CoHyFuse: Condition-wise Hypergraph Fusion with Global Connectome in Task-fMRI

Boseong Kim; Haejun Chung; Ikbeom Jang

cs.LGOct 5, 2026

CoHyFuse: Condition-wise Hypergraph Fusion with Global Connectome in Task-fMRI

Authors: Boseong Kim, Haejun Chung, Ikbeom Jang

Organizations: Hanyang University, Seoul, Republic of Korea · Hankuk University of Foreign Studies, Yongin, Republic of Korea

Abstract

Task-fMRI connectomes reveal state-dependent neural reconfigurations, yet conventional methods marginalize these signals by aggregating distinct conditions into static pairwise graphs, thereby obscuring condition-specific multi-ROI organization. We introduce CoHyFuse, a condition-aware ROI-centered hypergraph framework that constructs a task-state-specific incidence matrix from condition-wise functional connectivity (FC)-profile embeddings, allowing the same ROI to form different multi-ROI hyperedges across task phases. Condition-specific neighborhood sizes $K_q$ further adapt the hyperedge scale to each task state, and the resulting condition embeddings are fused with a complementary whole-session FC branch for prediction. In the AABC cohort (N=1,074), CoHyFuse achieved the best mean out-of-fold predictive performance among evaluated baselines on FACENAME Fluid Cognition Composite (FCC) prediction (7.83 $\pm$ 0.10 MAE, 0.439 $\pm$ 0.026 $R^2$ ) and VISMOTOR age prediction (7.52 $\pm$ 0.37 MAE, 0.592 $\pm$ 0.022 $R^2$ ). In an auxiliary CMI-HBN attention-deficit/hyperactivity disorder (ADHD) classification benchmark (N=223), CoHyFuse obtained 72.0 $\pm$ 2.1% macro-AUC and 74.2 $\pm$ 2.9% accuracy. Ablation studies support the contributions of condition-wise incidence construction and dual-view fusion, suggesting that state-resolved ROI-set structure provides complementary predictive information beyond whole-session FC alone. Occlusion analysis identifies the Distraction condition as the primary driver of model prediction, pointing toward the Salience/Ventral Attention Network (SAN)--FrontoParietal Network (FPN) and within-SAN hyperedge-defined ROI-set motifs as candidate model-relevant patterns. This framework provides an interpretable, state-resolved view of the connectome for downstream cohort analysis.

Figures & tables

Fig. 1: Overview of the proposed CoHyFuse. Condition-wise ROI-centered hypergraphs are encoded by HGNN branches, fused by multi-head attention and jointly integrated with a whole-session FC branch for prediction.

Subjects (N)	Sex (M/F)	Age (Mean $\pm$ Std)	FCC (Mean $\pm$ Std)
1,074	495 / 579	$60.53\pm 15.16$	$96.41\pm 13.34$

TABLE I: Demographics of the AABC cohort.

Group	Subjects (N)	Sex (M/F)	Age (Mean $\pm$ Std)
NoDx	68	37 / 31	$8.86\pm 1.75$
ADHD-I	75	45 / 30	$9.48\pm 1.45$
ADHD-C	80	42 / 38	$8.58\pm 1.76$

TABLE II: Demographics of the CMI-HBN cohort.

Model	Year	FACENAME			VISMOTOR
Model	Year	MAE	RMSE	$R^{2}$	MAE	RMSE	$R^{2}$
SVM	1995	$8.21\!\pm\!0.15$	$10.35\!\pm\!0.32$	$0.386\!\pm\!0.036$	$8.57\!\pm\!0.54$	$10.83\!\pm\!0.64$	$0.475\!\pm\!0.045$
MLP	1986	$8.31\!\pm\!0.10$	$10.45\!\pm\!0.21$	$0.374\!\pm\!0.036$	$8.20\!\pm\!0.52$	$10.09\!\pm\!0.65$	$0.544\!\pm\!0.046$
GCN [ 7 ]	2017	$8.32\!\pm\!0.11$	$10.49\!\pm\!0.26$	$0.369\!\pm\!0.033$	$8.24\!\pm\!0.69$	$10.44\!\pm\!0.95$	$0.510\!\pm\!0.046$
GAT [ 8 ]	2018	$8.29\!\pm\!0.15$	$10.48\!\pm\!0.29$	$0.370\!\pm\!0.032$	$8.17\!\pm\!0.38$	$10.10\!\pm\!0.32$	$0.544\!\pm\!0.034$
HGNN [ 5 ]	2019	$8.25\!\pm\!0.19$	$10.46\!\pm\!0.38$	$0.373\!\pm\!0.044$	$8.05\!\pm\!0.38$	$10.09\!\pm\!0.45$	$0.544\!\pm\!0.038$

TABLE III: Regression results (mean

\pm

std) on FACENAME and VISMOTOR task-fMRI. Best and second-best are bold and underlined .

