List of Papers Browse by Subject Areas Author List
Abstract
Hyperpolarized 129Xe lung magnetic resonance imaging (MRI) offers a method for visualizing human lung function. However, its long imaging time limits widespread research and clinical adoption. Deep learning has shown significant potential in addressing undersampled MRI reconstruction challenges. Yet, the clinical novelty of hyperpolarized 129Xe lung MRI results in a particular lack of raw k-space data. To address this, we propose a Noise-Controllable Complex-Valued Diffusion Model (NC-CDM) to augment the available data from limited clinical training set. Specifically, complex-valued convolutional kernels replace traditional ones, enhancing feature extraction and data utilization efficiency by learning rich representations from k-space. In addition, a noise-controllable module is introduced to mitigate estimation biases caused by thermal noise during MRI acquisition in the training phases. Experiments compare the proposed NC-CDM with other state-of-the-art models. Fréchet Inception Distance (FID) and Inception Score (IS) metrics show that our method obtains higher image quality. The generated data, mixed with real data, are subsequently applied to downstream MRI reconstruction task using two deep learning-based MRI reconstruction methods: CasNet and KIKI-net. The results show that reconstruction networks trained on our generated data exhibit superior reconstruction performance.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/2171_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
https://github.com/TmpAccount25/NC_CDM
Link to the Dataset(s)
N/A
BibTex
@InProceedings{HanLin_NoiseControllable_MICCAI2025,
author = { Han, Linxuan and Xiao, Sa and Li, Muhong and Liu, Jinghan and Zhou, Xin},
title = { { Noise-Controllable Complex-Valued Diffusion Model for k-Space Data of Hyperpolarized 129Xe Lung MRI Generation } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
year = {2025},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15969},
month = {September},
page = {359 -- 368}
}
Reviews
Review #1
- Please describe the contribution of the paper
This study proposes a noise-controllable, complex-valued diffusion model for generating k-space data in hyperpolarized 129Xe MR imaging. Experimental results show that incorporating the generated k-space data with real data enhances the performance of two MR image reconstruction networks.
- 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 study presents an effective strategy for generating k-space data using a complex-valued diffusion model, which leverages complex-valued operations to more accurately represent the intrinsic properties of k-space. Further more, considering the limitations of traditional generative quality metrics such as the FID and IS scores, the authors evaluated the quality of the generated data by training MR reconstruction networks with a mixture of real and synthetic k-space data. They compared the reconstruction performance using different generative models.
- 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 implementation details are insufficient, particularly for the comparison methods.
In conventional practice, complex MRI data are typically separated into real and imaginary components and processed independently. A comparison between the proposed approach and this common implementation should be conducted.
The performance of the proposed method should also be evaluated on out-of-distribution data, such as images acquired from patients.
- 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 has provided an anonymized link to the source code, dataset, or any other dependencies.
- 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 am wondering whether the generated data are multi-coil or single-coil k-space data.
Characterizing k-space is inherently challenging due to its complex distribution. An alternative approach is to generate complex-valued images and then transform them back into k-space, which may be more straightforward. I suggest comparing these two strategies.
- 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?
Generating high-quality k-space data is crucial for deep learning-based MR reconstructions, particularly for applications where raw k-space data are unavailable. The proposed method directly characterizes the k-space data distribution using diffusion models and enhances performance by employing a complex-valued network. Additional details should be provided to help readers better understand the strengths of the method.
- 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
Review #2
- Please describe the contribution of the paper
This work use diffusion model to synthesize the k-space data for hyperpolarized 129Xe lung MRI. It uses the complex-valued convolutions and include a nose-controllable module to mitigate the thermal noise. The work also validate the proposed methods by integrating the synthetic data and real data for image reconstruction.
- 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.
- This work applied diffusion model for hyperpolarized 129Xe lung MRI, which is a novel application.
- It also show that the proposed methods can significantly improve performance compare to other SOTAs.
- The ablation studies are able to support their hypothesis.
- 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 noise-controllable module is not clearly introduced. No explaination what are the alphas and why thermal noise will increase as t decrease. And where lamba_t is a hyperparameter or a trainable parameter.
- 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 has provided an anonymized link to the source code, dataset, or any other dependencies.
- 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?
Please provide a clear explanation of the noise control module to understand the novelty of this work.
- 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
This study proposed a novel k-space data generation method that can potentially alleviate the problem of lacking sufficient clinical data. A detailed comparison of various methods and evaluation on a downstream task demonstrated have been carried out.
- 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.
-
A novel way to generate synthetic k-space data, which holds potential to outperform conventional methods that only generate image-domain data.
-
Good evaluation - In addition to comparing different approaches for data generation, the authors also do evaluation using the downstream task.
-
- 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.
