List of Papers Browse by Subject Areas Author List
Abstract
Graph-based models have shown promise in neuroimaging by identifying abnormal functional brain connectivity in psychiatric disorders. However, current approaches for functional Magnetic Resonance Imaging analysis have several limitations. Most methods estimate connectivity only from time-domain correlations, ignoring frequency dependent phase synchronization that provides complementary information about neural communication. In addition, existing models typically learn representations using diagnostic labels. For disorders such as Major Depressive Disorder, these labels are heterogeneous and site-dependent, which limits generalization across institutions and populations when labeled data are scarce. Finally, reliability and equity are rarely considered during representation learning, leading to demographic bias and inconsistent subgroup performance. To address these challenges, we propose RESOLVE (Robust and Equitable Self supervised learning for functiOnaL connectiVity modEling), a self-supervised multi-domain connectome representation learning framework. RESOLVE employs dual graph encoders to learn from correlation- and coherence-based connectivity and promotes consistency across the two domains. We further introduce a topology-preserving reconstruction objective that maintains the intrinsic organization of the brain network. To improve reliability, we learn demographically invariant representations by aligning subgroup distributions in the latent space using an Optimal Transport objective. Experiments on the REST-meta-MDD dataset demonstrate the effectiveness of the proposed framework in balancing diagnostic performance and demographic bias compared to existing methods. Source code is available at: https://github.com/shashivipul/RESOLVE_2026.git.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/4867_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
https://github.com/shashivipul/RESOLVE_2026.git
Link to the Dataset(s)
N/A
BibTex
@InProceedings{BarJyo_EquityAware_MICCAI2026,
author = { Barman, Jyotismita AND Singh, Vipul Kumar AND Yusuf, Mohammad AND Gandhi, Tapan K. AND Jayadeva AND Kumar, Sandeep},
title = { { Equity-Aware Self-supervised Multi-domain Connectome Representation Learning for Major Depressive Disorder Diagnosis } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 16880},
month = {September},
page = {pending}
}
Reviews
Review #1
- Please describe the contribution of the paper
This paper presents RESOLVE, a self-supervised graph representation learning framework that leverages functional connectivity in both the time and frequency domains. To ensure reliability and equity, it incorporates a topology-preserving reconstruction objective to maintain brain network organization and an Optimal Transport-based alignment module to produce demographically invariant representations, thereby reducing bias across subgroups. Experiments on the REST-meta-MDD dataset show the superiority of the proposed method.
- 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.
This method incorporates frequency-domain coherence and time-domain correlations, which may capture a more comprehensive representation of neural communication compared to single-domain approaches.
- 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）Most of the comparison models chosen by the authors were published in/before 2022.The article lacks comparisons with more recent technologies. Furthermore, the lack of comparisons with cross-domain (or domain-adaptive/generalization) methods raises concerns about the robustness and reliability of the reported results, especially when distributions shift. It is recommended to enhance this aspect of the paper by incorporating comparisons with state-of-the-art techniques to improve the overall innovativeness of the study. 2) The authors compute coherence between ROIs only within the 0.01–0.1 Hz frequency band. It is unclear whether the same band-pass filtering (0.01–0.1 Hz) was already applied during preprocessing. If so, the subsequent analysis is essentially conducted within the same frequency range, meaning that the frequency setting is consistent and does not introduce any additional information filtering step. This raises the question of whether potentially informative higher-frequency fluctuations are unintentionally excluded. A discussion or ablation analysis on different frequency bands would strengthen the paper. 3) Experiments are conducted only on REST-meta-MDD, which may limit generalizability to other disorders, such as ASD and ADHD. 4) Hyperparameters, such as the k-nearest-neighbor value (k=10) and the coherence threshold (0.3), are selected empirically without a detailed sensitivity analysis. 5) How can the data be split within a 5-fold cross-validation to obtain 70% training, 10% validation, and 20% testing sets?
- 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 has provided an anonymized link to the source code, dataset, or any other dependencies.
- 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 is generally well organized and clearly written, with a logical flow from method and experimental evaluation. However, the reviewer still has some concerns regarding the methodological motivation and certain experimental results. If the authors can adequately address the aforementioned issues, the reviewer would be willing to consider raising the score.
- 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.
All concerns have been addressed.
Review #2
- Please describe the contribution of the paper
1.It employs dual graph encoders to learn from correlation- and coherence-based connectivity and promotes consistency across the two domains. 2.It introduces a topology-preserving reconstruction objective that maintains the intrinsic interactions between brain regions across connectivity domains. 3.The framework incorporates an optimal transport-based latent alignment module, effectively reducing demographic bias.
