List of Papers Browse by Subject Areas Author List
Abstract
Selective serotonin reuptake inhibitors (SSRIs) are recognized as the first-line treatment for major depressive disorder; however, the characterization of their microstructural effects on white matter (WM) is still limited. This study presents a novel path signature (PS) framework to quantify longitudinal WM plasticity utilizing clinical-grade diffusion magnetic resonance imaging (MRI) data. This approach overcomes the limitations of conventional diffusion metrics, achieving a sensitivity of 1 mm³ without requiring high-resolution imaging. By combining rough path theory with super-resolution mapping, significant SSRI-induced reorganization is found in the transverse pontine tract, left anterior limb of the internal capsule, and splenium of the corpus callosum in MDD patients. Changes in PS features in these fiber bundles and the left corticospinal tract correlate positively with reductions in the 17-item Hamilton Depression Rating Scale scores, providing preliminary evidence of a relationship between WM alterations and clinical outcomes. The findings establish PS analysis as a promising tool for detecting macrostructural plasticity in WM due to SSRIs, thereby bridging the critical gap between microstructural diffusion metrics and circuit-level reorganization, and providing a novel insight into comprehensive biomarkers for precision antidepressant therapy.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/1330_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
https://github.com/qinjiaolong/PS
Link to the Dataset(s)
https://github.com/qinjiaolong/PS
BibTex
@InProceedings{QinJia_Path_MICCAI2025,
author = { Qin, Jiaolong and Dong, Weihong and Ni, Huangjing and Yao, Zhijian and Lu, Qing and Wu, Ye},
title = { { Path Signature Features Revealed SSRI-Induced White Matter Morphological Reorganization in Depressions } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
year = {2025},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15972},
month = {September},
page = {445 -- 455}
}
Reviews
Review #1
- Please describe the contribution of the paper
The paper proposes using path signature (PS) to characterize neural fiber pathways based on diffusion MRI data. It was then applied to study MDD patient data. The paper claims that the multiscale feature of the PS method may help to identify macrostructural markers at a finer resolution (1mm^3) based on raw data acquired at a coarse resolution level (1.875x1.875x3mm^3).
- 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 method uses path signature as the major feature to characterize fiber tracts, The PS has a unique hierarchical structure.
- 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.
- By concept, PS is not a new idea, which basically is similar to the hierarchical “vector feature” that has been used for decades in image understanding and patter recognition. However, the parametric feature of PS seems to be a 1-dimensional feature within the streamline, while the vector feature preserves spatial information in the 3D space.
- Many mathematical expressions are inadequately explained. Many items need further clarity. Section 2.1: what exactly are a & b? They seem to be the 2 parametric start and ending points. If so, please clearly state it. Equation 1: it defines a<s<t, so that the signature would always be positive. Why later there are cases with negative PS values? Is any information missing here? More detailed explanations are needed. Please clarify. Equation 3: what exactly is the physical meaning of the 3-rd order term? Please explain to facilitate understanding. This applies also to Equation 4 for the k-th order terms. Section 2.2, “…compute the PS features at each point…”: the sampling strategy concerning deciding each point or the sliding window is unclear. Equations 7&8: as mentioned, it is not clear why PS values could be negative when computing PS always starts from a to b. Same place: As a and b are parametric starting and ending points, the fiber tracts in 3d space may grow from the same voxel to all different spatial directions. Averaging PS within the same voxel does not seem to make much sense, except to retain an average running distance from the voxel. The PS computed within the 1-dimensional streamline seems unable to characterize the 3D spatial characteristics of fiber tracts, but only captures 1-dimensional distance information from the starting point a.
- PS seems only to work with the simplified diffusion tensor model, which assumes just one underlying direction of the neural fibers at each location. While in fact the situation is much more complex in the brain that many areas exist crossing, kissing, fending, bifurcation (merging, splitting) fibers, PS seems helpless in these cases.
- Some of the claims may be problematic. For example: The paper claims the method may work out results at a resolution of 1 mm^3. However, the procedure was based on raw data of a resolution of 1.875x1.875x3 mm^3. While super-sampling the data will not make up additional finder information, claiming resampling and processing at a small step would generate the results at a finer resolution at 1mm^3 is not deemed correct. Otherwise, anybody may resample and process the data and generate results at whatever levels of resolution they want.
- The analysis results based on MDD patient data thus obtained by using this method therefore does not seem convincing and valid.
- Please rate the clarity and organization of this paper
Poor
- 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.
(2) Reject — should be rejected, independent of rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
Method description seems problematic. Certain claims may be wrong. Please see details outlined in Item#7.
- Reviewer confidence
Very confident (4)
- [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.
The concerns are not satisfactorily addressed, for example, the issue of resolution.
Review #2
- Please describe the contribution of the paper
This study presents a path signature (PS) framework to quantify longitudinal WM plasticity utilizing clinical-grade diffusion magnetic resonance imaging (MRI) data.
