the Global Diffusion Discriminative Generation (GDDG) fine module to enhance volume global consistency and discrimination. The noise diffusion is used to transform the global intensity discriminant information in 3D volume. The experiments on two real-world fetal MRI datasets demonstrate that RDGM achieves state-of-the-art results.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2024/paper/3884_paper.pdf
SharedIt Link: https://rdcu.be/dV5Dc
SpringerLink (DOI): https://doi.org/10.1007/978-3-031-72104-5_32
Supplementary Material: N/A
Link to the Code Repository
N/A
Link to the Dataset(s)
N/A
BibTex
@InProceedings{Tan_Fetal_MICCAI2024,
author = { Tan, Junpeng and Zhang, Xin and Qing, Chunmei and Yang, Chaoxiang and Zhang, He and Li, Gang and Xu, Xiangmin},
title = { { Fetal MRI Reconstruction by Global Diffusion and Consistent Implicit Representation } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2024},
year = {2024},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15007},
month = {October},
page = {329 -- 339}
}
Reviews
Review #1
- Please describe the contribution of the paper
This paper propose a method for SVR. The authors adopted INR with hash encoding and added a double-spatial component to address the regional consistency. A diffusion process is used to further improved the performance. The method is evaluated on both realistic and simulated data.
- 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.
- There are plenty experiments and results on both simulated and realistic data.
- The proposed method made additional changes to the baselines such as NeSVoR.
- The performance of the proposed method shows significant improvements.
- 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 motivation of combining diffusion model and INR is not clearly told. It looks more like simply mixing two popular methods. I’m confused about the combination. NeSVoR is proposed to accelerate the recon process with less memory consumption, the authors combine it with diffusion model which will absolutely make the recon process much longer.
- In equation 10 and 11, where is the sampled noise term? It looks like the diffusion process that the authors used is deterministic.
- Even though the quantitative metrics indicate improvements, there should be more reconstructed figures for each ablation experiments. For example, I’d like to see the difference between NeSVoR and baseline + double-spatial component.
- 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.
- 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
- There are many typos such as Page 4, ‘wihch’.
- The running time of the propose method should be given. NeSVoR is an efficient method in terms of recon time.
- More recon results are needed for ablation experiments as I mentioned in the weakness section.
- 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?
Even though this paper shows significant improvements on both realistic and simulated data, I found the motivation of combining diffusion model and INR not clear. I’d like to hear more from the authors in the rebuttal to address the concerns I mentioned in the weakness and comments sections.
- Reviewer confidence
Somewhat confident (2)
- [Post rebuttal] After reading the author’s rebuttal, state your overall opinion of the paper if it has been changed
Weak Reject — could be rejected, dependent on rebuttal (3)
- [Post rebuttal] Please justify your decision
I will keep the same score.
Review #2
- Please describe the contribution of the paper
The paper introduces the RDGM, a novel self-supervised model for fetal brain MRI that utilizes a Consistent Implicit Neural Representation (CINR) network to tackle regional intensity heterogeneity and a Global Diffusion Discriminative Generation (GDDG) module to ensure global consistency. It achieves state-of-the-art results across three real-world fetal MRI datasets.
- 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.
- This paper introduces a novel, self-supervised, coarse-to-fine reconstruction strategy that effectively handles the complexities of fetal MRI.
- A coarse SVR 3D training strategy, named CINR, involves proposing a 3D voxel batch association map from two distinct voxel mapping spaces.
- A Global Diffusion Discriminative Generation (GDDG) mechanism is proposed for generating fine high-fidelity fetal brain MRI 3D volume.
- 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 author did not conduct a thorough sensitivity analysis of hyperparameters, such as \lambda, \lambda_V， and \lambda_B.
- The author did not provide a detailed description of the experimental setup, which may make it difficult for readers to replicate the method. For example, the author used a diffusion model but did not specify the number of sampling steps, noise scale, etc.
- The author did not provide the number of parameters and the specific inference time of the proposed model.
- 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 submission does not mention open access to source code or data but provides a clear and detailed description of the algorithm to ensure reproducibility.
- 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
- Conduct a sensitivity analysis concerning the hyperparameters \lambda，\lambda_V，\lambda_B.
- The author should provide a detailed description of the experimental setup, such as the number of sampling steps and noise scale in the diffusion model, as well as the number of neurons in the hidden layers of the MLP.
- Provide the number of parameters and the specific inference time for both the proposed model and the comparative methods.
- 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?
N/A
- 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
N/A
- [Post rebuttal] Please justify your decision
N/A
Review #3
- Please describe the contribution of the paper
They propose a new framework for detecting fetal cerebellar landmarks based on 3D MRI data, utilizing publicly available segmentation data to guide the landmark detection task through pre-training.
They designed a pre-training method based on boundary distance maps, which enhances the landmark detection network’s ability to identify landmark positions.
Extensive experiments conducted on datasets from diverse backgrounds and different domains demonstrate the high effectiveness of our method, surpassing the accuracy of primary human experts.
- 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 contribution of the paper is centered around addressing significant challenges in fetal MRI, specifically motion correction, artifact removal, regional intensity heterogeneity, and global consistency in 3D space. Here are the key contributions of the paper:
Coarse-to-Fine Self-Supervised Framework (RDGM): The paper introduces the Radiation Diffusion Generation Model (RDGM), a novel coarse-to-fine self-supervised framework tailored for fetal brain MRI. This framework is designed to handle the complexities of fetal MRI data, which is often marred by motion artifacts and varying intensity distributions.
