List of Papers Browse by Subject Areas Author List
Abstract
Understanding how intrinsic brain activity constrains stimulus-evoked responses is a central challenge in cognitive neuroscience. Task-based fMRI measures localized evoked activation but is costly and difficult to scale, whereas resting-state fMRI (rs-fMRI) is easier to acquire yet lacks direct stimulus interpretability. We study resting-conditioned synthesis of visually evoked fMRI (ve-fMRI): given an rs-derived intrinsic functional prior and a visual stimulus image, we predict the corresponding trial-wise activation volume. We propose Rest2Visual, a 3D encoder–decoder network that encodes the resting prior into hierarchical volumetric features and injects stimulus embeddings into the decoder via adaptive group normalization to produce stimulus-specific maps. Using large-scale rs–stimulus–evoked triplets, Rest2Visual improves voxel-wise and representational agreement over stimulus-agnostic and modality-ablated baselines, and preserves subject-specific response structure. We further use image reconstruction as a downstream probe to assess whether the synthesized activations retain stimulus information. The results suggest that individualized intrinsic organization can be transformed into stimulus-aligned activation patterns for scalable functional modeling.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/2424_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
N/A
Link to the Dataset(s)
N/A
BibTex
@InProceedings{ZhoChu_Rest2Visual_MICCAI2026,
author = { Zhou, Chuyang AND Ji, Ziao AND Liu, Daochang AND Wang, Dongang AND Liu, Xiaochen AND Yong, Ken-Tye AND Wang, Chenyu AND Xu, Chang},
title = { { Rest2Visual: Predicting Visually Evoked fMRI from Resting-State Scans } },
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
This paper studies the synthesis of visually evoked fMRI, where a subject-specific intrinsic prior derived from resting-state fMRI is combined with a visual stimulus to predict trial-wise evoked activation maps. To address this problem, the authors propose Rest2Visual, a conditional 3D encoder-decoder that preserves volumetric resting-state structure while injecting stimulus information through bottleneck conditioning and AdaGN-based decoder modulation. The paper further contributes a practical triplet-construction pipeline on the Natural Scenes Dataset (NSD), pairing resting-state priors, visual stimuli, and evoked task-fMRI scans, as well as an empirical study showing that the synthesized maps improve voxel-wise and representational agreement over simple baselines, preserve subject-specific response structure, and remain informative for downstream image reconstruction. Overall, the paper’s contribution lies in showing that individualized resting-state organization can be transformed into stimulus-aligned task responses through a multimodal conditional generation framework.
- 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 proposes an interesting multimodal formulation that combines resting-state priors and visual stimuli to synthesize trial-wise visually evoked fMRI, which is a meaningful extension beyond standard stimulus-to-brain encoding alone. The model design is well aligned with this goal, using a 3D conditional architecture that preserves volumetric structure while modulating predictions with image-derived features. The evaluation is also reasonably comprehensive, covering voxel-wise accuracy, representational structure, regional analysis, ablations, qualitative inspection, subject-specificity, and downstream reconstruction. Altogether, these elements make the work a convincing proof of concept.
- 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 experimental evaluation has several important limitations: i). Although the full model outperforms random prediction, the resting-state-only baseline, and noise-based ablations, these comparisons are relatively weak and do not establish that the method is competitive with more realistic learned alternatives.
ii). The results suggest limited cross-subject generalization. As shown in Table 3, performance on an unseen subject drops substantially in terms of Pearson correlation, RSA, and MAE relative to the standard setting. This indicates that the method relies heavily on subject-specific patterns observed during training and does not yet learn a robust subject-agnostic mapping. Because one of the paper’s motivations is to reduce the burden of task-fMRI acquisition, weak generalization to new subjects significantly limits the practical utility of the approach.
iii). Although, the model improves voxel-wise and representational metrics, the downstream reconstruction results remain noticeably worse than those obtained from ground-truth fMRI. As shown in Table 2, reconstructions from the predicted maps lose information relative to reconstructions from real NSD beta maps, both in low-level similarity and in higher-level semantic measures. This weakens the claim that the synthesized activations can serve as a close substitute for true task-evoked fMRI. They are clearly useful, but they still remain meaningfully inferior to actual measured responses.
iv). The empirical evidence remains largely proof of concept on a small and highly controlled dataset. The study is based on only eight NSD subjects, all collected under a high-quality experimental protocol. While the in-distribution results are promising, this is not sufficient to support stronger claims about robustness, scalability, or clinical relevance. The weak performance on unseen subjects further reinforces this concern.
- 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
The experimental section would be stronger with more competitive baselines. While comparisons to random prediction, resting-only prediction, and modality-noise ablations are useful sanity checks, they do not show how the method performs relative to stronger learned alternatives.
Generalization is another important issue. Although results on unseen stimuli are encouraging, the substantial drop on unseen subjects suggests that the method still depends heavily on subject-specific patterns seen during training. Since one motivation is to reduce the burden of task-fMRI acquisition, improving robustness to new subjects seems important.
The discussion of practical or clinical relevance should also be more cautious. Because the experiments are conducted on a small, highly controlled NSD cohort of healthy participants, the work is better interpreted as a proof of concept than as evidence of clinical feasibility.