- 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 designs detailed and thorough experiments for evaluation. 2.The research problem is well-motivated, and the proposed method demonstrates strong novelty.
- 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 naming convention for the proposed algorithm’s abbreviation is unclear. 2.The background information in the Introduction is overly detailed. It is recommended to condense this section to focus on the core problem. 3.The feature fusion strategy is relatively simplistic.
- 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 has provided an anonymized link to the source code, dataset, or any other dependencies.
- 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 provides solid experimental evaluation and sound research motivation.
- 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.
I recommend acceptance, as the authors have satisfactorily justified their feature fusion design and committed to addressing the presentation and naming issues in the final revision.
Review #3
- Please describe the contribution of the paper
The authors introduce RESOLVE, a fairness-aware, self-supervised framework for Major Depressive Disorder diagnosis. Additionally, they propose a topology-preserving reconstruction objective that retains intrinsic interactions between brain regions across connectivity domains.
- 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 experimental results reflect experiments on multiple sites as well as cross-site generalizability, indicating the strength of the model in terms of potential clinical viability.
- The use of the FATE metric in the evaluation of RESOLVE and the compared baselines plays a major role in validating the author’s claim that their framework reduce disparities across demographic subgroups and promotes fair predictions.
- Figure 2 is a good addition, as it shows that the authors performed numerous ablation experiments, robustness analysis, and evaluated performance against numerous labeled data settings.
- The authors conducted sufficient statistical analysis to demonstrate that their results were actually significant, and through Figure 3, they demonstrated that their method is capable of producing biologically meaningful results.
- 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.
- Even though the authors do evaluate on multiple sites from the REST-meta-MDD dataset, the work is still slightly limited by the fact that the authors do not evaluate on datasets besides this one. It would have been nice to see a bit more discussion on how the authors think their method would generalize to different datasets, particularly in terms of how well the fairness-aware aspect of their work would translate to datasets with different demographic populations.
- I think at times the presentation is a little weak. I imagine that a lot of this is due to space constraints, but I feel that if accepted the paper could benefit from another pass through to make it slightly better to read. Just as an example of the type of thing I’m talking about, in Section 2, the authors introduce “Problem 1”, but never formalize any other problems. It seems odd to formalize something as Problem 1 if that is not followed by other problems.
- 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 has provided an anonymized link to the source code, dataset, or any other dependencies.
- 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
I do have one question for the authors. I’m not entirely sure I understood what the READOUT operator the authors used actually did. I would appreciate if the authors could prove a more layman explanation for what this component does.
- 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?
I think this is a good paper that has potential benefits in a clinical setting, especially due to the fairness-aware aspect. There are some minor things I think could be addressed, but I can see myself boosting my rating higher if the authors provide a strong rebuttal to my and other reviewer’s comments.
- 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.
I thank the authors for their diligent rebuttal. My initial review skewed positive, with just some minor comments that I felt were easy to address in the rebuttal. I do feel like those comments were adequately addressed by the authors, and I feel like their responses to the other reviewers’ concerns for satisfactory as well. Because of that, I am recommending that this paper be accepted.
Author Feedback
We thank reviewers for constructive comments. We are encouraged that they recognized strong motivation(R4,R5), clarity(R1), novelty(R4), thorough evaluation/ablations(R4,R5) & clinical relevance of equity-aware diagnosis with biologically meaningful results(R5).
Q1:Baseline. GCDA[24] & CoRE-BOLD[22] are both 2025 SOTA. GCDA is an SOTA SSL benchmark designed for cross-site generalization. RESOLVE outperforms both in utility & FATE under cross-site distribution shift & Gaussian noise. Prior generalization-based approaches don’t explicitly address fairness.
Q2:L_DP loss. Section 3.1 provides 3 quantitative validations of L_DP: (i)Table 1 shows RESOLVE is the only method achieving positive FATE on both cross-site protocols, reducing demographic parity difference more effectively than baselines.(ii)In Fig.2(c),RESOLVE w/o L_DP shows a clear rise in ΔDP, showing its contribution to reducing subgroup disparity.(iii)Fig.2(b) shows RESOLVE preserves best utility-fairness tradeoff across labeled-data fractions.
