Abstract

Tissue microstructure information reconstructed from diffusion magnetic resonance imaging (MRI) provides crucial brain tissue information for brain disease analysis. However, clinical imaging time constraints often limit the availability of diffusion MRI, thus prompting research into tissue microstructure reconstruction from clinically feasible MRI modalities, such as T1-weighted MRI. Recent Transformer-based generative adversarial networks demonstrate potential by capturing long-range dependencies via self-attention in general MRI synthesis tasks, yet the significant gap between diffusion and T1-weighted MRI limits their ability to achieve optimal performance, leading to anatomical inconsistency in the reconstructed tissue microstructure maps. To address the problem, we propose a hierarchical anatomy-aware guidance (HAAG) framework for brain tissue microstructure reconstruction from T1-weighted MRI. First, we consider a two-level strategy to introduce the anatomical priors for the Transformer. At the input level of the Transformer, we propose an adaptive semantic embedding module that seamlessly integrates anatomical structure category information, providing semantic-level guidance for tissue microstructure reconstruction. At the feature modeling level of the Transformer, we propose a distance-guided self-attention mechanism to achieve effective information fusion of anatomical structures that balances both global and local contexts. Then, we consider a more general approach to verify the anatomical consistency at the output level of the whole synthesis network. We develop an anatomy-aware discriminative loss that encourages anatomical consistency between the input and output modalities. HAAG was validated on a public brain MRI dataset for reconstruction of tissue microstructure from T1-weighted MRI. The results demonstrate that our method significantly improves the quality of tissue microstructure reconstruction.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/2966_paper.pdf

SharedIt Link: Not yet available

SpringerLink (DOI): Not yet available

Supplementary Material: Not Submitted

Link to the Code Repository

https://github.com/Winterborner/HAAG

Link to the Dataset(s)

HCP dataset: https://db.humanconnectome.org/data/projects/HCP_1200

BibTex

@InProceedings{LiYux_Hierarchical_MICCAI2025,
        author = { Li, Yuxing and Ye, Chuyang},
        title = { { Hierarchical Anatomy-Aware Guidance for Brain Tissue Microstructure Reconstruction from T1-weighted MRI } },
        booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
        year = {2025},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 15962},
        month = {September},
        page = {251 -- 261}
}


Reviews

Review #1

  • Please describe the contribution of the paper

    This paper proposes a hierarchical anatomy-aware guidance (HAAG) framework for brain tissue microstructure reconstruction from T1-weighted MRI. The framework introduces anatomical priors into the Transformer model, providing semantic-level guidance through an adaptive semantic embedding module, and achieves effective fusion of anatomical structure information via a distance-guided self-attention mechanism. Additionally, an anatomy-aware discriminative loss is used to ensure anatomical consistency between the input and output modalities. Experimental results demonstrate that the HAAG framework significantly improves the quality of tissue microstructure 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.
    1. Integration of anatomical prior knowledge: The Hierarchical Anatomy-Aware Guidance (HAAG) framework incorporates anatomical structure information at input, feature modeling, and output levels to improve anatomical consistency.

    2. Moderate computational efficiency: Transformer-based GANs are practical in resource-limited clinical settings.

    3. Improved synthesis performance: Outperforms existing methods in reconstructing NODDI microstructure maps (e.g., ICVF, OD) on public datasets.

  • 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.
    1. The paper does not provide a detailed explanation of the data preprocessing process.

    2. The method presented in the paper is proposed to address the lengthy clinical imaging time in diffusion MRI; however, the experimental section of the paper does not demonstrate the time cost of this method.

    3. The authors are encouraged to discuss the limitations and other implications of its method, or directions for future research.

    4. It is unclear whether the data used in this method comes from different patient groups, which would help in understanding the generalizability of the method.

  • 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.

  • 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.

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

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

    The method proposed in this paper is interesting; however, the article has deficiencies in its organization, which impacts its readability and reproducibility.

  • Reviewer confidence

    Confident but not absolutely certain (3)

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

    Reject

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

    After carefully cheking the rebutall and other reviewers’ comments, I still have major concerns regarding the evaluation of the method and its real clinical application. Therefore, my find decision is rejection.



Review #2

  • Please describe the contribution of the paper

    This paper presents HAAG (Hierarchical Anatomy-Aware Guidance), a framework designed to improve brain tissue microstructure reconstruction—specifically, NODDI-derived ICVF and OD maps—using clinically available T1-weighted MRI, without the need for time-consuming diffusion MRI scans.

  • 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.
    1. Use a hierarchical anatomy-aware prior for synthesis.
    2. The topic of brain tissue microstructure reconstruction is interesting.
    3. The model outperforms other approaches on the HCP dataset across multiple metrics.
  • 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.
    1. While the paper makes a strong case for anatomical prior integration, it does not sufficiently review earlier works that similarly integrate anatomy or hierarchical priors into medical image synthesis. Some examples are listed below. The authors should review and discuss these related studies to better position their contribution. [1] Structure-preserving image translation for multi-source medical image domain adaptation[J]. Pattern Recognition, 2023, 144: 109840. [2] HiFi-Syn: Hierarchical granularity discrimination for high-fidelity synthesis of MR images with structure preservation[J]. Medical Image Analysis, 2025, 100: 103390.

    2. There is a lack of baseline comparison with more recent models, particularly those involving structurally-aware or hierarchical modeling approaches.
    3. The setting of the ablation study is confusing, and there is no clear baseline backbone (e.g., the model without any of the proposed modules). Does the row marked with “ASE” represent the “Ours1” model without the ASE module? The authors should clarify this and reorganize the table accordingly.
    4. Quantitative MRI synthesis is a rigorous task, as its numerical outputs are often closely linked to clinical conclusions or pathological assessments. The paper lacks clinically relevant evaluation strategies, such as expert ratings, which limits the credibility of the method in real-world medical applications.
  • 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

    Please refer to the comments regarding the weaknesses.