Method	MAE	$R^{2}$
w/o Whole FC	$8.20\!\pm\!0.26$	$0.375\!\pm\!0.050$
w/o Cond. branch	$8.31\!\pm\!0.10$	$0.374\!\pm\!0.036$
w/o ENC cond.	$8.10\!\pm\!0.17$	$0.397\!\pm\!0.036$
w/o DIST cond.	$8.17\!\pm\!0.25$	$0.397\!\pm\!0.031$
w/o REC cond.	$8.08\!\pm\!0.15$	$0.398\!\pm\!0.036$
Proposed	$\mathbf{7.83\!\pm\!0.10}$	$\mathbf{0.439\!\pm\!0.026}$

TABLE IV: Component ablation results on the FACENAME task.

Variant	MAE	$R^{2}$
CoHyFuse-MLP	$8.11\!\pm\!0.19$	$0.404\!\pm\!0.023$
CoHyFuse-GCN	$8.13\!\pm\!0.14$	$0.400\!\pm\!0.033$
CoHyFuse-GAT	$8.07\!\pm\!0.13$	$0.406\!\pm\!0.028$
CoHyFuse-HGNN	$\mathbf{7.83\!\pm\!0.10}$	$\mathbf{0.439\!\pm\!0.026}$

TABLE V: Within-framework control results on the FACENAME task.

$K$ Scheme	$K(E,D,R)$	MAE	$R^{2}$
Shared- $K$	$(5,\,5,\,5)$	$7.98\!\pm\!0.09$	$0.408\!\pm\!0.036$
	$(10,\,10,\,10)$	$8.03\!\pm\!0.14$	$0.405\!\pm\!0.035$
	$(50,\,50,\,50)$	$8.05\!\pm\!0.12$	$0.406\!\pm\!0.029$
Cond- $K_{q}$	$(45,\,50,\,5)$	$7.85\!\pm\!0.15$	$0.433\!\pm\!0.024$
	$\mathbf{(50,\,45,\,5)}$	$\mathbf{7.83\!\pm\!0.10}$	$\mathbf{0.439\!\pm\!0.026}$
	$(50,\,50,\,10)$	$7.84\!\pm\!0.11$	$0.432\!\pm\!0.025$

TABLE VI: Top-3

K

settings on the FACENAME task-fMRI.

Fig. 2: $K_{q}$ sensitivity heatmaps (validation MAE) on the FACENAME task-fMRI. Each panel fixes $K_{\mathrm{REC}}$ and varies $K_{\mathrm{ENC}}$ and $K_{\mathrm{DIST}}$ .

Distance	MAE	$R^{2}$
Euclidean	$\mathbf{7.83\!\pm\!0.10}$	$\mathbf{0.439\!\pm\!0.026}$
Cosine	$7.94\!\pm\!0.08$	$0.404\!\pm\!0.023$
Correlation	$7.96\!\pm\!0.09$	$0.414\!\pm\!0.021$

TABLE VII: Distance metric sensitivity on the FACENAME task.

Fig. 3: Interpretability analyses at the Yeo-7 networks: Visual ( VIN ), Somatomotor ( SMN ), Dorsal Attention ( DAN ), Salience/Ventral Attention ( SAN ), Limbic ( LIN ), FrontoParietal ( FPN ), and Default Mode ( DMN ). (a) Condition-wise top-5 hyperedge subnetworks; triangles mark anchors, and ribbons summarize dominant network pairs. (b) Condition-wise BOLD signal. (c) Subnetwork strength by FCC for each condition.

Model	mAUC (%)	ACC (%)	SEN N (%)	SEN I (%)	SEN C (%)
SVM	62.8 $\pm$ 2.5	63.6 $\pm$ 3.3	58.2 $\pm$ 5.1	62.1 $\pm$ 4.9	69.5 $\pm$ 5.3
MLP	63.9 $\pm$ 3.0	65.2 $\pm$ 3.8	61.2 $\pm$ 5.4	63.4 $\pm$ 5.1	70.2 $\pm$ 5.4
GCN [ 7 ]	60.6 $\pm$ 3.4	61.2 $\pm$ 4.2	57.5 $\pm$ 5.8	59.8 $\pm$ 5.6	65.5 $\pm$ 6.0
GAT [ 8 ]	61.2 $\pm$ 3.2	62.6 $\pm$ 4.1	59.3 $\pm$ 5.5	60.8 $\pm$ 5.4	66.9 $\pm$ 5.8
HGNN [ 5 ]	67.1 $\pm$ 2.8	68.4 $\pm$ 3.5	65.3 $\pm$ 4.7	66.7 $\pm$ 4.6	72.6 $\pm$ 5.0
HNHN [ 10 ]	66.8 $\pm$ 2.8	68.0 $\pm$ 3.6	65.1 $\pm$ 4.8	66.4 $\pm$ 4.6	71.8 $\pm$ 5.1

TABLE VIII: Performance comparison on the CMI-HBN benchmark.

