MCDA: Multi-View Graph Contrastive Learning with Dynamic Topology Augmentation for Functional MRI Analysis

Hao Zhang; Xiaoyun Liu; Shuo Huang; Yue Ma; Liang Cao; Mingxia Liu; Li Zhang

cs.CVOct 19, 2026

MCDA: Multi-View Graph Contrastive Learning with Dynamic Topology Augmentation for Functional MRI Analysis

Authors: Hao Zhang, Xiaoyun Liu, Shuo Huang, Yue Ma, Liang Cao, Mingxia Liu, Li Zhang

Organizations: College of Biomedical Engineering, Fudan University, Shanghai 200433, China · College of Information Science and Technology & Artificial Intelligence, Nanjing Forestry University, Nanjing 210037, China · Department of Psychosomatics and Psychiatry, Zhongda Hospital, School of Medicine, Jiangsu Provincial Key Laboratory of Brain Science and Medicine, Southeast University, Nanjing 210009, China · College of Clinical Medicine, Jiangsu Health Vocational College, Nanjing 210018, China ·

Abstract

Resting-state fMRI (rs-fMRI) is a vital non-invasive tool for diagnosing brain disorders and uncovering pathological mechanisms. However, existing graph contrastive learning (GCL) methods often underexploit augmentation histories across training epochs and the relationships among different augmented views. In this paper, we introduce a topology-guided Multi-view graph Contrastive learning framework with Dynamic graph topology Augmentation (MCDA) for fMRI analysis. Unlike traditional GCL that uses fixed noise or edge-dropping, MCDA employs a Dynamic Augmentation Update (DAU) module that iteratively refines the augmented topology. Specifically, we incorporate self-aware and cross-sample augmentation strategies to simultaneously capture individual-specific functional signatures and inter-subject heterogeneity. We then design a topology-guided multi-view graph contrastive learning scheme that explores interrelationships not only between the original network and its augmented views, but also across different augmented views to maximize structural information preservation. We characterize the score-level optimization directions and establish a lower bound for the contrastive objective. The framework combines classification and contrastive losses for brain disorder diagnosis and examines discriminative functional connectivity patterns as potential biomarkers. Experiments on two public rs-fMRI datasets comprising 1,236 subjects demonstrate that MCDA outperforms the state-of-the-art methods. The code has been made publicly available at: https://github.com/goodcodingagoodboy/MCDA.

Figures & tables

Fig. 1: Overview of the proposed MCDA framework for rs-fMRI analysis. (a) The framework comprises brain network construction, dynamic graph augmentation, and topology-guided multi-view graph contrastive learning. (b) Self-aware augmentation provides subject-specific modulation, while cross-sample augmentation introduces spatial and temporal perturbations from other subjects. (c) The DAU module uses a shared BiGRU to summarize cross-epoch augmentation histories and sigmoid-gated fusion to refine current augmented views using branch-specific historical references.

Fig. 2: The spatio-temporal GC Encoder consists of spatial and temporal graph convolution parts. Here, TP refers to time points, and ROI represents brain region nodes.

Fig. 3: Illustration of the DAU module. (a) Whole-graph projection and construction of a descriptor history window across training epochs. (b) Shared BiGRU encoding followed by decoding and sigmoid gate generation. (c) Gated fusion of the current augmentation with the preceding epoch’s pre-refinement graph. The self-aware and cross-sample branches share BiGRU parameters while maintaining separate histories, projections, and decoders.

Dataset	Site	Category	Number	Gender(M/F)	Age
REST-MDD	20	MDD	282	99/183	38.74 $\pm$ 13.74
	20	HC	251	87/164	39.64 $\pm$ 15.87
	21	MDD	86	38/48	34.71 $\pm$ 12.63
	21	HC	70	31/39	36.13 $\pm$ 12.64
	25	MDD	89	21/68	65.60 $\pm$ 6.75
	25	HC	63	29/34	69.63 $\pm$ 5.86

TABLE I: Demographic Information of studied subjects from the REST-MDD and ABIDE datasets. The age is reported as mean

\pm

standard deviation. M: Male; F: Female; MDD: Major depressive disorder; ASD: Autism Spectrum Disorder; HC: Healthy control.

