List of Papers Browse by Subject Areas Author List
Abstract
T2-weighted FLAIR MRI at 5T reveals cortical laminar patterns via a distinct hypointense band, offering enhanced contrast over 3T and greater clinical feasibility than 7T. Accurate whole-brain reconstruction of this band is crucial for analyzing signal-defined intracortical laminar morphology but faces a trade-off: optimization-based methods achieve high fidelity yet are computationally prohibitive, while deep learning methods are fast but are typically designed only to reconstruct white matter and pial surfaces. Here, we propose BiC-ODE (Bidirectionally Coupled Neural ODEs), the first unified deep learning framework to achieve fast and accurate surface reconstruction of the intracortical hypointense band from 5T MR images, bridging the gap between high-resolution intracortical-band surface reconstruction and clinical feasibility. Our core idea is to jointly deform the white matter surface outward and the pial surface inward, until they align with the inner and outer boundaries of the intracortical hypointense band in 5T FLAIR images, respectively. Extensive experiments demonstrate that BiC-ODE: (1) achieves high reconstruction fidelity by precisely aligning with the intracortical hypointense band, fully leveraging the enhanced laminar contrast and clinical accessibility of 5T MRI; (2) delivers rapid inference speed, enabling efficient processing of large-scale datasets; and (3) promotes consistent multi-surface modeling through a bidirectional coupling strategy, reducing ordering violations among the nested surfaces. Thus, our framework enables reliable analysis of signal-defined intracortical morphology across large-scale 5T cohorts. Our code is publicly available at https://github.com/shuicao-water/BiC-ODE-code.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/1454_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
https://github.com/shuicao-water/BiC-ODE-code
Link to the Dataset(s)
N/A
BibTex
@InProceedings{CaoShu_BiCODE_MICCAI2026,
author = { Cao, Shui AND Teng, Lin AND Shi, Lu AND Lian, Zifeng AND Liu, Jiameng AND Shi, Feng AND Sun, Kaicong AND Meng, Xiangshui AND Shen, Dinggang},
title = { { BiC-ODE: Bidirectionally Coupled Neural ODEs for Rapid Laminar Surface Reconstruction of the Human Cortex from 5T MRI } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 16894},
month = {September},
page = {pending}
}
Reviews
Review #1
- Please describe the contribution of the paper
Recent advances in MR technology and imaging protocols allow zooming into the cortical layer structure. This submission uses FLAIR-weighted images acquired at 5T at a spatial resolution of ~0.5 mm to detect the “intracortical hypointense band”. A well-known, OED-based method to reconstruct the GM/WM and WM/CSF interfaces is employed here, but optimzed using at neural network. This approach provides - first of all - a considerably improved computational efficiency over a conventional approach, developed by the same group and provided here as an unpublished supplement.
- 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 within the scope of this conference and of potential interest to its audience. The work described here is an incremental but noteworthy advance over previous work by a similar group of authors. The text is straightforward to understand for a reader with knowledge in recent approaches for image 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.
For more medically-inclined readers, it would be interesting to learn which anatomical layer(s) are represented by this somewhat ominous “hypointense band”. What is the anatomical (clinical) relevance of this work?
- 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.
- Based on your review and your understanding of the MICCAI Scientific Code of Ethics, do you believe this submission may involve a potential ethics concern or violation?
N/A
- 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/2026/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 — marginally below the acceptance threshold, but would not mind if accepted, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
This submission has a strong overlap with the (unpublished? submitted?) manuscript provided as supplemental material.
- 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
(1)This is the first study to propose a deep learning surface reconstruction method specifically targeting the cortical low-signal band in 5T MRI FLAIR images. (2)A Bidirectional Coupled Neural ODE (BiC-ODE) framework is proposed, which simultaneously drives the white matter surface outward and the pial surface inward to reconstruct the inner and outer boundaries of the low-signal band. (3)Three task-specific loss functions (depth consistency, layer thickness, and intensity band loss) are designed, combined with supervision from a “silver standard” generated by an optimization-based method. (4)Validation was performed on a relatively large 5T MRI dataset (270 cases), demonstrating fast inference speed and a low Surface Ordering Violation Ratio (SOVR).
