arxiv: 2604.10116 · v1 · submitted 2026-04-11 · 💻 cs.CV · cs.AI

Recognition: unknown

A Dual Cross-Attention Graph Learning Framework For Multimodal MRI-Based Major Depressive Disorder Detection

Nojod M. Alotaibi , Areej M. Alhothali

Authors on Pith no claims yet

Pith reviewed 2026-05-10 15:50 UTC · model grok-4.3

classification 💻 cs.CV cs.AI

keywords multimodal MRImajor depressive disordercross-attentiongraph learningfusion frameworkstructural MRIfunctional MRIMDD classification

0 comments

The pith

A dual cross-attention framework fuses structural and functional MRI scans to detect major depressive disorder by modeling their bidirectional interactions.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper presents a fusion method for structural MRI and resting-state functional MRI that uses dual cross-attention to let each modality directly attend to relevant patterns in the other. This replaces simpler approaches like joining features at the input level. A reader would care because depression involves changes that neither scan type fully captures alone, so explicit interaction modeling could yield more accurate classification from available imaging data. Tests on a large public dataset with multiple brain atlases and repeated cross-validation show the method matches or exceeds basic fusion, especially when using functional atlases. If the approach holds, it points to cross-modal attention as a practical way to combine complementary brain measurements for psychiatric classification tasks.

Core claim

The authors develop a dual cross-attention graph learning framework that explicitly models bidirectional interactions between structural MRI and resting-state functional MRI representations. On the REST-meta-MDD dataset, using both structural and functional brain atlases under 10-fold stratified cross-validation, the best configuration reaches 84.71 percent accuracy and outperforms conventional feature-level concatenation for functional atlases while remaining comparable for structural ones.

What carries the argument

Dual cross-attention graph learning framework that allows each MRI modality to attend to information from the other modality in both directions before final classification.

If this is right

The method delivers robust results across both structural and functional atlas choices.
Explicit bidirectional modeling improves results over feature concatenation specifically when functional atlases are used.
The framework supports multimodal neuroimaging classification without requiring one modality to dominate the other.
Performance metrics remain stable under stratified cross-validation on a large cohort.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The same attention structure could be tested on other psychiatric conditions that use paired structural and functional scans.
If attention maps consistently highlight the same brain connections, they might guide targeted follow-up studies of those regions.
Deployment would still require checking whether the learned interactions transfer to scanners or patient groups outside the training distribution.

Load-bearing premise

The performance gain stems mainly from the bidirectional cross-modal interactions captured by attention rather than from other modeling choices, and the pattern will hold on new data, atlases, or validation schemes.

What would settle it

Running the same 10-fold cross-validation on the identical dataset and atlases but replacing the dual cross-attention blocks with simple feature concatenation and obtaining equal or higher accuracy would show the interactions are not the main driver.

Figures

Figures reproduced from arXiv: 2604.10116 by Areej M. Alhothali, Nojod M. Alotaibi.

read the original abstract

Major depressive disorder (MDD) is a prevalent mental disorder associated with complex neurobiological changes that cannot be fully captured using a single imaging modality. The use of multimodal magnetic resonance imaging (MRI) provides a more comprehensive understanding of brain changes by combining structural and functional data. Despite this, the effective integration of these modalities remains challenging. In this study, we propose a dual cross-attention-based multimodal fusion framework that explicitly models bidirectional interactions between structural MRI (sMRI) and resting-state functional MRI (rs-fMRI) representations. The proposed approach is tested on the large-scale REST-meta-MDD dataset using both structural and functional brain atlas configurations. Numerous experiments conducted under a 10-fold stratified cross-validation demonstrated that the proposed fusion algorithm achieves robust and competitive performance across all atlas types. The proposed method consistently outperforms conventional feature-level concatenation for functional atlases, while maintaining comparable performance for structural atlases. The most effective dual cross-attention multimodal model obtained 84.71% accuracy, 86.42% sensitivity, 82.89% specificity, 84.34% precision, and 85.37% F1-score. These findings emphasize the importance of explicitly modeling cross-modal interactions for multimodal neuroimaging-based MDD classification.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