Method	Site 20				Site 21				Site 25
Method	ACC	AUC	PRE	REC	ACC	AUC	PRE	REC	ACC	AUC	PRE	REC
BrainGNN	63.6 $\pm$ 1.7 *	65.5 $\pm$ 3.3	57.6 $\pm$ 5.9	69.8 $\pm$ 14.2	65.2 $\pm$ 6.2	67.0 $\pm$ 2.9	58.8 $\pm$ 7.2	74.2 $\pm$ 7.5	66.2 $\pm$ 8.3	64.8 $\pm$ 7.6	59.3 $\pm$ 7.2	68.0 $\pm$ 8.2
LCAF	61.6 $\pm$ 1.9 *	62.8 $\pm$ 4.0	54.3 $\pm$ 2.7	68.9 $\pm$ 7.3	61.1 $\pm$ 2.9 *	55.5 $\pm$ 6.6	55.3 $\pm$ 10.1	57.3 $\pm$ 12.7	69.0 $\pm$ 7.3	67.6 $\pm$ 3.6	63.0 $\pm$ 6.8	70.0 $\pm$ 6.1
TSDA	62.1 $\pm$ 2.4 *	60.9 $\pm$ 4.6	56.1 $\pm$ 3.9	62.9 $\pm$ 7.8	59.0 $\pm$ 2.7 *	55.0 $\pm$ 5.8	52.6 $\pm$ 7.9	62.9 $\pm$ 17.0	62.7 $\pm$ 5.7 *	54.2 $\pm$ 4.1	52.1 $\pm$ 1.1	63.8 $\pm$ 13.9
GCDA	63.5 $\pm$ 1.0 *	62.2 $\pm$ 4.0	59.4 $\pm$ 4.2	64.5 $\pm$ 14.3	67.3 $\pm$ 4.1	65.2 $\pm$ 6.5	60.0 $\pm$ 10.5	69.4 $\pm$ 7.8	69.3 $\pm$ 3.3	68.8 $\pm$ 6.3	70.7 $\pm$ 7.6	72.6 $\pm$ 9.6
UCGL	63.7 $\pm$ 2.7 *	60.7 $\pm$ 2.7	57.9 $\pm$ 2.9	66.2 $\pm$ 11.1	62.8 $\pm$ 3.3 *	61.2 $\pm$ 5.7	52.0 $\pm$ 8.3	63.4 $\pm$ 13.1	63.3 $\pm$ 1.9 *	58.0 $\pm$ 4.4	57.1 $\pm$ 9.8	67.2 $\pm$ 12.3

TABLE II: Results (%) of seven methods in MDD vs. HC classification with mean

\pm

standard deviation. Best results are shown in bold. Here, * denotes a statistically significant ( p <0.05) difference between MCDA and a competing method in terms of accuracy.

Method	Site NYU				Site UM				Site USM
Method	ACC	AUC	PRE	REC	ACC	AUC	PRE	REC	ACC	AUC	PRE	REC
BrainGNN	73.6 $\pm$ 7.8 *	70.3 $\pm$ 5.9	70.3 $\pm$ 5.8	49.9 $\pm$ 7.5	70.9 $\pm$ 6.1	70.4 $\pm$ 5.6	81.0 $\pm$ 14.9	51.3 $\pm$ 14.2	75.2 $\pm$ 9.1 *	71.1 $\pm$ 10.2	69.2 $\pm$ 13.9	84.7 $\pm$ 11.5
LCAF	71.7 $\pm$ 4.6 *	72.2 $\pm$ 4.8	77.6 $\pm$ 9.2	64.9 $\pm$ 12.4	71.2 $\pm$ 5.2	73.0 $\pm$ 5.7	72.2 $\pm$ 13.4	61.8 $\pm$ 8.9	71.5 $\pm$ 5.1 *	73.9 $\pm$ 11.2	70.6 $\pm$ 15.9	73.5 $\pm$ 12.4
TSDA	68.3 $\pm$ 7.0 *	69.3 $\pm$ 7.3	73.1 $\pm$ 9.8	66.6 $\pm$ 8.5	69.5 $\pm$ 4.2 *	70.8 $\pm$ 9.9	72.5 $\pm$ 17.7	68.8 $\pm$ 15.9	62.4 $\pm$ 4.5 *	57.4 $\pm$ 5.4	66.1 $\pm$ 6.4	69.2 $\pm$ 2.2
GCDA	76.4 $\pm$ 2.4	65.7 $\pm$ 6.0	76.2 $\pm$ 8.3	66.7 $\pm$ 15.6	71.2 $\pm$ 13.4	74.5 $\pm$ 14.8	65.3 $\pm$ 12.1	77.9 $\pm$ 16.2	77.8 $\pm$ 8.6	73.9 $\pm$ 14.1	81.7 $\pm$ 6.9	78.6 $\pm$ 15.1
UCGL	73.0 $\pm$ 6.6 *	70.7 $\pm$ 8.2	78.9 $\pm$ 8.2	59.6 $\pm$ 17.5	74.5 $\pm$ 5.2	71.8 $\pm$ 7.7	81.1 $\pm$ 6.5	58.2 $\pm$ 18.7	77.2 $\pm$ 7.5	78.6 $\pm$ 9.0	78.0 $\pm$ 9.4	78.1 $\pm$ 9.4