- 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)The method is specifically designed for the unique high-contrast low-signal band present in 5T FLAIR images. (2)The proposed SOVR metric is reasonable and practical, effectively quantifying the preservation of the nested ordering of the four cortical surfaces. (3)The experimental comparisons are comprehensive, including multiple deep learning methods and ablation studies, with results showing advantages in metrics such as ASSD, HD90, and SOVR.
- 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 claims that “bidirectional coupled deformation” is the core innovation. However, in the actual implementation, it merely concatenates the features of the WM and pial surfaces and feeds them into a shared network. This “coupling” approach is relatively simplistic and lacks theoretical depth or significant mechanistic innovation, differing little from standard multi-task joint prediction. The technical contribution of the “Cross-Surface Coupling” section appears significantly overstated. (2)The authors only cite a paper currently “under review” as the method for generating the silver standard, without providing any specific description, validation, or open-source code for that method. This substantially undermines the reproducibility and credibility of the experimental results. Reviewers cannot determine whether the silver standard itself is accurate, nor can they assess whether BiC-ODE offers genuine improvement over that optimization method. (3)The method was validated exclusively on 5T FLAIR data from a single institution and was not tested on 3T, 7T, or 5T data from other vendors. For a deep learning model, this is a major limitation; it fails to demonstrate the method’s generalizability across different acquisition conditions, field strengths, or populations. (4)The authors did not directly compare BiC-ODE against the optimization method used to generate the silver standard (e.g., in terms of accuracy, inference time, robustness). Readers cannot judge whether the deep learning method genuinely outperforms the traditional approach, nor evaluate the suitability of the silver standard as a reasonable supervision target.
- 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.
- Based on your review and your understanding of the MICCAI Scientific Code of Ethics, do you believe this submission may involve a potential ethics concern or violation?
N/A
- 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/2026/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 — marginally below the acceptance threshold, but would not mind if accepted, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The paper addresses a noteworthy topic, and the preliminary results show some potential. However, the model’s novelty is limited, the origin of the silver standard is opaque, generalizability has not been validated on external datasets, and a direct comparison with the optimization method used to generate the silver standard is lacking.
- 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 authors have addressed most of my concerns. Overall, despite some limitations, the method proposed is meaningful.
Review #3
- Please describe the contribution of the paper
This paper introduces a novel method for modeling the hypointense band of the cerebral cortex in 5T FLAIR MRI. While existing cortical surface reconstruction methods typically model the inner and outer cortical surfaces (white matter and pial), they do not account for the hypointense band visible in T2-weighted 5T FLAIR images. The proposed method deforms pre-existing white matter and pial surfaces to estimate the inner and outer cortical band surfaces using a neural ODE-based framework with bidirectional feature exchange between the two surfaces. Three task-specific loss functions are introduced to guide the modeling of the hypointense band. The method is compared against deep learning-based cortical surface reconstruction approaches, outperforming them in accuracy and producing fewer surface ordering errors.
- 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 introduces several novel components, particularly the bidirectional coupled information exchange between surfaces and the dedicated loss functions for hypointense band modeling.
- This appears to be one of the first approaches specifically designed for cortical hypointense band modeling, making it a genuinely novel contribution to the field.
- 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.
- Silver standard ground truth. The authors use silver standard hypointense band surfaces generated by an optimization method that is currently under review (provided as supplementary material). While using an algorithmic silver standard is generally acceptable when it is well-validated, relying on an unpublished and unreviewed method as the evaluation reference raises concerns about the reliability of the results. Since the referenced paper is anonymized — suggesting it may be from the same group — the authors presumably have access to manual annotations. I would encourage the authors to use manual labels as ground truth, or at minimum provide a more thorough justification for why the silver standard is sufficiently reliable for this purpose.