Dual cross-attention fusion beats simple concatenation on functional atlases for MDD detection but the multi-site 10-fold CV leaves the gains open to site-leakage explanations.

read the letter

The main point is that this framework applies dual cross-attention plus graph learning to fuse sMRI and rs-fMRI for MDD classification and reports better numbers than feature concatenation on functional atlases in the REST-meta-MDD dataset, reaching 84.71% accuracy. That specific bidirectional modeling is the concrete step they take beyond prior concatenation work in this area, and testing across atlas types on a large collection is a reasonable move. They also keep performance comparable on structural atlases, which shows the method is not just overfitting one data view. The abstract gives clear metrics and a straightforward claim that explicit cross-modal interaction helps, so the core idea is easy to follow. The soft spots are more noticeable. No architecture details, loss functions, ablations, or statistical tests appear in the abstract, making it hard to judge how much the graph component or the dual attention actually drives the lift. The bigger issue is the evaluation: REST-meta-MDD pools scans from many sites with scanner differences, yet the 10-fold stratified CV shows no site harmonization, site-stratified folds, or leave-one-site-out checks. In that setup the model can pick up site covariance instead of true sMRI-rs-fMRI interactions, which directly weakens the claim of robust performance from the fusion mechanism. This paper is mainly for people already working on multimodal neuroimaging for psychiatric classification who want an example of cross-attention applied to this exact pair of modalities. A reader could pull the fusion idea for their own experiments, but anyone planning to cite or extend the results would need the full methods to verify the implementation and rerun with proper site controls. It deserves a serious referee to check the missing details and test whether the reported edge survives better generalization protocols.

Referee Report

2 major / 2 minor

Summary. The manuscript proposes a dual cross-attention graph learning framework that explicitly models bidirectional interactions between sMRI and rs-fMRI representations for MDD detection. It evaluates the approach on the REST-meta-MDD dataset across structural and functional atlases using 10-fold stratified cross-validation, reporting that the best dual cross-attention model achieves 84.71% accuracy, 86.42% sensitivity, 82.89% specificity, 84.34% precision, and 85.37% F1-score while outperforming simple feature concatenation on functional atlases.

Significance. If the performance improvements can be attributed to the bidirectional cross-attention mechanism rather than dataset artifacts, the work would provide a concrete advance in multimodal neuroimaging fusion for psychiatric classification by addressing the challenge of integrating structural and functional data. The graph-based formulation and atlas-agnostic testing are positive elements that could influence subsequent studies on cross-modal interaction modeling.

major comments (2)

[Abstract] Abstract (and Methods cross-validation description): The 10-fold stratified cross-validation procedure does not incorporate site harmonization, site-stratified folds, or leave-one-site-out testing despite the multi-site composition of REST-meta-MDD. This is load-bearing for the central claim of 'robust' performance and generalization, as the dual cross-attention model could learn scanner-specific covariance patterns instead of true sMRI–rs-fMRI interactions.
[Results] Results section (performance tables and comparisons): The reported outperformance over concatenation (e.g., 84.71% accuracy) is presented without statistical significance tests, confidence intervals, or ablation studies isolating the contribution of the dual cross-attention components versus the graph learning backbone. This weakens the attribution of gains specifically to bidirectional interaction modeling.

minor comments (2)

[Abstract] Abstract: Adding the total number of subjects, number of sites, and demographic details from REST-meta-MDD would improve context for interpreting the reported metrics.
[Methods] Notation: The description of the dual cross-attention mechanism would benefit from an explicit equation or diagram showing how the bidirectional attention weights are computed and fused, to aid reproducibility.

