MAGNET: Manifold-Aware Graph Diffusion Network for Connectome Generation

Protyay Dey; Ayush Roy; Hyuna Cho; Won Hwa Kim; Vishnu Suresh Lokhande

cs.LGOct 12, 2026

MAGNET: Manifold-Aware Graph Diffusion Network for Connectome Generation

Authors: Protyay Dey, Ayush Roy, Hyuna Cho, Won Hwa Kim, Vishnu Suresh Lokhande

Organizations: University at Buffalo, SUNY, USA · National Institutes of Health, USA · POSTECH, South Korea

Abstract

Functional brain connectome represents neural connectivity as a matrix of pairwise interactions between brain regions. Generation of functional connectomes is not only a question of validity; rather, having satisfied the constraints on the correlation matrix, the next step is to recover the class-conditional geometry buried under coarse labels. We propose MAGNET, a Manifold-Aware Graph Diffusion Network which uses a normalized-Cholesky representation of the manifold of correlation matrices that guarantees validity. MAGNET lifts noisy latent states into ROI-level region tokens and performs denoising with a relational inductive bias over brain atlas regions. To deal with structural problems induced by coarse labels, MAGNET employs class-anchored conditioning, amortized structural bridge, and relevance-preserving corruption. Across ABIDE, ADNI, and OASIS-3, MAGNET consistently achieves favorable results compared to previous manifold-aware approaches, demonstrating improvements of 7-21% in class-conditional fidelity ( $α,β$ -F1) across three cohorts and better sampling efficiency. Moreover, while training only with strict binary labels, MAGNET is capable of maintaining clinical heterogeneity through fine substructure of the connectomes in a zero-shot setting, improving subclass covariance alignment ( $λ$ -MSE) by over 30%. These results suggest that geometric validity is a necessary but insufficient condition for clinical utility in connectome synthesis. Moreover, efforts in making the diffusion denoising class-conditional manifold aware finds utility beyond the highly curved brain connectome generation as this is a critical problem in various general settings. The code repository can be found at https://github.com/protyayofficial/MAGNET.

Figures & tables

Figure 1: 1D manifold projection. Density estimates along the primary manifold direction across ABIDE, ADNI, and OASIS-3. A consistent diagnostic shift along the dominant geometric axis confirms that valid manifold structure carries clinical signal. Healthy controls are anchored in green.

Figure 2: Latent structural heterogeneity. 2D projections of connectomes across ABIDE, ADNI, and OASIS-3. Key observation: The extensive vertical spread (residual variation) and severe overlap within individual clinical labels demonstrate that coarse categories collapse diverse subpopulations. This off-axis variation proves classes lack a single prototype, visually confirming the core challenge of conditional support recovery.

Figure 3: Illustration of MAGNET Architecture. The pipeline maps raw correlation matrices to validity-preserving coordinates ( $x_{t}$ ), which are tokenized and merged with additive context embeddings (time, subtype, and bridge). The Transformer core then performs graph-aware denoising, modulated by three dedicated controllers: Class-Anchored Conditioning, Amortized Structural Bridge, and Relevance-Preserving Corruption . Finally, row projection and symmetrization guarantee the predicted structure remains on the valid correlation manifold.

Dataset	Method	Quality Metrics			CAS Metrics		Time (s.)
Dataset	Method	$\alpha$ -Precision $\uparrow$	$\beta$ -Recall $\uparrow$	$\alpha,\beta$ -F1 $\uparrow$	ROC-AUC $\uparrow$	F1 $\uparrow$	Sampling $\downarrow$
ABIDE	Real Data	0.80 $\pm$ 0.04	0.81 $\pm$ 0.04	0.80 $\pm$ 0.01	0.66 $\pm$ 0.04	0.58 $\pm$ 0.06	N/A
	TriangCFM	0.21 $\pm$ 0.21	0.24 $\pm$ 0.35	0.20 $\pm$ 0.29	0.50 $\pm$ 0.05	0.40 $\pm$ 0.13	7.16 $\pm$ 1.17
	TriangDDPM	0.25 $\pm$ 0.04	0.22 $\pm$ 0.06	0.23 $\pm$ 0.04	0.52 $\pm$ 0.05	0.43 $\pm$ 0.08	5.45 $\pm$ 0.19
	DiffeoGauss	0.51 $\pm$ 0.04	0.23 $\pm$ 0.03	0.32 $\pm$ 0.03	0.66 $\pm$ 0.05	0.56 $\pm$ 0.06	0.06 $\pm$ 0.01
	DiffeoCFM	0.81 $\pm$ 0.05	0.50 $\pm$ 0.04	0.62 $\pm$ 0.05	0.60 $\pm$ 0.04	0.52 $\pm$ 0.07	9.04 $\pm$ 0.86