  • 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.

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

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

    The proposed framework for brain tissue microstructure reconstruction holds clinical value; however, the lack of review, comparison, and citation of related works with similar ideas in this field weakens the completeness of the paper and undermines the clarity of its contribution.

  • Reviewer confidence

    Confident but not absolutely certain (3)

  • [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.

    Thank you for the detailed rebuttal. The authors have addressed most of my concerns satisfactorily. I appreciate the clinical-driven contributions, which are well-motivated and clearly articulated. The discussion further supports the clinical relevance and coherence of the proposed framework, demonstrating its potential value in real-world applications.



Review #3

  • Please describe the contribution of the paper

    The authors propose framework with two hierarchical anatomy-aware guidance for brain tissue microstructure reconstruction from clinically feasible MRI. They adapted SOTA transformer-based architecture and provided adequate experiments.

  • 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 is well-organised and presented method is clearly described, followed by good evaluation and experimental part including publicly available dataset. Also the authors implementation is based on publicly available implementation of ResViT and the authors are going to publish their code. I don’t find the method overcomplicated, which I consider an advantage.

  • 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.

    Check the typos and formatting of text. Otherwise, I have no complaints.

  • 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.

  • 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

    No.

  • 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?

    Mentioned above.

  • 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




Author Feedback

Reviewer #1
Q1​​: Lack of data preprocessing details. ​​A1​​: Preprocessing includes: (1) computing tissue microstructure maps from diffusion MRI, and (2) co-registering diffusion MRI/microstructure maps with structural MRI. Details will be added in revision. ​​Q​​2: Method’s time reduction claim lacks experimental validation. ​​A​​2: Our method reduces clinical time by replacing lengthy diffusion MRI (≈30min for 270 gradients) with structural MRI (≈2min). High-quality results are achieved with ​​1/15 data collection time​​ (implied time saving≈28min). Q​​3: Need limitations/future work discussion. ​​A​​3: Future work focuses on validating clinical utility on patient cohorts. We will supplement limitation discussion (e.g., generalizability to pathological cases) in revision. Q​​4: Data source lacks diverse patient groups. ​​A​​4: As prior works [1, 2] have indicated transferability of deep learning diffusion MRI analysis to patients, and our work first reconstructs tissue microstructure from structural MRI, we have focused on the basic evaluation on healthy subjects. Also, the ​​anatomy​​ shared between healthy/diseased brains suggests method applicability on patients. We will test clinical cohorts in future work. Reviewer #2
Q5​​: Typos/formatting need checking. ​​A​​5: All typos and formatting inconsistencies have been corrected in revision. Reviewer #3 Q​​6: Insufficient discussion of anatomical prior integration in related works. ​​A​​6: A discussion will be added to clarify that cited works (e.g., mutual information loss in [3], hierarchical granularity in [4]) focus on ​​structural preservation in image synthesis​​, whereas our innovation is the ​​first to leverage structural MRI for tissue microstructure reconstruction​​ with explicit anatomical priors. This distinction highlights our clinical-driven contribution. ​​Q​​7: Insufficient comparison with recent structurally-aware models. A​​7: Our baseline selection prioritizes ​​brain MRI synthesis methods​​ due to the ​​ill-posed nature​​ of microstructure reconstruction. The omitted study [3] focuses on non-brain domains (e.g., fundus/prostate), whose anatomical complexity differs fundamentally from brain tissue mapping. Concurrent work [4] (released during our submission preparation) will be benchmarked in an extended analysis. Q​​8: Unclear ablation study table labeling. ​​A​​8: The table labeling confusion arises from inverted module exclusions: the “ASE” row represents “Ours1” ​​without RDG-SA​​ (not ASE), and “RDG-SA” indicates ​​without ASE​​. We will revise the table with unambiguous notations (e.g., “w/o ASE”) and explicitly reference the baseline backbone from Table 1 to eliminate ambiguity. ​​Q​​9: Lacks clinical evaluation (e.g., expert ratings). ​​A​​9: For healthy subjects’ tissue microstructure maps, expert visual ratings are infeasible due to ​​sub-voxel structural homogeneity​​. Current validation relies on ​​gold-standard metrics​​ computed with diffusion MRI. As 1) recent work [1] has already indicated that tissue microstructure estimated by deep learning (from undersampled diffusion MRI) has the capability of detecting disease-related brain tissue alterations​​, and 2) our work is the first to reconstruct tissue microstructure maps from structural MRI, the validation focused on the primary evaluation of reconstruction consistency with gold standard instead of clinical evaluation. References [1] Li et al., Deep learning enables accurate brain tissue microstructure analysis based on clinically feasible diffusion magnetic resonance imaging, NeuroImage 2024. [2] Zhang et al., DDParcel: deep learning anatomical brain parcellation from diffusion MRI, IEEE TMI 2023. [3] Kang et al., Structure-preserving image translation for multi-source medical image domain adaptation, Pattern Recognition 2023. [4] Yu et al., HiFi-Syn: Hierarchical granularity discrimination for high-fidelity synthesis of MR images with structure preservation, MedIA 2025.




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’

    N/A



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’

    The manuscript was rejected due to the author’s inadequate response to key concerns. It fails to provide a strong alternative or clarify its primary objective, leaving important conceptual issues unresolved. It lacks details on data preprocessing, fails to demonstrate the method’s time cost, does not discuss limitations and future research, and does not clarify the patient groups involved. These concerns hinder its evaluation and clinical applicability.



back to top