Simulated Author's Rebuttal

2 responses · 0 unresolved

Thank you for the constructive feedback on our manuscript. We address each major comment point by point below, indicating planned revisions to strengthen the work.

read point-by-point responses

Referee: [Abstract] Abstract (and Methods cross-validation description): The 10-fold stratified cross-validation procedure does not incorporate site harmonization, site-stratified folds, or leave-one-site-out testing despite the multi-site composition of REST-meta-MDD. This is load-bearing for the central claim of 'robust' performance and generalization, as the dual cross-attention model could learn scanner-specific covariance patterns instead of true sMRI–rs-fMRI interactions.

Authors: We acknowledge that the multi-site nature of REST-meta-MDD requires explicit handling of site effects to support claims of robustness. Our current 10-fold stratified cross-validation ensures class balance but does not stratify by site or apply harmonization. In the revision, we will update the Methods to include site-stratified folds and leave-one-site-out experiments, report the corresponding metrics, and discuss potential scanner-specific patterns. If site labels permit, we will also explore harmonization (e.g., ComBat) to isolate true cross-modal interactions. revision: yes
Referee: [Results] Results section (performance tables and comparisons): The reported outperformance over concatenation (e.g., 84.71% accuracy) is presented without statistical significance tests, confidence intervals, or ablation studies isolating the contribution of the dual cross-attention components versus the graph learning backbone. This weakens the attribution of gains specifically to bidirectional interaction modeling.

Authors: We agree that statistical rigor and targeted ablations are needed to attribute gains specifically to the dual cross-attention mechanism. The revised manuscript will add paired statistical tests (e.g., t-tests or McNemar’s test across folds) comparing the proposed model to concatenation, report 95% confidence intervals for all metrics, and include ablation studies that isolate the bidirectional cross-attention components from the graph backbone. These will be presented in updated Results tables and text. revision: yes

Circularity Check

0 steps flagged

No significant circularity; empirical ML evaluation stands independently of inputs

full rationale

The paper presents an empirical machine-learning framework (dual cross-attention graph fusion of sMRI and rs-fMRI) whose central claim is improved classification accuracy on the REST-meta-MDD dataset under 10-fold stratified CV. Performance numbers are obtained by training on held-out folds and comparing against a simple concatenation baseline; no equation, parameter, or prediction is shown to reduce by construction to the input features or to a self-citation. No self-definitional loops, fitted-input-as-prediction, uniqueness theorems imported from the authors' prior work, or ansatzes smuggled via citation appear in the derivation. The reported superiority is therefore an independent empirical result, not a tautology.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract provides no explicit free parameters, axioms, or invented entities; typical deep learning hyperparameters are implicitly present but unspecified.

pith-pipeline@v0.9.0 · 5528 in / 1179 out tokens · 46540 ms · 2026-05-10T15:50:56.422508+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

45 extracted references · 3 canonical work pages · 2 internal anchors

[1]

Major depressive disorder: hypothesis, mechanism, prevention and treatment,

L. Cui, S. Li, S. Wang, X. Wu, Y . Liu, W. Yu, Y . Wang, Y . Tang, M. Xia, and B. Li, “Major depressive disorder: hypothesis, mechanism, prevention and treatment,”Signal transduction and targeted therapy, vol. 9, no. 1, p. 30, 2024

2024
[2]

Functional mri in major depressive disorder: A review of findings, limitations, and future prospects,

J. Pilmeyer, W. Huijbers, R. Lamerichs, J. F. Jansen, M. Breeuwer, and S. Zinger, “Functional mri in major depressive disorder: A review of findings, limitations, and future prospects,”Journal of Neuroimaging, vol. 32, no. 4, pp. 582–595, 2022

2022
[3]

Multimodal neuroimaging: basic concepts and classification of neuropsychiatric diseases,

E. E. Tulay, B. Metin, N. Tarhan, and M. K. Arıkan, “Multimodal neuroimaging: basic concepts and classification of neuropsychiatric diseases,”Clinical EEG and neuroscience, vol. 50, no. 1, pp. 20–33, 2019

2019
[4]

The rise and fall of mri studies in major depressive disorder,

C. Zhuo, G. Li, X. Lin, D. Jiang, Y . Xu, H. Tian, W. Wang, and X. Song, “The rise and fall of mri studies in major depressive disorder,”Translational psychiatry, vol. 9, no. 1, p. 335, 2019