TABLE III: Results (%) of seven methods in ASD vs. HC classification with mean

\pm

standard deviation. Best results are shown in bold. Here, * denotes a statistically significant ( p <0.05) difference between MCDA and a competing method in terms of accuracy.

Method	Site 20				Site 21				Site 25
Method	ACC	AUC	PRE	REC	ACC	AUC	PRE	REC	ACC	AUC	PRE	REC
MCDA-I	61.5 $\pm$ 2.6 *	50.9 $\pm$ 3.2	53.8 $\pm$ 2.7	63.8 $\pm$ 8.4	63.2 $\pm$ 5.4 *	65.4 $\pm$ 2.5	54.0 $\pm$ 8.5	66.6 $\pm$ 12.3	63.3 $\pm$ 1.9 *	58.8 $\pm$ 4.6	52.7 $\pm$ 5.8	63.8 $\pm$ 11.8
MCDA-II	60.8 $\pm$ 3.6 *	63.5 $\pm$ 7.3	58.9 $\pm$ 4.5	53.8 $\pm$ 10.7	68.0 $\pm$ 4.9 *	69.3 $\pm$ 6.5	52.4 $\pm$ 11.6	60.9 $\pm$ 19.0	68.0 $\pm$ 8.1	64.1 $\pm$ 10.7	59.5 $\pm$ 18.5	54.1 $\pm$ 7.8
MCDA-III	58.2 $\pm$ 5.6 *	55.6 $\pm$ 6.5	54.3 $\pm$ 4.9	49.3 $\pm$ 8.9	65.4 $\pm$ 4.9 *	53.7 $\pm$ 12.5	57.5 $\pm$ 10.4	58.7 $\pm$ 13.1	64.4 $\pm$ 4.6 *	67.5 $\pm$ 5.4	68.5 $\pm$ 2.6	66.3 $\pm$ 3.2
MCDA	70.3 $\pm$ 2.9	74.5 $\pm$ 2.8	72.3 $\pm$ 4.0	65.6 $\pm$ 8.4	71.8 $\pm$ 6.4	70.2 $\pm$ 5.3	63.9 $\pm$ 4.8	69.5 $\pm$ 8.8	73.4 $\pm$ 3.4	73.8 $\pm$ 2.5	66.6 $\pm$ 6.5	79.4 $\pm$ 9.8

TABLE IV: Results (%) of MCDA and its three variants in MDD vs. HC classification. Best results are shown in bold. Here, * denotes a statistically significant ( p <0.05) difference between MCDA and a variant in terms of accuracy.