-
- Cross-surface coupling is not evaluated in the ablation study. The bidirectional coupled information exchange is a core architectural contribution, yet its individual impact is not isolated in the ablation study. It would be informative to understand how much of the performance gain is attributable to this design choice versus the specialized loss functions.
- 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.
- Based on your review and your understanding of the MICCAI Scientific Code of Ethics, do you believe this submission may involve a potential ethics concern or violation?
N/A
- 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/2026/en/REVIEWER-GUIDELINES.html
- Page 2: The authors assert that deep learning methods for cortical surface reconstruction “treat the cortex as a homogeneous signal layer… failing to capture the distinct contrast of the hypointense band.” This claim is not entirely convincing as it is not immediately clear why such methods would fail if they were simply trained on laminar band surfaces instead. Additionally, the claim that existing methods lack multi-surface modeling overlooks the Vox2Cortex family of methods [1, 2], which perform joint modeling of cortical surfaces.
-
Ablation study: It is notable that even without the band-specific loss, the proposed method still outperforms the baselines in Table 1.This raises an interesting question about how much of the gain comes from the architecture versus the loss functions. For example, how would the method perform with only the MSE and Laplacian losses, bringing it closest in design to the baselines? Conversely, how would the baseline architectures perform when augmented with the proposed depth, thickness, and band losses? These experiments are not strictly necessary, but would be interesting to add in a journal version.
- Page 8: The authors discuss the computational efficiency of their method, but the analysis does not account for the upstream requirement of running FreeSurfer to extract the white matter and pial surfaces.
Could the authors provide a brief medical motivation for reconstructing the hypointense laminar band? For example, cortical surfaces are widely used for thickness analysis in neurodegenerative disease research. Clarifying what additional clinical value the laminar band offers beyond standard white matter and pial surfaces would help strengthen the paper’s motivation.
[1] Bongratz, Fabian, et al. “Vox2Cortex: Fast explicit reconstruction of cortical surfaces from 3D MRI scans with geometric deep neural networks.” Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2022.[2] Bongratz, Fabian et al. “Neural deformation fields for template-based reconstruction of cortical surfaces from MRI.” Medical Image Analysis 93 (2024): 103093.
- 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?
This paper makes a meaningful contribution by tackling a task that has received little attention to date. The proposed method performs well compared to relevant baselines and introduces a clean, effective architecture alongside novel loss functions tailored to the hypointense band modeling task. Addressing the concerns about the silver standard ground truth would further solidify the work.
- 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.
This work addresses a novel task, and outperforms baselines. Validation against manual ground truth would make it stronger, but I believe the silver standard comparison is well motivated.
Author Feedback
We sincerely thank the Reviewers and AC for their constructive feedback. We address the core concerns below. 1.Silver-Standard Reliability & Comparison (AC, R1-3) The silver-standard surfaces were generated by an energy-function optimization method (detailed in the supplement) that incorporates geometry- and image-guided terms. Crucially, this optimization method demonstrates excellent agreement with clinician-validated manual annotations (Dice ~96.0%/~95.4% on the primary/cross-site datasets) and high test-retest repeatability (ASSD <0.1 mm). Since manual labels are only local masks, this highly accurate method serves as our whole-cortex evaluation ground truth. This optimization method is not a competing baseline but our supervisory signal. BiC-ODE’s primary contribution is successfully matching this clinician-validated standard (Tables 1-2, Fig. 2) while reducing the optimization’s hours-long computation to seconds-level inference for large 5T cohorts. 2.Rationale & Ablation of bidirectional coupling (AC, R2, R3) We will refine claims about the coupling module. The inner and outer hypointense-band surfaces are adjacent boundaries defining a thin intracortical band. If each surface deforms independently using only local features, they develop inconsistent relative positions and severe ordering violations. BiC-ODE therefore explicitly conditions each surface’s velocity field on its counterpart to reconstruct a strictly coupled pair. We report the requested coupling ablation here. Removing cross-surface coupling (each surface concatenates only its own features) worsened ASSD/HD90 and increased SOVR to ~17% compared with BiC-ODE. This confirms that bidirectional conditioning is physically essential for preserving relative positions, proving the performance gain is not merely due to specialized loss functions. 