- 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.
writing and method.
- 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.
- Figure 1 in section 2.2 is not fully described clearly. How to achieve equal length sampling?
- Section 2.2, rewrite the first S p1, p3.
- Although the PS is computed over the interval [p1,p3], the formula should mention p2.
- For section 3.2, authors should provide all 39 results like figure 3 as the supplement material.
- For section 3.3, authors should provide all 39 results like figure 4 as the supplement material.
- The authors did not mention the patient’s sex and age group information. I believe there should be some more comparative experiments are needed to further validate the effectiveness 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 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.
(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?
method
- 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 work proposes a novel path signature framework for characterizing the geometric properties of diffusion MRI fiber paths throughout the brain. The proposed PS method uses a sliding window to traverse all fiber paths in a tractogram, and estimates varying directional and geometric properties of the tract based on the window’s spatial position. These feature maps are then applied to a study of patients with MDD given SSRI treatments, and new potential insights into such treatments are shown.
- 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 method is, conceptually at least, fairly intuitive and understandable. Furthermore, it is readily applicable to any number of tractography-based studies.
-
The experimental results provide a preliminary model validation, as supported by the literature, and also point towards new directions for studying the effectiveness of SSRIs in treating MDD.
-
- 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.
-
While I like the direct application of a novel method to a clinically-oriented problem, the single experiment does not provide enough methodological validation on its own. The primary validation of the proposed method is that the statistically different tracts found with these PS features match those previously reported in the MDD literature. However, many of these previous works are based on simpler diffusion measurements such as FA, which prompts the question: what do the PS feature maps show us that previous metrics (diffusion or tract-based) cannot? It would be beneficial to show additional validation in a more simple experimental setting. Then, we can be more confident that PS features tell us things that other measurements cannot and vice-versa (e.g. what are the limitations of PS features).
-
Related to the previous comment, it is not clear to me how PS features compare to tract-based features. For example in the TWI paper [1] (among other works), local curvature is shown as a track-weighted feature map. What is shown by PS features that is not visible in averaged curvature, or even simple FA? If the MDD analysis is performed with these other tract or image-based features, are the same tracts still statistically different?
-
There are a couple of confounds in the MDD analysis that may be worth considering. To rule out inter-session variability as a confound, it would be informative to see the MDD analysis performed on a control group over a similar time period. Also, correction of magnetic susceptibility distortion was not mentioned in the preprocessing. Susceptibility distortion effects the resulting tractography [2], and it may be the case that PE features are particularly sensitive to geometric distortions. So it would be worth correcting, if possible.
[1] F. Calamante, “Track-weighted imaging methods: extracting information from a streamlines tractogram,” Magn Reson Mater Phy, vol. 30, no. 4, pp. 317–335, Aug. 2017, doi: 10.1007/s10334-017-0608-1.
[2] M. O. Irfanoglu, L. Walker, J. Sarlls, S. Marenco, and C. Pierpaoli, “Effects of image distortions originating from susceptibility variations and concomitant fields on diffusion MRI tractography results,” NeuroImage, vol. 61, no. 1, pp. 275–288, May 2012, doi: 10.1016/j.neuroimage.2012.02.054.
-
- 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
Specific comments:
-
The “PS_n_pos” notation used in the manuscript is difficult to read, especially in sections with multiple PS maps listed together such as Sections 3.2 and 3.3.
-
Figure 3 is difficult to understand. It is small and somewhat visually cluttered, and the caption does not explain what is being displayed. The different elements of the figure also show noticeable compression artifacts.
-
I would like to see a visual example of at least some of these PS feature maps to gain a better intuition for what features they are capturing. There are two examples in Figure 1, but I believe the figure only shows the same positive and negative feature map copied and pasted.
-
The use of the term “super-resolution” in the manuscript is somewhat confusing. The PS feature maps are calculated from the underlying tractograms, which can be sampled at any spatial resolution. However, the tractograms are still limited by the spatial resolution of the dMRIs, and (usually trilinear) interpolation is used to bridge the gaps. So, the accuracy of the PS maps is directly dependent upon the tractography accuracy, which can change based on resolution (and other parameters) [1-2]. I would not consider the underlying tractogram to be “super-resolved” in the typical sense, and I would apply the same conclusion to the PS maps.
-
Section 2.3, 2nd paragraph: I do not believe iFOD2 uses RK4 integration, as it is a probabilistic tractography method [3].
[1] K. G. Schilling et al., “Fiber tractography bundle segmentation depends on scanner effects, vendor effects, acquisition resolution, diffusion sampling scheme, diffusion sensitization, and bundle segmentation workflow,” NeuroImage, vol. 242, p. 118451, Nov. 2021, doi: 10.1016/j.neuroimage.2021.118451.
[2] L. C. Liebrand, G. A. van Wingen, F. M. Vos, D. Denys, and M. W. A. Caan, “Spatial versus angular resolution for tractography-assisted planning of deep brain stimulation,” NeuroImage: Clinical, vol. 25, p. 102116, Jan. 2020, doi: 10.1016/j.nicl.2019.102116.
