List of Papers Browse by Subject Areas Author List
Abstract
Diffusion MRI (dMRI) is an important neuroimaging technique with high acquisition costs. Deep learning approaches have been used to enhance dMRI and predict diffusion biomarkers through undersampled dMRI. To generate more comprehensive raw dMRI, generative adversarial network based methods are proposed to include b-values and b-vectors as conditions, but they are limited by unstable training and less desirable diversity. The emerging diffusion model (DM) promises to improve generative performance. However, it remains challenging to include essential information in conditioning DM for more relevant generation, i.e., the physical principles of dMRI and white matter tract structures. In this study, we propose a physics-guided diffusion model to generate high-quality dMRI. Our model introduces the physical principles of dMRI in the noise evolution in the diffusion process and introduce a query-based conditional mapping within the difussion model. In addition, to enhance the anatomical fine detials of the generation, we introduce the XTRACT atlas as prior of white matter tracts by adopting an adapter technique. Our experiment results show that our method outperforms other state-of-the-art methods and has the potential to advance dMRI enhancement.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2024/paper/2374_paper.pdf
SharedIt Link: https://rdcu.be/dV1Oq
SpringerLink (DOI): https://doi.org/10.1007/978-3-031-72069-7_33
Supplementary Material: N/A
Link to the Code Repository
https://github.com/Caewinix/Phy-Diff
Link to the Dataset(s)
https://db.humanconnectome.org/
BibTex
@InProceedings{Zha_PhyDiff_MICCAI2024,
author = { Zhang, Juanhua and Yan, Ruodan and Perelli, Alessandro and Chen, Xi and Li, Chao},
title = { { Phy-Diff: Physics-guided Hourglass Diffusion Model for Diffusion MRI Synthesis } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2024},
year = {2024},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15002},
month = {October},
page = {345 -- 355}
}
Reviews
Review #1
- Please describe the contribution of the paper
The paper presents Phy-Diff, a novel model for synthesizing high-quality diffusion MRI (dMRI) images, and the key contribution is underlying its physics-guided: It introduces a physics-guided noise evolution process and a query-based conditional mapping to generate dMRI images.
- 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 paper introduces three novel approaches: 1) Integrates dMRI physics into noise evolution 2) Incorporates q-space sampling 3) Uses XTRACT atlas to enhance anatomical details
- 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.
1) There is a lack of description regarding the settings of the comparative experiments. 2) More detailed ablation experiments are needed, such as retaining only the results of individual modules.
- 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?
Due to the complexity of the article with numerous modules, including a demo for demonstration purposes would greatly enhance its accessibility.
- 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) Although the derivation of Equation 3 has been described, it is still not clear why it is defined in this form. Additionally, there is a lack of explanation for the definition of ¯αt, such as why it is exp(-2bnDn). 2) In the author’s description, “conventional positional embedding is only applicable to integer scalar index categories,” but it seems that positional encoding can still operate on non-integer values. Is there an issue with this statement? 3) It seems that the authors are working with dMRI images containing real numbers. Therefore, referring to them as a “real-number embedding function” in Section 2.2 is not clear. Furthermore, there is a lack of specific description for the settings of the MLP. 4) It is not clear what “Arbitrary bn” refers to in Table 1. Does it mean that images with any b-value can be generated? Or is there any constraint on bn during training? 5) There are similar questions regarding “any b-value” in the “Quantitative Results” 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 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 innovativeness, workload, and expression of the article contribute to my assessment score.
- 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 #2
- Please describe the contribution of the paper
Authros propose a physics-guided conditional diffusion model to generate high-quality dMRI. The method demonstrates robust performance when having inputs with different b-vectors, b-values, and slices.
- 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 paper is well-oragnized
- The proposed method is interesting and novel. It offers a promising new apporach for dMRI synthesis
- The comparison experiments are solid
- 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.
Only one test dataset with nine subjects is used
- Please rate the clarity and organization of this paper
Excellent
- 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
The method is novel and promising. I am looking forward to seeing it applied to more datasets with diverse demographics. In this paper, only one dataset with nine subjects is used. It is good for demonstrating it in a conference paper, hope authors can apply this method to more datasets in the future work.