- 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 — marginally above the acceptance threshold, but would not mind if rejected, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
While the proposed idea is interesting and technically feasible, the experimental results highlight several limitations that weaken the overall empirical case and prevent a clear recommendation for acceptance. Nevertheless, as an initial proof of concept, the work explores a promising direction and may therefore warrant a weak accept.
- 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 proposes Rest2Visual, a method to predict visually evoked fMRI beta maps using intrinsic functional priors from resting-state fMRI and external visual stimuli. The model outperforms baseline methods, enables meaningful visual image reconstruction from predicted brain activity, and generalizes well to novel stimuli. By linking resting-state brain organization to task-driven responses, the work provides a new way to estimate task activations without explicit task scanning, which is valuable for clinical and neuroscientific applications. This multi-level evaluation supports the effectiveness of the proposed method and rules out explanations based on noise or intrinsic topology alone.
- 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 introduces a novel paradigm of resting-conditioned prediction of visually evoked fMRI activity, which aims to synthesize single-trial beta maps using only a resting-state fMRI intrinsic prior and a visual stimulus. This is conceptually interesting and practically valuable because it connects the brain’s intrinsic functional architecture (from rs-fMRI) to stimulus-driven responses, potentially reducing the need for lengthy task fMRI scans. The idea of using resting-state as an individual prior for task prediction is under-explored and represents a clear conceptual advance. The authors compare against two strong non-parametric baselines (Random and Resting), assess performance at both voxel-wise and representational similarity levels, and further validate results using a downstream image reconstruction model (MindEye2).
- 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.Some experimental details require further clarification, including key parameters such as Ds, Hs, Ws, as well as the number of training samples. 2.The paper uses only 8 participants, a small sample size that may limit the reliability of the reported performance and the generalizability of the proposed method. 3.The results in Table 3 show very poor cross-subject generalization, which largely reduces the ability of model. As shown in Table 3, the model exhibits substantially poor cross-subject generalization, which considerably weakens its practical utility and reliability. 4.The fMRI signal of the PCC region already reflects the BOLD hemodynamic response; convolving it with a canonical HRF during seed-based GLM analysis appears redundant and physiologically unjustified. 5.While predicted activation maps are qualitatively similar to ground truth, noticeable inconsistencies exist in Figure 2.The feasibility of replacing real task-fMRI with predicted data has not been sufficiently established.
- 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.
(4) Weak Accept — marginally above the acceptance threshold, but would not mind if rejected, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
This paper presents a novel and well-motivated framework to predict visually evoked fMRI from resting-state priors and visual stimuli, with a rigorous multi-level evaluation including strong baselines and image reconstruction. It offers clear conceptual value and potential for clinical applications. However, the score is limited by notable weaknesses: a small sample size of 8 subjects, very poor cross-subject generalization, unclear experimental parameters, a physiologically redundant HRF convolution step, and visible inconsistencies in predicted activation maps. These issues restrict practical applicability but do not invalidate the core contribution, resulting in a solid but not outstanding paper.
- 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
The paper introduces a new fMRI analysis pipeline for visual stimuli. This new pipeline, called Rest2Visual, uses resting-state fMRI data to extract subject-specific priors that, when paired with the stimulation image, allow for the extraction of statistical parametric maps for the task (what the authors call “predicted fMRI”). This predicted fMRI is then validated against the maps yielded by a GLM-based analysis. This paper is particularly interesting as it demonstrates that resting-state fMRI data may be used to supplement task fMRI analyses, for example by establishing priors.
- 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 grounded and the methods are solid. It explores an interesting approach of using data from rs-fMRI to establish priors that are then used to estimate GLM-based maps for task-fMRI. The paper also explores both a full brain analysis and an analysis restricted to only the visual cortex.
- 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.
Firstly, in my opinion it is important to underscore that this method is not estimating actual BOLD signal measured in a fMRI acquisition but rather the statistical parametric maps that are the output of a GLM-based analysis. This confusion is accentuated by the authors calling the output of their pipeline “predicted fMRI”. This distinction is important as the fMRI signal is much more complex as it shows variations in oxygenated blood in the brain rather than statistical linear correlations between this signal and the paradigm design. Although the contribution of this work is still valid, estimating GLM-based statistical maps and estimating whole-brain BOLD signal are different challenges. Secondly, I think the paper could be improved by using a simple quantitative measure of the overlap between the predictions and the NSD-provided ground truth beta maps. In Fig. 2 there could be a common simple metric (overlap, Jaccard, Dice-Sørensen, etc.) that allows readers to quickly understand how much the predicted maps overlap with to the ground truth. Lastly, I understand that the paper is already on the limit of the allowed length, but I think the issue of limited subject diversity, and, thus, generalization. Particularly when the pipeline showed such poor performance for the unseen subject test (which may have been expected), I believe this requires further explanation (i.e. why the priors are so subject-specific).
- 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.
(4) Weak Accept — marginally above the acceptance threshold, but would not mind if rejected, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The paper is overall well grounded and built upon solid methods. It tackles a relevant challenge. However, in my opinion, the three main points raised in the weaknesses section (estimating GLM-based maps and not fMRI signal, not having a simple measure of overlap between estimated and ground-truth maps, and not exploring the poor performance from the unseen subject test) hold it back from an “Accept”.
- 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
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 reviewers were uniformly enthusiastic about the paper. However, they did note a few areas of clarification that the authors should address in the final version.