3.Clinical relevance (R1, R3) We define the hypointense band as an MRI-visible, signal-defined band on 5T FLAIR, not as a specific anatomical layer. Nevertheless, prior studies (Refs.[1–3]) suggest that intracortical contrast reflects regional cyto-/myeloarchitectonic variation. The supplement shows preliminary clinical relevance: epilepsy patients had lower hypointense-layer thickness ratio than healthy controls, suggesting that these multi-layer metrics may complement conventional WM/pial morphometry for cortical microstructural disorders. 4.Cross-site generalizability & Clarifications on claims & Other concerns (R2,R3) (a) We tested BiC-ODE on the supplement’s cross-site 5T dataset, which was not used for training, obtaining comparable inner/outer ASSD (~0.14/0.17mm), HD90 (~0.31/0.36mm), and SOVR (~2.5%), supporting cross-site generalizability; broader multi-vendor, 3T/7T validations are vital future work. (b) We will refine the Page 2 statement. While existing methods can be retrained on laminar-band targets, retraining alone does not explicitly model the two coupled hypointense-band boundaries and the required four-surface ordering. We will properly acknowledge joint surface methods such as Vox2Cortex. (c) WM/pial surface extraction is a shared upstream step; our runtime comparison focuses on hypointense-band reconstruction. (d) Detailed coupling/loss ablations will be considered in a journal version. (e) BiC-ODE and the supplement code will be released upon acceptance.
Meta-Review
Meta-review #1
- Your recommendation
Invite for Rebuttal
- Please justify your decision. In case you deviate from the reviewers’ recommendations, explain in detail the reasons why. In case of an invitation for rebuttal, clarify which points are important to address in the rebuttal.
The most critical points to address in the rebuttal are the concerns regarding the silver standard ground truth and the lack of a direct comparison with the optimization method used to generate it. You must clearly justify its use or provide evidence of its reliability, as this undermines the credibility and reproducibility of the results. Second, the novelty of the bidirectional coupling lacks any theoretical justification or an ablation experiment.
- 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.
The authors provided validation metrics on an external dataset and detailed the high agreement between their silver standard and manual annotations. An ablation experiment was presented to demonstrate that removing cross-surface coupling increases surface ordering violations. The final recommendation is to accept the manuscript.
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.
This paper proposes BiC-ODE, a bidirectional coupled neural ODE framework for reconstructing the inner and outer surfaces of the intracortical hypointense band in 5T FLAIR MRI, supervised by an optimization-derived silver standard and guided by task-specific depth, thickness, and intensity-band losses. The reviewers agree that this is among the first approaches to target hypointense band modeling, that the task is genuinely novel, and that the method performs well against relevant baselines while preserving the nested ordering of cortical surfaces. The principal concern across the reviews and the meta-review was the reliance on a silver standard generated by an as-yet-unpublished optimization method; the rebuttal addressed this by characterizing its agreement with clinician-validated annotations and clarifying that it serves as the supervisory signal rather than a competing baseline, with BiC-ODE’s contribution being to match this reference at a fraction of the computational cost, and the reviewer who weighted this concern most heavily found it sufficiently motivated. The authors are encouraged to retain the clearer, more measured description of the bidirectional coupling module and to incorporate the requested clinical motivation and acknowledgment of prior joint-surface methods in the camera-ready version.
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.
This submission proposes a bidirectional neural ODE for reconstructing cortical hypointense bands in 5T MRI. R3 likved the methodologly on neural ODE. R1 and R2 noted issues with an unvalidated silver standard and missing ablation studies. The rebuttal clarified these concerns by showing agreemenrt between the silver standard and manual annotations, with an additional ablation study (note: no new experiments is allowed in the rebuttal, unfair to other submissions, discarding this study in this decision). The consensus is a positive impact and novelty in the method. For all these reasons, and with respect to the other submissions, the recommendation is towards Acceptnace.