- 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?
The paper is well-oragnized. The method is novel and compared experiments are relatively solid.
- 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 #3
- Please describe the contribution of the paper
This article proposes a physics-guided diffusion model to generate high-quality dMRI, aiming to address the lack of consideration for physical principles such as dMRI and white matter tract structure in existing methods for synthetic diffusion MRI.
- 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 article proposes a physics-guided diffusion model to generate high-quality dMRI data, which demonstrates novelty in both methodology and application. Specifically, it integrates the physical principles of dMRI into the model construction and incorporates prior information about white matter tracts, resulting in generated images that better match real dMRI data.
- 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.
1、In the experiment, this article utilized data from only nine subjects, which is relatively limited in quantity. Additionally, the data was divided into the usual proportions of 7:1:2. 2、I hope the authors clarify the experimental setup, especially regarding the training and testing processes. 3、Table 1 is not specifically referenced in the main text. It would be helpful if the authors could provide a reference. Additionally, it would be beneficial if the authors could provide specific values for arbitrary bn. 4、I hope the authors can correct the punctuation in section 3.2. 5、I hope the authors can provide further explanation for the statement, “However, PSNR is slightly inferior, …. high PSNR excessively smoothing the image.” 6、The authors stated in the qualitative results section that their method obtained the minimum error map. However, based on the error maps in Figure 2, it appears that their method’s error map is not the smallest. Perhaps providing a color bar would be helpful.
- 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 provide sufficient information for reproducibility.
- Do you have any additional comments regarding the paper’s reproducibility?
I hope the authors can open-source their code and provide a more detailed description of the training and testing processes.
- 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 hope the authors can provide relevant statistical analysis, such as t-tests, which might be more informative.
- 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?
To enhance the synthesis quality of dMRI, this paper integrates the physical principles of dMRI imaging along with corresponding prior information, which is crucial for clinical applications.
- 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
Author Feedback
We would thank all the reviewers for their constructive comments.
R1Q1/R3Q1: Test dataset. We trained our model based on 2D dMRI from the HCP S1200 dataset. Each subject contains 135 b-vectors with 110 slices at each b-vector. Therefore, we have included 1351109 slices altogether. Due to the long training time, we were only able to include nine subjects, in the same settings as the previous works (Ren et al., 2021). We will include more datasets in our future works.
R2Q1: Lack of description of the comparative experiments. We would thank the reviewer for this comment. Due to the page limit, we were unable to include these details. In the accepted version, we will provide a more detailed description.
R2Q2: More detailed ablation experiments. Our model contains three main modules, therefore we performed three ablation studies. We would thank the reviewer for the insightful comments and will consider adding more ablation experiments in the future.
R3Q2: Experimental setup. We would thank the reviewer for this comment. In the training phase, we set the maximum epochs of 80 and used the early stopping techniques to avoid overfitting. For the objective functions, we included these in section 2. In the testing phase, we tested a subject using b-values of 1000, 2000 and 3000 individually, with their combined set. We will add these details in section 3 in the accepted version.
R3Q3: Table 1 reference. Specific values for arbitrary bn. We will reference Table 1 in section 3.3, and provide more detailed clarification. Arbitrary bn means we used all b-values in {1000, 2000, 3000} when training and testing, instead of using specific b-values. The selection of these 3 values is due to the setting of the dataset, we plan to test more datasets with more b-values in the future.
R3Q5: Explanation for PSNR. PSNR calculates the pixel-squared error, and high values may blur important edges and texture details, while SSIM considers the structure of the image, similar to human vision. In our experiments, we observed that images with high PSNR and low SSIM appeared overly smooth. We inferred that if the SSIM significantly decreases after processing (e.g. using L2 optimisation, which is relevant to PSNR), it usually indicates that the structural information has been somewhat damaged. This is likely due to excessive smoothing, resulting in the loss of details and textures. In clinical diagnosis, using images losing important anatomical details is problematic. We will further explain this in the accepted version.
R3Q6: Error map. We would like to thank the reviewer for this comment. In the accepted version, we will further improve this figure and add a color bar.
Meta-Review
Meta-review not available, early accepted paper.