Regionally Consistent Implicit Neural Representation (CINR): A significant contribution is the development of the CINR network. This self-supervised model utilizes a novel double-spatial voxel association consistency mechanism that effectively addresses the problem of regional intensity heterogeneity in fetal MRIs. By enhancing 3D voxel association and complementarity through two-voxel mapping spaces, CINR generates a coarse representation of the MRI, setting a solid foundation for further refinement.
Enhanced Training Convergence: The paper proposes modifications to the training process, including fine-tuning the weighted slice reconstruction loss and introducing new voxel regularization techniques. These modifications are aimed at improving the training convergence speed, making the model more efficient and effective in learning from complex MRI data.
Global Diffusion Discriminative Generation (GDDG): As a part of the fine module of the RDGM, the GDDG approach is introduced to enhance the global consistency and discrimination of the volumetric data. This module uses noise diffusion processes to transform and refine the coarse MRI data into a high-quality output that maintains global structural consistency.
Comprehensive Validation: Through extensive experiments on diverse datasets, the paper validates the effectiveness of the proposed methods. The RDGM model not only addresses the inherent challenges in fetal MRI analysis but also surpasses the accuracy of primary human experts in detecting and correcting inconsistencies.
- 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 main weaknesses of the paper, as inferred from the provided context, could include the following aspects:
Lack of Hyperparameter Analysis: The paper does not provide a detailed analysis of the hyperparameters associated with the loss function. Understanding the impact of different hyperparameter settings is crucial for optimizing model performance and ensuring robustness across diverse datasets.
Limited Explanation of the Double-Spatial Voxel Association Mechanism: While the paper introduces a novel double-spatial voxel association consistency mechanism, there might be insufficient details on how this association is implemented and its specific advantages over traditional methods. This lack of detailed explanation can hinder the reproducibility and understanding of the model’s core features.
- 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 submission does not mention open access to source code or data but provides a clear and detailed description of the algorithm to ensure reproducibility.
- Do you have any additional comments regarding the paper’s reproducibility?
Please provide a link to your code repository, or alternatively, package your code into a Python package (similar to nesvor). I would like to manually test your code.
- 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
I am giving this paper a weak accept. The technological foundation of the paper is sound, and the improvements it introduces are clear. However, I would consider giving it a strong accept and recommend it for an oral presentation if I had the opportunity to test your code and verify the claimed improvements firsthand.
- 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?
I am giving this paper a weak accept. The technological foundation of the paper is sound, and the improvements it introduces are clear. However, I would consider giving it a strong accept and recommend it for an oral presentation if I had the opportunity to test your code and verify the claimed improvements firsthand.
- 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
i hold the same score
Author Feedback
N/A
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 reviewers are concerned about the lack of hyperparameter analysis, ablation study and further information on the methodological details. No rebuttal is provided. However: The scores are in majority in the “accept range”, the paper provides a new direction and angle for fetal brain reconstruction and the reviewers agree on a solid method and new direction - all great ingredients for a paper attracting interest at MICCAI. I therefore would despite the missing rebuttal recommend acceptance.
- 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 reviewers are concerned about the lack of hyperparameter analysis, ablation study and further information on the methodological details. No rebuttal is provided. However: The scores are in majority in the “accept range”, the paper provides a new direction and angle for fetal brain reconstruction and the reviewers agree on a solid method and new direction - all great ingredients for a paper attracting interest at MICCAI. I therefore would despite the missing rebuttal recommend acceptance.
Meta-review #2
- 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’
Reviewers are unclear about the motivation for combining diffusion models and Implicit Neural Representations (INR). The rationale behind integrating these methods is not well explained, making it seem like a mix of popular techniques without a clear purpose. There is insufficient detail on certain implementation aspects, such as the sampled noise term in equations 10 and 11, and the number of sampling steps and noise scale in the diffusion model. This lack of detail hampers reproducibility. The paper lacks adequate reconstructed figures to support the quantitative metrics, especially for ablation experiments. Reviewers also express concerns about the small variance reported in comparison tables, suggesting potential issues with the reported results. The paper does not provide a thorough sensitivity analysis of the hyperparameters, which is crucial for understanding the robustness and performance optimization of the proposed model. The paper does not mention open access to source code or data, which limits the ability of others to replicate the study’s findings. Code should be published with such a 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).
Reviewers are unclear about the motivation for combining diffusion models and Implicit Neural Representations (INR). The rationale behind integrating these methods is not well explained, making it seem like a mix of popular techniques without a clear purpose. There is insufficient detail on certain implementation aspects, such as the sampled noise term in equations 10 and 11, and the number of sampling steps and noise scale in the diffusion model. This lack of detail hampers reproducibility. The paper lacks adequate reconstructed figures to support the quantitative metrics, especially for ablation experiments. Reviewers also express concerns about the small variance reported in comparison tables, suggesting potential issues with the reported results. The paper does not provide a thorough sensitivity analysis of the hyperparameters, which is crucial for understanding the robustness and performance optimization of the proposed model. The paper does not mention open access to source code or data, which limits the ability of others to replicate the study’s findings. Code should be published with such a paper.
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’
The majority of the reviewers voted for accept. The proposed approach is interesting and can have an impact on the community beyond the specific application.
- 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 the reviewers voted for accept. The proposed approach is interesting and can have an impact on the community beyond the specific application.