Q3:Dataset. MDD is used as a proof-of-concept due to established gender bias. However, RESOLVE is disease/cohort agnostic. Our cross-site validation (S20→S14 & S20→S21) already evaluates realistic distribution shifts & demographic composition, where RESOLVE achieves SOTA performance. The OT alignment performs non-parametric matching between two subgroup distributions, making it applicable to arbitrary binary subgroup populations. We also validated RESOLVE on ABIDE dataset with consistent improvement. Due to rebuttal policy,results will be detailed in GitHub repo upon acceptance.
Q4:Novelty. RESOLVE is, to our knowledge, 1st equity-aware multi-domain SSL framework for connectomes without task labels, a setting previously unaddressed in neuroimaging. It delivers 3 connectome specific contributions: (a)Representation-level fairness in SSL: since diagnostic labels are unavailable during pretraining, RESOLVE enforces demographic invariance in fused latent space via OT alignment, unlike prior methods that mainly operate in supervised prediction space (b)Joint multi-domain reconstruction: RESOLVE reconstructs both correlation & coherence based connectomes from a fused embedding, rather than a single connectivity view, introducing a novel SSL pretext (c)Graph-level cross-domain SSL: Jointly modeling time & frequency domain connectivity while preserving ROI interactions. No prior method unifies SSL pretraining, multi-domain connectome learning, & equity-aware latent representation learning.
Q5:Frequency. The 0.01–0.1 Hz range is widely established as Low-Frequency Fluctuation(LFF) band in seminal rsfMRI literature, capturing dominant spontaneous BOLD fluctuations while attenuating low frequency drift(<0.01Hz) & physiological noise(>0.1Hz)[C1,C2]. The same filter is applied during preprocessing to isolate this signal. RESOLVE’s frequency encoder captures phase synchronization within this band, which a time-domain encoder doesn’t explicitly model. This will be clarified in camera ready.
Q6:k & threshold. Following prior literature, we tuned k∈[5,15] & threshold∈[0.1,0.6] via grid search on validation data. Due to page limit we didn’t provide the analysis.
Q7:READOUT. READOUT (Global mean pooling,in Sec.3) aggregates ROI(node)-level embeddings into a subject(graph)-level vector.
Q8:CV split. We use standard 5-fold CV where each fold holds 20% for testing; rest 80% is split as 90:10 into training and validation, yielding a 72/8/20 split. We will correct it in camera ready.
Q9:Fusion. RESOLVE aim to derive its gains from objectives in Eq. 9 rather than fusion complexity, yet outperforms SOTA baselines including transformer-based architectures. Exploring attention-based fusion is a promising future direction.
Q10:Presentation. RESOLVE acronym is expanded in Introduction; we will additionally include it in Abstract. We will rename Problem 1 to Problem Statement.
C1: Cordes et al.,AJNR, 2001 C2: Biswal et al.,Magn Reson Med. 1995
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.
Based on the three reviews, the paper has notable strengths. However, R1 raises substantial concerns about outdated baseline comparisons (mostly pre‑2022), lack of comparison with domain adaptation/generalization methods, limited frequency‑band analysis, evaluation on only one dataset (REST‑meta‑MDD), and missing hyperparameter sensitivity analysis. R2 suggests improving the fusion strategy and presentation clarity. R3 is generally positive but recommends testing on additional datasets and polishing the writing. Besides, novelty is limited. The proposed framework combines existing techniques—dual‑domain (time/frequency) connectivity, contrastive learning, reconstruction loss. More important, the paper asserts that minimizing the optimal transport loss LDP aligns subgroup representations and promotes demographic‑invariant embeddings, thereby reducing disparities. However, no solid evidence is provided to support this claim.
- 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 rebuttal does a good job in clarifying the comparison baselines and model novelty in the paper. Post rebuttal, all reviewers lean towards accepting the paper and I am in agreement as well. All major concerns seem to be addressed.
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 contributes an equity-aware self-supervised representation learning framework for connectome-based MDD diagnosis, and its attempt to treat fairness as a central design objective is a meaningful strength. The rebuttal appears to have responded effectively by clarifying why fairness is not merely an auxiliary analysis and by better grounding the method in the observed utility-fairness behavior, which helped maintain this paper qualified 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.
The paper addresses a clinically relevant and timely problem in fairness-aware self-supervised connectome learning. The rebuttal adequately addresses the main concerns about recent baselines, the demographic alignment loss, frequency-band choice, hyperparameter selection, fusion design, and cross-validation split. All active reviewers support acceptance after rebuttal, and the remaining issues are mainly presentation and clarification rather than core validity.