List of Papers Browse by Subject Areas Author List
Abstract
The diffusion MRI signals in the human cerebral cortex are strongly associated with neurodegenerative diseases. Although models like NODDI have been extensively used to characterize cortical microstructure degeneration, they fall short in capturing detailed, orientation-specific connectivity changes within the cortex. In this study, we introduce a method to decompose cortical tissue diffusion signal to radial and tangential components. Our approach uses data from multi-shell diffusion imaging and combines it with anatomical information from brain surfaces. By applying a GPU accelerated probabilistic optimization framework, we can accurately and efficiently estimate these diffusion components while keeping the results smooth and consistent with the cortical anatomy. We test our method on data from HCP subjects and a clinical dataset of patients with autosomal dominant Alzheimer’s Disease (ADAD) subjects. Our results demonstrate that the proposed method can more effectively reveal cortical gray matter connectivity changes related to tau pathology than metrics from the NODDI model. Our codebase is publicly available at https://github.com/Haibaobob/FOD-ctx-decomp.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/2438_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
https://github.com/Haibaobob/FOD-ctx-decomp
Link to the Dataset(s)
N/A
BibTex
@InProceedings{ZhaHon_GPU_MICCAI2025,
author = { Zhang, Hongbo and Nie, Xinyu and Yue, Jiaxin and Li, Yuan and Ringman, John and Shi, Yonggang},
title = { { GPU Accelerated Modeling of Cortical Radial and Tangential Connectivity Changes in Neurodegeneration } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
year = {2025},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15971},
month = {September},
page = {342 -- 351}
}
Reviews
Review #1
- Please describe the contribution of the paper
This paper describes a computational framework for decomposing multi-shell diffusion MR measurements in the cortex into radial and tangential components via a reconstruction of the cortical surface. It is based on free water aware fODF estimation and a Markov Random Field that classifies the resulting fODF peaks as radial or tangential, and is optimized on the GPU. Results are evaluated in terms of smoothness on HCP data, and in terms of effect sizes and association with tau pathology in a cohort of subjects with Alzheimer’s disease.
- 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 proposed approach is plausible and technically sound
-
Experimental results on clinical data seem usable for cortical microstructure analysis
-
The observed correlations with Tau-PET imaging are interesting
-
- 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 main contribution is in the implementation. Novelty in terms of ideas is limited, given that fODF estimation has been studied in great detail, that MRFs are widely used for brain image analysis, and that they have even been explored for fODF peak selection in [10].
-
Given the previous point, and the fact that the GPU-based implementation is not described in full detail (e.g., choice of regularization parameter, optimizer and its hyperparameters), it is somewhat disappointing that there do not appear to be any plans to release the code.
-
The presented evaluation did not convince me of the practical relevance of the proposed MRF-based assignment compared to a straightforward nearest neighbor based approach. Table 1 shows a small benefit in terms of smoothness, but I doubt whether that really makes a difference in the downstream analyses in Fig. 3/4/5.
-
Moreover, the main benefit compared to nearest neighbors appears to be handling cases in which the radial peak is absent. I wonder whether this might simply be addressed by imposing an angular threshold, i.e., assuming the radial peak is absent whenever the angular deviation towards the nearest neighbor becomes too large.
-
- 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 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
Why does Fig.3 show individual subjects rather than p values from a group comparison? Have those two subjects been cherry-picked?
How can we be sure that there is always at most one tangential peak direction?
The correct size of parentheses can be adjusted automatically in LaTeX via \left and \right
- 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?
Given that the methods used in this work are very classical ones, and that the evaluation does not make a strong point for the practical benefit over a straightforward solution of the same problem, I expect that the interest in this work at the MICCAI main conference will be limited. However, this work should be suitable for a more specialized workshop.
- 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.
The rebuttal addresses a key concern by promising that the code will be made available. I am still not fully convinced by the claimed practical significance and clinical value, given the lack of direct comparison to a straightforward nearest neighbor based approach, but having the code available will permit the community to try this out for themselves.
Review #2
- Please describe the contribution of the paper
This paper introduces a novel method of characterizing cortical diffusion from multi-shell diffusion MRI data with GPU accelerated optimization. The authors then apply this method to the HCP dataset and ADAD dataset.
- 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.
Overall, the method proposed is very interesting and addresses a gap in the current knowledge. Making the optimization GPU-amenable also increases the generalizability since the computation time is greatly reduced.
- 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.
There are two primary weaknesses in this work. First, there is not enough evidence provided to show that this method can capture the laminar connectivity in the cortex. The resolution inherently captures multiple layers since the cortex is 2.5-3mm on average and the resolution is 1.25mm^3. These layers are also dominated by radially oriented neurons. With these two things in mind, I am skeptical that we can effectively quantify tangentially-oriented intra-laminar interneurons. Second, I am not convinced by the metrics chosen to show to the validity of the method. I’m not sure smoothness should be an inherent feature of these metrics. Related to this, I would like to see some examples of how well the model fits the signal, especially compared to NODDI and other related methods.
There is no limitations section and no open-source code mentioned.
- 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.