Explore similar work

Aug 31, 2026eess.SP

Benchmarking External Generalization of SPD Matrix Learning for Resting-State fMRI Connectome Prediction

Resting-state functional magnetic resonance imaging (rs-fMRI) functional connectivity (FC) matrices are widely used for individual-level prediction, but strong performance within one cohort may not generalize to a new cohort. We ask whether within-dataset performance remains when the test data come from an entirely held-out rs-fMRI dataset. Each scan is represented as a regularized symmetric positive definite (SPD) correlation connectome, which allows methods to use the geometry of the SPD manifold. We introduce a reproducible age-prediction benchmark across six rs-fMRI datasets: COBRE, ADNIDOD, Cam-CAN, ABIDE, OASIS-3, and ADNI. The benchmark compares a vectorized correlation baseline, Tangent-Space Ridge, SPDNet, and split-wise Riemannian harmonization under within-dataset GroupKFold, pooled GroupKFold, and leave-one-dataset-out (LODO) evaluation. Within-dataset and pooled GroupKFold results are substantially more favorable than LODO results. When an entire dataset is held out, prediction error increases, differences among methods narrow, and performance is strongly affected by age-range mismatch and cohort heterogeneity. The benchmark provides common inputs, model settings, data splits, and analysis scripts so that future SPD matrix learning methods can be evaluated under the same external-validation protocol.

Ce Ju, Antoine Collas, Florent Bouchard +1

Inria, CEA, Universit´e Paris-Saclay, Palaiseau, France · Universit´e Paris-Saclay, CNRS, CentraleSup´elec, L2S, France

Human Connectome Project Full-Rank Correlation Matrices Vascular Age Kernel Ridge Regression Matrix

Ask PDF

May 26, 2026cs.LG

FM-fMRI: Event Conditioned Flow Matching for Rest-to-Task fMRI Time-Series Synthesis

Task-based fMRI provides a direct readout of task-evoked neural dynamics, but it is expensive and difficult to acquire at scale, motivating rest-to-task synthesis from widely available resting-state fMRI (rsfMRI). We propose FM-fMRI, an event-conditioned flow-matching model that learns a continuous-time conditional vector field to generate task ROI time series from a subject's rsfMRI and the task event information. The formulation enables fast ODE-based sampling and flexible conditioning over heterogeneous event schedules. Rather than optimizing for pointwise reconstruction, we evaluated generated signals using complementary criteria that probe temporal and spectral structure, subject and group-level connectome consistency, and distributional alignment. On the public Human Connectome Project and internal BioPoint autism cohort, FM-fMRI achieves the strongest spectral and connectivity agreement and improved distribution-level matching over conditional diffusion, generative adversarial networks (GANs), and variational autoencoders (VAEs) baselines. Furthermore, we augment the BioPoint cohort by synthesizing task-fMRI ROI time series with our method, improving downstream autism classification and demonstrating practical utility in data-limited clinical settings. The code will be available on GitHub.

Peiyu Duan, Jiyao Wang, Nicha C. Dvornek +4

Department of Biomedical Engineering, Yale University, New Haven, CT, USA · Department of Radiology & Biomedical Imaging, Yale University, New Haven, CT, USA · Department of Electrical Engineering, Yale University, New Haven, CT, USA

Functional Magnetic Resonance Imaging Time-Series Generation Neural Dynamics Autism Spectrum Disorder

Ask PDF

Sep 29, 2026cs.AI

Flattening the Connectome Spectrum: A Spectral Filter for FC Induces a Pretraining Target for fMRI Encoders

Self-supervised pretraining reshaped prediction in language and vision, and brain foundation models (BFMs) inherited its promise. Representations learned from large unlabelled corpora should capture individual functional dynamics and generalise across cohorts. However, kernel ridge regression (KRR) fitted on functional connectivity (FC) matrices still predicts individual phenotypes more accurately than any BFM we tested. In this paper, we show that KRR is weighted by the eigenvalues of the FC which are miscalibrated for phenotype prediction. We apply an efficient spectral filter to recalibrate the eigenvalues of each subject's FC matrix, enabling the model to exploit more inter-individual variance. Across the 5 datasets, 11 parcellations and 6 prediction targets we tested, we match or exceed the KRR baseline. Based on this finding, we then pretrain a small encoder model on about 4,000 hours of fMRI from 162 open datasets, whereby we align the pairwise similarities between the embeddings of recording snippets with those between the recalibrated connectomes. Our model performs on par with the best of the 6 published BFMs we tested while having an order of magnitude fewer parameters. Our encoder performs better than FC on short scans and in smaller cohorts, especially in fingerprinting. We release the pretrained model weights, the code and the pretraining data, preprocessed and parcellated.

Giovanni Marraffini, Victoria Shevchenko, Carlo Alberto Barbano +1

UNITO · MIND

Functional Connectivity Human Connectome Project Functional Magnetic Resonance Imaging Connectome Pretraining Kernel Ridge Regression+2

Ask PDF