2019
[5]

Brain structural and functional changes in patients with major depressive disorder: a literature review,

L. Dai, H. Zhou, X. Xu, and Z. Zuo, “Brain structural and functional changes in patients with major depressive disorder: a literature review,”PeerJ, vol. 7, p. e8170, 2019

2019
[6]

3d convolutional neural network model for detection of major depressive disorder from grey matter images,

B. AR, A. Adiga, B. Mahanand, and D. Consortium, “3d convolutional neural network model for detection of major depressive disorder from grey matter images,”Applied Sciences, vol. 15, no. 19, p. 10312, 2025

2025
[7]

Classification of major depressive disorder using an attention-guided unified deep convolutional neural network and individual structural covariance network,

J. Gao, M. Chen, D. Xiao, Y . Li, S. Zhu, Y . Li, X. Dai, F. Lu, Z. Wang, S. Cai,et al., “Classification of major depressive disorder using an attention-guided unified deep convolutional neural network and individual structural covariance network,”Cerebral Cortex, vol. 33, no. 6, pp. 2415–2425, 2023

2023
[8]

3dvit-gat: a unified atlas-based 3d vision transformer and graph learning framework for major depressive disorder detection using structural mri data,

N. M. Alotaibi, A. M. Alhothali, and M. S. Ali, “3dvit-gat: a unified atlas-based 3d vision transformer and graph learning framework for major depressive disorder detection using structural mri data,”Scientific Reports, 2026. 17 arXivTemplateA PREPRINT

2026
[9]

Diagnosis of major depressive disorder based on anomaly detection with smri gray matter slices,

Y . Xiao, X. Liu, Y . Qian, B. Huang, P. Tang, and Z. Zhang, “Diagnosis of major depressive disorder based on anomaly detection with smri gray matter slices,” in2024 IEEE Smart World Congress (SWC), pp. 1271–1277, IEEE, 2024

2024
[10]

Multi-atlas ensemble graph neural network model for major depressive disorder detection using functional mri data,

N. M. Alotaibi, A. M. Alhothali, and M. S. Ali, “Multi-atlas ensemble graph neural network model for major depressive disorder detection using functional mri data,”Frontiers in Computational Neuroscience, vol. 19, p. 1537284, 2025

2025
[11]

Depressiongraph: a two-channel graph neural network for the diagnosis of major depressive disorders using rs-fmri,

Z. Xia, Y . Fan, K. Li, Y . Wang, L. Huang, and F. Zhou, “Depressiongraph: a two-channel graph neural network for the diagnosis of major depressive disorders using rs-fmri,”Electronics, vol. 12, no. 24, p. 5040, 2023

2023
[12]

Spectral graph neural network-based multi-atlas brain network fusion for major depressive disorder diagnosis,

D.-J. Lee, D.-H. Shin, Y .-H. Son, J.-W. Han, J.-H. Oh, D.-H. Kim, J.-H. Jeong, and T.-E. Kam, “Spectral graph neural network-based multi-atlas brain network fusion for major depressive disorder diagnosis,”IEEE Journal of Biomedical and Health Informatics, vol. 28, no. 5, pp. 2967–2978, 2024

2024
[13]

Spatial-temporal data-augmentation-based functional brain network analysis for brain disorders identification,

Q. Liu, Y . Zhang, L. Guo, and Z. Wang, “Spatial-temporal data-augmentation-based functional brain network analysis for brain disorders identification,”Frontiers in Neuroscience, vol. 17, p. 1194190, 2023

2023
[14]

Advances in multimodal data fusion in neuroimaging: overview, challenges, and novel orientation,

Y .-D. Zhang, Z. Dong, S.-H. Wang, X. Yu, X. Yao, Q. Zhou, H. Hu, M. Li, C. Jiménez-Mesa, J. Ramirez,et al., “Advances in multimodal data fusion in neuroimaging: overview, challenges, and novel orientation,”Information Fusion, vol. 64, pp. 149–187, 2020