Method	Site NYU				Site UM				Site USM
Method	ACC	AUC	PRE	REC	ACC	AUC	PRE	REC	ACC	AUC	PRE	REC
MCDA-I	76.0 $\pm$ 7.1	77.5 $\pm$ 8.3	70.2 $\pm$ 11.9	58.0 $\pm$ 10.4	70.9 $\pm$ 7.6	73.0 $\pm$ 8.3	78.0 $\pm$ 18.0	59.4 $\pm$ 26.2	74.5 $\pm$ 2.6	76.1 $\pm$ 4.1	76.0 $\pm$ 6.9	74.7 $\pm$ 6.8
MCDA-II	73.6 $\pm$ 6.8 *	76.0 $\pm$ 7.4	81.7 $\pm$ 7.6	57.6 $\pm$ 14.2	72.7 $\pm$ 4.5	75.7 $\pm$ 7.2	81.9 $\pm$ 6.4	61.3 $\pm$ 16.3	70.2 $\pm$ 5.3 *	70.6 $\pm$ 7.9	64.9 $\pm$ 9.9	78.4 $\pm$ 9.1
MCDA-III	69.0 $\pm$ 9.0 *	68.0 $\pm$ 6.5	77.0 $\pm$ 16.3	61.5 $\pm$ 14.9	65.4 $\pm$ 3.3 *	65.4 $\pm$ 1.7	65.7 $\pm$ 7.3	69.8 $\pm$ 2.1	67.4 $\pm$ 5.2 *	66.4 $\pm$ 8.0	66.2 $\pm$ 6.3	75.3 $\pm$ 10.6
MCDA	81.1 $\pm$ 2.1	82.9 $\pm$ 2.4	76.1 $\pm$ 9.9	68.1 $\pm$ 8.0	75.0 $\pm$ 2.6	78.4 $\pm$ 7.2	87.4 $\pm$ 3.6	61.2 $\pm$ 9.6	80.2 $\pm$ 7.1	82.1 $\pm$ 5.8	77.9 $\pm$ 8.9	86.2 $\pm$ 8.4

TABLE V: Results (%) of MCDA and its three variants in ASD vs. HC classification. Best results are shown in bold. Here, * denotes a statistically significant ( p <0.05) difference between MCDA and a variant in terms of accuracy.

Fig. 4: Visualization of the top 10 ROI-pair associations identified by MCDA in two classification tasks: (a) MDD vs. HC classification (Site 20) on REST-MDD, and (b) ASD vs. HC classification (Site NYU) on ABIDE.

Fig. 5: Visualization of dynamic augmented brain networks by DAU module at every 20 epochs in (a) self-aware and (b) cross-sample augmentation.

Fig. 6: Results of MCDA-GRU, MCDA-LSTM and MCDA on Site 20 from the REST-MDD dataset.

Method	ACC	AUC	PRE	REC
MCDA-None	59.8 $\pm$ 4.3	60.0 $\pm$ 6.8	52.8 $\pm$ 8.3	53.7 $\pm$ 7.3
MCDA-Ra-GConv	64.2 $\pm$ 3.7	66.2 $\pm$ 4.2	57.1 $\pm$ 6.2	67.9 $\pm$ 5.9
MCDA-2DCNN	62.6 $\pm$ 3.0	65.0 $\pm$ 4.6	58.3 $\pm$ 4.4	66.0 $\pm$ 3.9
MCDA	70.3 $\pm$ 2.9	74.5 $\pm$ 2.8	72.3 $\pm$ 4.0	65.6 $\pm$ 8.4

TABLE VI: Results (%) of different augmentation encoders on Site 20 from the REST-MDD dataset.

Fig. 7: Accuracy and AUC of MCDA with different Topology-guided Multi-view Contrastive Loss parameters $\alpha$ and $\beta$ on Site 20 from the REST-MDD dataset.

Fig. 8: Results of MCDA with different loss parameter

\lambda

on Site 20 from the REST-MDD dataset.

$K$	ACC	AUC	PRE	REC
1	60.2 $\pm$ 3.1	56.4 $\pm$ 3.7	56.0 $\pm$ 5.7	65.5 $\pm$ 2.4
3	64.4 $\pm$ 0.8	59.2 $\pm$ 3.5	61.0 $\pm$ 3.5	61.4 $\pm$ 6.7
5	70.3 $\pm$ 2.9	74.5 $\pm$ 2.8	72.3 $\pm$ 4.0	65.6 $\pm$ 8.4
7	67.0 $\pm$ 2.9	66.6 $\pm$ 9.6	67.7 $\pm$ 6.9	62.1 $\pm$ 6.5
9	69.0 $\pm$ 3.5	68.0 $\pm$ 8.5	66.9 $\pm$ 6.5	65.8 $\pm$ 5.1

TABLE VII: Results (%) of MCDA with different history window lengths

K

on Site 20 from the REST-MDD dataset.

