Abstract

Recent advances in diffusion-based generative models have substantially improved image reconstruction from fMRI signals. However, most existing approaches rely on visual cortical activity, flatten voxel responses into unstructured representations, and require subject-specific adaptation. Here we propose ConnecToMind2, a whole-brain neural decoding framework that models fMRI activity at the region level and incorporates structural connectivity (SC) priors. Whole-brain signals are first aligned to a shared anatomical space and projected into regional embeddings that preserve cortical organization. These embeddings are then aligned to the CLIP image embedding space via a Connectome-guided Q-Former, where SC priors guide cross-attention to enforce inter-regional interactions. We evaluate our model on the Natural Scenes Dataset (NSD) under intra-subject and inter-subject settings. ConnecToMind2 achieves competitive performance across perceptual and semantic metrics. Ablation analysis confirms that essential information for reconstruction is distributed throughout the whole-brain rather than confined to the visual network. All codes for this study are publicly available at GitHub (https://github.com/aimed-gist/ConneToMind2).

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/1776_paper.pdf

SharedIt Link: Not yet available

SpringerLink (DOI): Not yet available

Supplementary Material: Not Submitted

Link to the Code Repository

https://github.com/aimed-gist/ConnecToMind2

Link to the Dataset(s)

N/A

BibTex

@InProceedings{BaeGun_ConnecToMind2_MICCAI2026,
        author = { Bae, Gunwoo AND Kim, Yeonwoo AND Bae, Junha AND Kim, Mansu},
        title = { { ConnecToMind2: Inter-Subject fMRI Decoding via Whole-Brain Connectome-Guided Alignment } },
        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

    Summary: This paper presents a very interesting inference about the way fMRI are used to reconstruct perceived images. Previous literature heavily focused on using FCs gathered from visual cortex region for image construction. From a neurological perspective (provided in the lit review), they argue that the perceived vision encompasses multiple brain regions.

  • 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 proposition in itself is pretty strong in my opinion, and the results also signify their hypothesis. It’s also a significant finding that, shifting our focus from designing subject-adaptive layers for fMRI reconstruction to inclusion of brain regions in the same anatomical space would be more beneficial to generalize the perceived vision across subjects. In 2 stage training (Section 2.4): Methodically, the freezing of self-attention and unfreezing for the correct attention computation specific to the task requirement was praiseworthy. Choice of intra-inter difference metrics for quantifying the results was also a major strength.

  • 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.There is no discussion on the performance results for the high-level metrics under the Brain Diffuser Protocol in table 1.

    2.No analysis as to what the high-level performance metric results in table 1 or figure 3 signify in terms of inclusion of brain regions in same anatomical space or their exclusion.

    3.It is evident that the reconstructed image without the associative networks lack clarity. Since the associative networks are doing a different task than capturing spatial details, what exactly is the associative networks capturing and why are they unable to produce the required clarity in the reconstructed images? 4.This paper requires a discussion section on (a) how/why Associative Networks for semantic details struggle in certain reconstructions. (b) How do the uncaptured semantic details of the images relate to the FCs in the same anatomical space?

    5.Results on other datasets will enable the community to evaluate the robustness and generalizability

  • 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.

    (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?

    There are significant methodological improvements, and the results are impressive.

  • 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 #2

  • Please describe the contribution of the paper

    This paper describes a novel approach towards the fMRI-to-image decoding task. The proposed approach enables region-specific and SC-aware during the encoding, and achieves significantly superior performance than baseline approaches in low-level evaluation metrics.

  • 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 clearly explained and well justified by experiments.

    2.The per-ROI embedding and SC-aware designs are novel and helpful for improving the explainability of networks.

    3.Bridging the fMRI-to-image task from 1-d flattened pixels from the whole image into ROI-specific embedding has broader implications to connect the decoding task with previous brain network studies.

  • 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 showing significant improvement on the low-level metrics, the performance of the proposed approach in high-level metrics are inferior to baselines. Discussions on potential causes would strengthen the argument.

    2.The interpretation based on atlas parcellation is great, but it would be interesting to check if the trained model weights show the same pattern of attention focus as prior knowledge of brain networks.

    3.More details on baseline implementations would facilitate understanding (e.g. which used DWI, which included atlas encoding).

  • 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.

  • 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

    Please see weakness

  • 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?

    Formulating fMRI-to-image decoding task as per-ROI embedding is important for connecting this field with brain network analysis. The experiments also show significant improvement in low-level metrics compared to baselines.

  • 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

    This paper proposes ConnecToMind2, a whole-brain fMRI-to-image reconstruction framework designed for inter-subject decoding. Instead of using flattened voxel responses from visual cortex alone, the method constructs region-level cortical embeddings in a shared anatomical space and aligns them to CLIP image representations through a connectome-guided Q-Former that incorporates structural connectivity priors. The aligned features are then used to condition a pretrained diffusion model for image reconstruction. Experiments are conducted on NSD under both intra-subject and inter-subject settings, together with a regional ablation study.

  • 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 paper addresses an important problem in neural decoding, namely improving cross-subject transfer rather than relying only on subject-specific reconstruction models. The motivation for moving beyond flattened voxel representations is reasonable, and the use of region-level embeddings provides a more structured formulation than standard vector-based pipelines. 2.The coherence of the proposed framework. The regional embedding module, connectome-guided alignment, and two-stage training strategy are consistent with the stated goal of anatomically grounded inter-subject decoding. Evaluation in both intra-subject and inter-subject settings is also a positive aspect. 3.The regional ablation excluding early visual versus higher-order regions is also useful. It provides some support for the broader claim that information relevant to reconstruction is distributed beyond the classical visual network.

  • 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 manuscript does not yet clearly separate the proposed contribution from closely related recent shared-subject and cross-subject decoding methods, so the novelty positioning remains somewhat incomplete. 2.The strongest gains are concentrated in low-level fidelity metrics, while semantic metrics are more mixed. 3.The proposed method differs from compared baselines in both architecture and input scope, making it difficult to isolate how much improvement comes from whole-cortex coverage versus the specific modeling design. 4.Results are averaged across only four NSD subjects without subject-wise breakdown or statistical analysis, and the practical meaning of “whole-brain” is narrower than the title suggests because the implementation is based on cortical Schaefer-100 parcels.

  • 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.

  • 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 is a well-motivated paper addressing an important direction in neural decoding, and the framework is technically coherent. The inter-subject setting is meaningful, and the low-level reconstruction gains are encouraging. However, the current version remains slightly below threshold for a methods-paper recommendation because the contribution is not yet positioned sharply enough relative to recent related work, the gains are uneven across metric families, and the current experiments do not fully isolate where the improvements come from. Limited subject-level statistical reporting is an additional weakness.

  • 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

N/A




Meta-Review

Meta-review #1

  • Your recommendation

    Provisional Accept

  • 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 significance of this work is high, as inter-subject neural decoding remains an important challenge, and moving beyond visual-cortex-only pipelines is scientifically meaningful. Although the reviewers raised concerns regarding the interpretation of mixed high-level metric results, limited details on baseline implementations, and insufficient analysis of the contributions of associative networks, they generally agree on the significance and potential of this work.



back to top