2020
[15]

Depression detection from smri and rs-fmri images using machine learning,

M. Mousavian, J. Chen, Z. Traylor, and S. Greening, “Depression detection from smri and rs-fmri images using machine learning,”Journal of Intelligent Information Systems, vol. 57, no. 2, pp. 395–418, 2021

2021
[16]

An attention-based multi-modal mri fusion model for major depressive disorder diagnosis,

G. Zheng, W. Zheng, Y . Zhang, J. Wang, M. Chen, Y . Wang, T. Cai, Z. Yao, and B. Hu, “An attention-based multi-modal mri fusion model for major depressive disorder diagnosis,”Journal of neural engineering, vol. 20, no. 6, p. 066005, 2023

2023
[17]

Cross-domain identification of multisite major depressive disorder using end-to-end brain dynamic attention network,

X. Yuan, M. Chen, P. Ding, A. Gan, A. Gong, Z. Chu, W. Nan, Y . Fu, and Y . Cheng, “Cross-domain identification of multisite major depressive disorder using end-to-end brain dynamic attention network,”IEEE Transactions on Neural Systems and Rehabilitation Engineering, vol. 32, pp. 33–42, 2023

2023
[18]

Classifying major depressive disorder using multimodal mri data: A personalized federated algorithm,

Z. Fan, J. Xu, J. Su, and D. Hu, “Classifying major depressive disorder using multimodal mri data: A personalized federated algorithm,”Brain Sciences, vol. 15, no. 10, p. 1081, 2025

2025
[19]

Adaptive multimodal neuroimage integration for major depression disorder detection,

Q. Wang, L. Li, L. Qiao, and M. Liu, “Adaptive multimodal neuroimage integration for major depression disorder detection,”Frontiers in Neuroinformatics, vol. 16, p. 856175, 2022

2022
[20]

Classification of mdd using a transformer classifier with large-scale multisite resting-state fmri data,

P. Dai, Y . Zhou, Y . Shi, D. Lu, Z. Chen, B. Zou, K. Liu, S. Liao, and R. meta MDD Consortium, “Classification of mdd using a transformer classifier with large-scale multisite resting-state fmri data,”Human brain mapping, vol. 45, no. 1, p. e26542, 2024

2024
[21]

Multivit: Multimodal vision transformer for schizophrenia prediction using structural mri and functional network connectivity data,

Y . Bi, A. Abrol, Z. Fu, and V . Calhoun, “Multivit: Multimodal vision transformer for schizophrenia prediction using structural mri and functional network connectivity data,” in2023 IEEE 20th International Symposium on Biomedical Imaging (ISBI), pp. 1–5, IEEE, 2023

2023
[22]

Ensemble of vision transformer architectures for efficient alzheimer’s disease classification,

N. Shaffi, V . Viswan, and M. Mahmud, “Ensemble of vision transformer architectures for efficient alzheimer’s disease classification,”Brain Informatics, vol. 11, no. 1, p. 25, 2024

2024
[23]

H. Ma, D. Zhang, D. Sun, H. Wang, and J. Yang, “Gray and white matter structural examination for diagnosis of major depressive disorder and subthreshold depression in adolescents and young adults: a preliminary radiomics analysis,”BMC medical imaging, vol. 22, no. 1, p. 164, 2022

2022
[24]

Mamf-gcn: Multi-scale adaptive multi-channel fusion deep graph convolutional network for predicting mental disorder,

J. Pan, H. Lin, Y . Dong, Y . Wang, and Y . Ji, “Mamf-gcn: Multi-scale adaptive multi-channel fusion deep graph convolutional network for predicting mental disorder,”Computers in biology and medicine, vol. 148, p. 105823, 2022

2022
[25]

Mmdd: A multimodal multitask dynamic disentanglement framework for robust major depressive disorder diagnosis across neuroimaging sites,