- It’s not entirely clear from the paper that whether the generated k-space is single-coil or multi-coil. Based on the provided code, it seems like only single-coil k-space data were synthesized and evaluated. Nowadays, most clinical MRI scans are carried out with multiple coil elements. More developments are required to assess the actual clinical applicability of the proposed method.
- 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 has provided an anonymized link to the source code, dataset, or any other dependencies.
- 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?
Although it may not be readily applicable to actual MRI data because the current method only generates single-coil k-space data, the paper still presents a new approach in k-space data generation. The authors also took efforts to do detailed evaluations to demonstrate the performance of the proposed method.
- Reviewer confidence
Somewhat confident (2)
- [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
Thank you to the editor in chief and all the reviewers for the recognition of this work. Q1: The implementation details are insufficient, particularly for the comparison methods. (R1) A1: We will include a detailed description of the comparative methods in the experiments section.
Q2: In conventional practice, complex MRI data are typically separated into real and imaginary components and processed independently. A comparison between the proposed approach and this common implementation should be conducted. (R1) A2: The method of separating complex-value k-space data into real and imaginary components for independent processing treats complex MRI as a dual channel real-value MRI and is not extract complex spatial features using complex-value convolution and proposed methods.
Q3: The performance of the proposed method should also be evaluated on out-of-distribution data, such as images acquired from patients. (R1) A3: Our dataset includes data from healthy subjects and patients. In the latest version of the article, we will include this section in the dataset description.
Q4: I am wondering whether the generated data are multi-coil or single-coil k-space data. (R1) A4: The proposed method is not limited to the acquisition method of MRI and is applicable to both multi-coil and single-coil MRI acquisition. However, the hyperpolarized 129Xe lung MRI data we used were collected using a single-coil. For specific collection details and experimental parameters, please refer to our article(Sci. Adv., doi: 10.1126/sciadv.abc8180).
Q5: Characterizing k-space is inherently challenging due to its complex distribution. An alternative approach is to generate complex-valued images and then transform them back into k-space, which may be more straightforward. I suggest comparing these two strategies. (R1) A5: Due to the reversible nature of Fourier transform and inverse Fourier transform, generating complex valued images and generating k-space data are equivalent. However, the real collected k-space data contains phase information, Which is not be preserved through Fourier and inverse transformations. So, directly generating complex value images is not align with the background.
Q6: It’s not entirely clear from the paper that whether the generated k-space is single-coil or multi-coil. Based on the provided code, it seems like only single-coil k-space data were synthesized and evaluated. Nowadays, most clinical MRI scans are carried out with multiple coil elements. More developments are required to assess the actual clinical applicability of the proposed method. (R2) R6: The proposed method is not limited to the acquisition method of MRI and is applicable to both multi-coil and single-coil MRI acquisition. However, the hyperpolarized 129Xe lung MRI data we used were collected using a single-coil. For specific collection details and experimental parameters, please refer to our article(Sci. Adv., doi: 10.1126/sciadv.abc8180). We are plan to make multi-coils for data acquisition. Subsequent work will expand to the generation of multi-coil k-space data, as well as the generation of more imaging sequences and target organs.
Q7: The noise-controllable module is not clearly introduced. No explaination what are the alphas and why thermal noise will increase as t decrease. And where lamba_t is a hyperparameter or a trainable parameter. (R3) R7: (1) Alpha is a parameter in the vanilla diffusion model. (2) MRI thermal noise is fixed for specific images and does not vary with t. The difference \sqrt{\bar{\alpha}_{t}} \eta^{\prime} is increase as t decreases.(3) Lambda_t is a non learnable parameter, and its calculation method is stated in formula (3) of the paper. To improve clarity,we will provide additional explanations for these formulas and symbols in the revised version.
Meta-Review
Meta-review #1
- Your recommendation
Provisional Accept
- 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”.
This paper introduces an innovative approach for hyperpolarized 129Xe lung MRI that addresses the common challenge of lack of raw k-space data. The method leverages complex-valued convolutional layers for more effective feature extraction and incorporates a noise-controllable module to reduce the impact of thermal noise, both of which are highly relevant advancements for this imaging modality. Strengths: The use of complex-valued convolution is well-motivated and enhances the model’s ability to process MRI data efficiently. The inclusion of a noise-controllable module directly targets a significant source of image degradation in hyperpolarized gas MRI. The proposed framework demonstrates promising results and could have a meaningful impact on the clinical utility of 129Xe lung MRI. Weaknesses: Some aspects, such as the workings of the noise-controllable module and whether the method is tailored for single-coil or multi-coil data, could benefit from further clarification. Quantitative results are presented only as averages, without reporting variance or statistical significance, which would strengthen the evaluation.
Overall, the paper presents a novel and practical solution to a pressing problem in lung MRI. The minor weaknesses identified by reviewers are primarily related to clarity and reporting and can be readily addressed in the camera-ready version.