[3] J.-D. Tournier, F. Calamante, and A. Connelly, “Improved probabilistic streamlines tractography by 2nd order integration over fibre orientation distributions,” p. 1, 2010.
-
- 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?
I like the general approach taken by this work, and I believe that it has the potential to provide a new set of tract-based features to enrich tract-based image analysis. However, the given analysis does not fully explain the benefits (or drawbacks) of PS features versus previous, more simple methods. Additionally, the main result is somewhat difficult to interpret given a lack of comparison methods and some unclear figure layout.
- 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.
The authors’ response to my 7-1 and 7-2 comments is encouraging. I would have liked to see more comparisons to more clearly understand the unique properties of the proposed feature maps, but that is difficult when the authors are not able to perform additional experiments. Assuming that the authors add the stated clarifications and supplemental content, I suggest accepting this work.
Author Feedback
Thanks for the reviewers & ACs. Code & related data will be released. To R1: [a&b meaning?] Right. We will add them to the final version. [Negative & positive PS values?] The sign arises from the directionality of the unified coordinate system, as PS inherently encodes orientation. [Meaning of 3rd order?] Encodes 3D torsion/spatial helicity. Positive value: right-handed twist. Negative value: left-handed twist. [Sampling method for point & window] Streamline resampled at 0.6 mm (<½ voxel size) using the tckresample command. The window length was 3 points with a step of 1. [Average PS within a voxel] Considering the sign of PS, we perform averaging separately for positive and negative values, which can enhance the stability within a voxel and reduce sensitivity to noise. Ignoring this distinction would render the averaging meaningless. [1D line cannot describe fiber] Length inherently aids curve characterization. PS (orders 1-3) completely encode fibers. [Current dMRI/tractography limitations fundamentally constrain fiber crossing and kissing] This is an unsolved technical challenge in dMRI, but our analysis focuses on examining tractogram-derived point-level features. [PS resolution] While tractogram accuracy is limited by dMRI resolution, PS maps can theoretically achieve super-resolution through continuous fiber—a benefit demonstrated in Figure 2. This approach significantly enhances clinical data utility. PS map sampling parameters were conservatively set based on native dMRI resolution, ensuring practical applicability. The term “super-resolution” follows the convention established by F. Calamante et al. [ref8]. [Data analysis lacks convincing validity] Our longitudinal MDD analysis employed standardized pipelines and FDR correction. It revealed treatment-related PS alterations correlating with clinical improvement—methodologically robust and biologically valid findings, lacking evidence for the concerns. To R2: [1] See [sampling method for point] of response to R1. [2] The contents will be rewritten. [3] Formula 6 is a concretization of formula 5 in this work, and this formulation is consistent with formula 2 in Ref11. [4] Figure 3 shows regions with statistically significant PS longitudinal differences in the MDD (scaled down for space). Complete results will be provided in the supplementary table. [5] We focus on analyzing only these differentially expressed regions (Figure 3) rather than all PS features, as demonstrating significant correlations yields meaningful clinical insights. [6] Age (x̄± s) 31.49±8.17; gender (M/F) 20/31. We focused on a challenging scenario [ref1]. We employed traditional statistics (not complex ML methods) using PS-derived feature maps and found pre-/post-treatment brain differences surviving FDR correction. These differences correlated significantly with clinical improvement rates, preliminarily validating the method’s efficacy. However, single-site data necessitate multi-center validation, a limitation acknowledged in the discussion. To R3: [7-1] We will cite our ISMRM-accepted abstract in the final version, providing another validation scenario—brain gender. [7-2] The dataset initially yielded no significant findings via FA map analysis, prompting our adoption of PS. Details will be provided in the final version of the supplementary materials. [7-3] For dMRI data, EPI correction was performed via the pnl_epi command. You are right – a double-masked trial is an ideal study design. However, given the nature of real-world clinical data, implementing such a design is impractical. Similar experimental setups are standard in the field, and we will address this as a limitation in the discussion. [10-1/-2/-3] We will provide a supplementary PS abbreviation table, optimize Figure 3 layouts with clarified region labeling in captions, and enhance Figure 1 with additional PS examples in the final version. [10-4] See response to R1’s [PS resolution]. [10-5] Standard iFOD2 was employed (no RK4).
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.
Reject
- 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’
This paper introduces a novel approach to quantify vertex-level uncertainty in cortical surface reconstruction using clinical MRI data. Reviewers positively acknowledged the method’s originality, clinical relevance, and clear empirical demonstration of improved robustness in downstream analyses, including Alzheimer’s disease classification. Although some concerns regarding clarity and partial volume effects were raised, the authors effectively addressed these issues in their rebuttal. Considering the method’s methodological novelty, computational efficiency, and practical clinical impact, the paper is recommended for 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’
N/A