Q. Chen, P. Dai, K. Huang, T. Hu, and S. Liao, “Mmdd: A multimodal multitask dynamic disentanglement framework for robust major depressive disorder diagnosis across neuroimaging sites,”Diagnostics, vol. 15, no. 23, p. 3089, 2025

2025
[26]

The direct consortium and the rest-meta-mdd project: towards neuroimaging biomarkers of major depressive disorder,

X. Chen, B. Lu, H.-X. Li, X.-Y . Li, Y .-W. Wang, F. X. Castellanos, L.-P. Cao, N.-X. Chen, W. Chen, Y .-Q. Cheng,et al., “The direct consortium and the rest-meta-mdd project: towards neuroimaging biomarkers of major depressive disorder,”Psychoradiology, vol. 2, no. 1, pp. 32–42, 2022

2022
[27]

Reduced default mode network functional connectivity in patients with recurrent major depressive disorder,

C.-G. Yan, X. Chen, L. Li, F. X. Castellanos, T.-J. Bai, Q.-J. Bo, J. Cao, G.-M. Chen, N.-X. Chen, W. Chen,et al., “Reduced default mode network functional connectivity in patients with recurrent major depressive disorder,” Proceedings of the National Academy of Sciences, vol. 116, no. 18, pp. 9078–9083, 2019. 18 arXivTemplateA PREPRINT

2019
[28]

Integrating machining learning and multimodal neuroimaging to detect schizophrenia at the level of the individual,

D. Lei, W. H. Pinaya, J. Young, T. Van Amelsvoort, M. Marcelis, G. Donohoe, D. O. Mothersill, A. Corvin, S. Vieira, X. Huang,et al., “Integrating machining learning and multimodal neuroimaging to detect schizophrenia at the level of the individual,”Human brain mapping, vol. 41, no. 5, pp. 1119–1135, 2020

2020
[29]

Harmonization techniques for machine learning studies using multi-site functional mri data,

A. El-Gazzar, R. M. Thomas, and G. van Wingen, “Harmonization techniques for machine learning studies using multi-site functional mri data,”bioRxiv, pp. 2023–06, 2023

2023
[30]

Adjusting batch effects in microarray expression data using empirical bayes methods,

W. E. Johnson, C. Li, and A. Rabinovic, “Adjusting batch effects in microarray expression data using empirical bayes methods,”Biostatistics, vol. 8, no. 1, pp. 118–127, 2007

2007
[31]

Automated anatomical labeling of activations in spm using a macroscopic anatomical parcellation of the mni mri single-subject brain,

N. Tzourio-Mazoyer, B. Landeau, D. Papathanassiou, F. Crivello, O. Etard, N. Delcroix, B. Mazoyer, and M. Joliot, “Automated anatomical labeling of activations in spm using a macroscopic anatomical parcellation of the mni mri single-subject brain,”Neuroimage, vol. 15, no. 1, pp. 273–289, 2002

2002
[32]

Gyri of the human neocortex: an mri-based analysis of volume and variance.,

D. N. Kennedy, N. Lange, N. Makris, J. Bates, J. Meyer, and V . S. Caviness Jr, “Gyri of the human neocortex: an mri-based analysis of volume and variance.,”Cerebral Cortex (New York, NY: 1991), vol. 8, no. 4, pp. 372–384, 1998

1991
[33]

Prediction of individual brain maturity using fmri,

N. U. Dosenbach, B. Nardos, A. L. Cohen, D. A. Fair, J. D. Power, J. A. Church, S. M. Nelson, G. S. Wig, A. C. V ogel, C. N. Lessov-Schlaggar,et al., “Prediction of individual brain maturity using fmri,”Science, vol. 329, no. 5997, pp. 1358–1361, 2010

2010
[34]

A whole brain fmri atlas generated via spatially constrained spectral clustering,

R. C. Craddock, G. A. James, P. E. Holtzheimer III, X. P. Hu, and H. S. Mayberg, “A whole brain fmri atlas generated via spatially constrained spectral clustering,”Human brain mapping, vol. 33, no. 8, pp. 1914–1928, 2012

1914
[35]