Batch Size	ACC	AUC	PRE	REC
30	59.1 $\pm$ 6.1	56.2 $\pm$ 4.6	53.2 $\pm$ 15.4	44.5 $\pm$ 14.7
35	61.6 $\pm$ 6.4	60.4 $\pm$ 5.3	57.3 $\pm$ 4.8	54.3 $\pm$ 8.8
40	58.9 $\pm$ 6.2	61.4 $\pm$ 10.1	55.3 $\pm$ 9.3	50.4 $\pm$ 7.8
45	70.3 $\pm$ 2.9	74.5 $\pm$ 2.8	72.3 $\pm$ 4.0	65.6 $\pm$ 8.4
50	64.2 $\pm$ 4.0	67.1 $\pm$ 7.5	52.7 $\pm$ 7.5	65.1 $\pm$ 8.2

TABLE VIII: Results (%) of MCDA with different batch sizes on Site 20 from the REST-MDD dataset.

Graph Contrastive Learning

Graph contrastive learning trains graph representations by aligning related nodes or graph views while separating structurally or semantically different examples.

See all 36 papers

Distance-Aware Joint Spatio-Temporal Graph Contrastive Learning for Major Depressive Disorder Diagnosis
May 22, 2026Muhammad Asif Hasan, Yanming Zhu, Xuefei Yin +1Functional Connectivity Depression Detection
School of Information Technology and Communication, Griffith University, Australia
Continual Learning for fMRI-Based Brain Disorder Diagnosis via Functional Connectivity Matrices Generative Replay
Apr 15, 2026Qianyu Chen, Shujian YuContinual Learning Replay-Based Continual Learning
†Nanyang Technological University · ‡VU Amsterdam · §UiT The Arctic University of Norway
Multi-Source Multi-View Graph Domain Adaptation with Hyperbolic Residual Encoding for Cross-Site MDD Identification from rs-fMRI
Jul 31, 2026Zhanpeng Zheng, Xiran Chen, Haiteng Jiang +6Multi-Source Domain Adaptation Domain Adaptation
School of Computer and Artificial Intelligence, Shandong Jianzhu University, Shandong, China. · School of Mathematics and Statistics, Chongqing Jiaotong University, Chongqing, China. · Institute of Fundamental and Frontier Sciences, University of Electronic Science and Technology of China, Chengdu, Sichuan, China. +4

May 22, 2026J/K move · Enter open · S save

Distance-Aware Joint Spatio-Temporal Graph Contrastive Learning for Major Depressive Disorder Diagnosis

Muhammad Asif Hasan, Yanming Zhu, Xuefei Yin, Alan Wee-Chung Liew

School of Information Technology and Communication, Griffith University, Australia

Functional Connectivity Depression Detection Graph Contrastive Learning Neuroimaging Spatiotemporal GNNs Dynamic Functional Connectivity

Open paper AskPDF arXiv

Major depressive disorder (MDD) is a common neuropsychiatric condition whose accurate diagnosis from resting-state functional magnetic resonance imaging (rs-fMRI) remains difficult. Dynamic functional connectivity (DFC) captures time-varying interactions among brain regions and provides rich spatio-temporal information, yet current DFC-based methods face three limitations: sliding-window Pearson correlation yields noisy estimates sensitive to window length and motion artifacts; correlation-derived node features do not fully exploit frequency-domain properties of blood-oxygen-level-dependent (BOLD) signals; and most spatio-temporal graph models handle spatial structure and temporal dynamics in separate stages, restricting their ability to represent coupled brain network evolution. To overcome these issues, we reformulate DFC learning as joint spatio-temporal graph representation learning under a Hawkes-process-inspired temporal dependency prior and propose HWSTCL, a two-stage framework built on a reliability-refined joint spatio-temporal graph with a kernel-weighted pretraining objective. Within each temporal window, BOLD signals are encoded as spectral node descriptors and functional edges are refined by an exponential distance-decay prior that down-weights less reliable long-range connections. The joint graph is then formed by linking each region to itself across future windows through a Hawkes-inspired exponential kernel, allowing spatial and temporal information to be propagated together during message passing. A kernel-weighted contrastive objective further promotes temporal consistency for each region across windows while reducing redundant similarity between different regions. Experiments on a benchmark rs-fMRI dataset show that HWSTCL outperforms recent baselines and yields coherent spatio-temporal representations for MDD diagnosis.

Abstract

Graph Contrastive Learning

Explore similar work

Distance-Aware Joint Spatio-Temporal Graph Contrastive Learning for Major Depressive Disorder Diagnosis

Continual Learning for fMRI-Based Brain Disorder Diagnosis via Functional Connectivity Matrices Generative Replay

Multi-Source Multi-View Graph Domain Adaptation with Hyperbolic Residual Encoding for Cross-Site MDD Identification from rs-fMRI