- 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 proposed method is interesting and could have widespread implications. But I think more justification and evidence need to be provided showing the validity of the method and to show it is really measuring what it claims to be measuring (laminar connectivity).
- 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 #3
- Please describe the contribution of the paper
Neurons in the human cortex are connected via highly organized fiber bundles than run radially (i.e., into the white matter) and tangentially (i.e., within the cortex). This submission proposes an approach to disentangle radial and tangential fiber components based on diffusion-weighted MRI acquisitions. A modification of a previously reported model is used for cortical (grey matter) data, and embedded in a probabilistic relaxation framework to distinguish both directional components. The method is evaluated on data acquired in healthy subjects and patients suffering from Alzheimer’s disease.
- 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 topic of this submission is scientifically interesting and of potential interest to the audience of this conference. The text is mostly well written (except where noted below) and straightforward to understand for a reader with a background in recent approaches for neuroimaging data analysis.
- 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.
A few, rather minor issues should be considered:
- Consider adding some word of caution re: tangential bundles. Except for the visual cortex, those bundles (or, rather layers) are less than 100 um thick - thus, much less than the typical voxel resolution acquired from in-vivo brains.
- For the curious: Is GPU acceleration faster than thread-based parallelization at the voxel level?
- Table 1 (and text): Please, specify exactly to what “Average Time” refers to.
- Fig.4 (and text): Your statement requires a statistical argument to be convincing.
- Fig.5 (and text): Same here. Correlations close to zero may not be significant after all.
- 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.
- 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?
Best submission of the batch.
- 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
We appreciate the reviewers’ valuable comments and insights. Our responses address the key points raised as follows: R1:
- Contribution and novelty: While existing techniques such as FOD estimation and MRFs are indeed established, our approach uniquely integrates these methods to address cortical microstructure analysis—a challenging task with significant anatomical and clinical implications. Despite its importance, this specific problem has been explored minimally in existing literature.
- Implementation detail: We used lambda_p = 0.5 (HCP) / 0.1(ADAD), lambda_reg = 1e-6 for both dataset. We used an Adam optimizer (lr = 0.1 (learning rate), beta1 = 0.9, beta2=0.999). We will include the code and complete hyperparameter settings in the revised manuscript.
- Smoothness evaluation: Although improvements in mCS appear modest, we highlight substantial improvements achieved in mSIG, which is not explicitly optimized within our MRF framework. The smoothness evaluation clearly supports our assumption of spatial coherence. The notable gains in mSIG, combined with demonstrable enhancements in downstream clinical analyses, underscore the practical significance and clinical value of our proposed method.
- Thresholded NN: We concur that an angular threshold NN could potentially tackle the missing radial issue. However, angular threshold methods still significantly depend on voxel-wise normal directions, which can be unreliable outliers in regions of complex cortical foldings. Our approach inherently mitigates such outliers by effectively enforcing spatial coherence.
- Figure 3 subject selection: The two subjects depicted in Fig. 3 exemplify typical cortical microstructural patterns differentiating CN and ADAD subjects. These illustrative examples reflect patterns consistently observed across our studied cohorts, notably prevalent among ADAD subjects. Comprehensive group-wise analyses supporting this observation are presented in Fig. 4.
- Tangential peaks: Currently, the model selects the largest available tangential peak per voxel to facilitate representative analyses, acknowledging that capturing all tangential peaks cannot be guaranteed at this stage. Future work will extend our model to accurately identify and analyze multiple distinct tangential peaks within each voxel, enabling a more comprehensive and rigorous representation. R2:
- Thread parallelism: While voxel-wise parallelism is limited due to MRF’s pairwise dependencies, we could do thread parallelism via matrix multiplications, which by experience, will slow down the runtime by ~30x compared to GPU acceleration on our device. This is why we incorporate GPU acceleration.
- Specification on time: “Average Time” pertains solely to the peak extraction step—radial peaks for all compared methods, and tangential peaks additionally for our method—conducted on one hemisphere. Pre- and post-processing steps (e.g. FOD integration), are excluded. 4&5. Statistical test: We appreciate the recommendation and will include corresponding statistical test results alongside visual plots in the revised manuscript. R8:
- Imaging resolution: We acknowledge that the voxel size of 1.25 mm limits laminar specificity. However, our model targets averaged signals in radial and tangential components rather than precise laminar structures. As demonstrated in Fig. 3, these averaged signals remain sufficient to capture cortical neurodegeneration. Future work will involve evaluating our method at higher imaging resolutions.
- Smoothness evaluation: The assumption of spatial smoothness is widely adopted in neuroimaging algorithms. Given the lack of ground truth, our smoothness evaluation is intended to demonstrate that our algorithm effectively captures the expected inherent spatial coherence.
- Model Fitting: The FOD model accurately fits the multishell dMRI signal, yielding results consistent with those in [8]. Future work will include a comprehensive comparison of different model fitting.
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’
Despite some limitations, all reviewers finally agree on acceptance.
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’
This paper addresses an important topic in FOD estimation for gray matter (GM). It is well-written, with methods presented clearly and effectively. The proposed approach was evaluated on both high-quality public datasets (HCP) and clinical data (ADAD), demonstrating its efficiency and high sensitivity in detecting Alzheimer’s disease.