An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale

A. Dosovitskiy, L. Beyer, A. Kolesnikov, D. Weissenborn, X. Zhai, T. Unterthiner, M. Dehghani, M. Minderer, G. Heigold, S. Gelly,et al., “An image is worth 16x16 words: Transformers for image recognition at scale,”arXiv preprint arXiv:2010.11929, 2020

work page internal anchor Pith review Pith/arXiv arXiv 2010
[36]

Semantics at an angle: When cosine similarity works until it doesn’t,

K. You, “Semantics at an angle: When cosine similarity works until it doesn’t,”arXiv preprint arXiv:2504.16318, 2025

work page arXiv 2025
[37]

Cosine similarity graph attention networks for autism spectrum disorder diagnosis,

Y . Zhang, Y . Fang, X. Yu, Y . Peng, and Z. Ju, “Cosine similarity graph attention networks for autism spectrum disorder diagnosis,” in2024 World Rehabilitation Robot Convention (WRRC), pp. 1–5, 2024

2024
[38]

Csepc: a deep learning framework for classifying small-sample multimodal medical image data in alzheimer’s disease,

J. Liu, X. Yu, H. Fukuyama, T. Murai, J. Wu, Q. Li, and Z. Zhang, “Csepc: a deep learning framework for classifying small-sample multimodal medical image data in alzheimer’s disease,”BMC geriatrics, vol. 25, no. 1, p. 130, 2025

2025
[39]

Gray matters: Vit-gan framework for identifying schizophrenia biomarkers linking structural mri and functional network connectivity,

Y . Bi, A. Abrol, S. Jia, J. Sui, and V . D. Calhoun, “Gray matters: Vit-gan framework for identifying schizophrenia biomarkers linking structural mri and functional network connectivity,”NeuroImage, vol. 297, p. 120674, 2024

2024
[40]

Chapter 18 - anatomical and functional brain network architecture in schizophrenia,

G. Collin and M. van den Heuvel, “Chapter 18 - anatomical and functional brain network architecture in schizophrenia,” inThe Neurobiology of Schizophrenia(T. Abel and T. Nickl-Jockschat, eds.), pp. 313–336, San Diego: Academic Press, 2016

2016
[41]

A method to assess randomness of functional connectivity matrices,

V . M. Vergara, Q. Yu, and V . D. Calhoun, “A method to assess randomness of functional connectivity matrices,” Journal of neuroscience methods, vol. 303, pp. 146–158, 2018

2018
[42]

Graph Attention Networks

P. Veliˇckovi´c, G. Cucurull, A. Casanova, A. Romero, P. Lio, and Y . Bengio, “Graph attention networks,”arXiv preprint arXiv:1710.10903, 2017

work page internal anchor Pith review arXiv 2017
[43]

Braingb: a benchmark for brain network analysis with graph neural networks,

H. Cui, W. Dai, Y . Zhu, X. Kan, A. A. C. Gu, J. Lukemire, L. Zhan, L. He, Y . Guo, and C. Yang, “Braingb: a benchmark for brain network analysis with graph neural networks,”IEEE transactions on medical imaging, vol. 42, no. 2, pp. 493–506, 2022

2022
[44]

Graph-to-text generation with bidirectional dual cross-attention and concatenation,

E. L. Jimale, W. Chen, M. A. Al-antari, Y . H. Gu, V . K. Agbesi, W. Feroze, F. Akmel, J. M. Assefa, and A. Shahzad, “Graph-to-text generation with bidirectional dual cross-attention and concatenation,”Mathematics, vol. 13, no. 6, p. 935, 2025

2025
[45]

Evolutionary neural architecture search for automated mdd diagnosis using multimodal mri imaging,

T. Li, N. Hou, J. Yu, Z. Zhao, Q. Sun, M. Chen, Z. Yao, S. Ma, J. Zhou, and B. Hu, “Evolutionary neural architecture search for automated mdd diagnosis using multimodal mri imaging,”Iscience, vol. 27, no. 10, 2024. 19

2024