Table 1: Quantitative comparison of validity-preserving generative models under matched low-step budgets. We compare MAGNET against GDD-proj [ 43 ] , TriangCFM, DiffeoGauss, and DiffeoCFM [ 21 ] on ABIDE [ 30 ] , ADNI [ 83 ] , and OASIS-3 [ 49 ] . We report

\alpha

-Precision,

\beta

-Recall, and

\alpha,\beta

-F1 [ 2 ] to evaluate class-conditional support fidelity, ROC-AUC and F1 to evaluate downstream utility, and wall-clock sampling time per split. The Real Data row provides the empirical oracle from split-wise real-data comparisons, and the MAGNET row highlights our method. Bold denotes the best generative model for each metric and underlining denotes the runner-up. Supporting geometric analyses are given in Appendix G – H .

Figure 4: Spatial mismatch of cortical hubs. Cortical surface maps illustrating the structural deviation of generated samples relative to a held-out real test scan (left). Brighter regions indicate a larger spatial mismatch compared to the real reference, while darker regions signify higher fidelity. MAGNET consistently produces the darkest surfaces across all cohorts, visually confirming superior support recovery and yielding the lowest quantitative hub errors (shown below each panel).

	Method	Quality (Macro-Averaged)			CAS	Distributional Preservation $\downarrow$
	Method	$\alpha$ -Prec. $\uparrow$	$\beta$ -Recall $\uparrow$	$\alpha,\beta$ -F1 $\uparrow$	Macro F1 $\uparrow$	Class $D_{KL}$	Class $D_{B}$	$\lambda$ -MSE
ADNI	TriangCFM	$\mathbf{0.51\pm 0.11}$	$0.21\pm 0.05$	$0.25\pm 0.05$	$0.19\pm 0.04$	$0.19\pm 0.07$	$0.05\pm 0.02$	$\underline{9.16\pm 0.16}$
	DiffeoGauss	$0.00\pm 0.00$	$\underline{0.54\pm 0.04}$	$0.01\pm 0.01$	$0.26\pm 0.03$	$\underline{0.05\pm 0.06}$	$0.01\pm 0.01$	$11.37\pm 0.64$
	DiffeoCFM	$\underline{0.38\pm 0.16}$	$0.48\pm 0.06$	$\underline{0.33\pm 0.07}$	$\underline{0.27\pm 0.03}$	$0.06\pm 0.08$	$\underline{0.01\pm 0.01}$	$14.30\pm 0.89$
	GDSS-proj	$0.27\pm 0.10$	$0.30\pm 0.06$	$0.20\pm 0.04$	$0.19\pm 0.03$	$0.34\pm 0.16$	$0.09\pm 0.03$	$13.47\pm 10.59$
	MAGNET (ours)	$0.34\pm 0.07$	$\mathbf{0.56\pm 0.14}$	$\mathbf{0.35\pm 0.04}$	$\mathbf{0.28\pm 0.02}$	$\mathbf{0.05\pm 0.03}$	$\mathbf{0.01\pm 0.01}$	$\mathbf{6.08\pm 0.53}$

Table 2: Zero-shot recovery of fine-grained clinical structure. Generators are trained using only binary labels, but evaluated on resolved four-class clinical spaces: CN/SMC/MCI/AD for ADNI and CDR

0/0.5/1/2+

for OASIS-3. Real held-out samples retain their metadata subclass labels, while generated samples are mapped into the same subclass space using a classifier trained only on real training data. Macro

\alpha

-precision,

\beta

-recall, and

\alpha,\beta

-F1 measure support recovery on the resolved subclass manifolds; multiclass CAS macro-F1 measures downstream utility; class KL divergence and Bhattacharyya distance measure preservation of subclass balance; and

\lambda

-MSE measures alignment of subclass-specific covariance structure. Bold and underlining denote the best and second-best generative models for each metric.

Dataset	Backbone	Controller				Quality			Sampling (s.) $\downarrow$
Dataset	Backbone	Subtype	Bridge	Noise	Guidance	$\alpha$ -Prec. $\uparrow$	$\beta$ -Rec. $\uparrow$	$\alpha,\beta$ -F1 $\uparrow$	Sampling (s.) $\downarrow$
ABIDE	Vector	✗	✗	✗	✗	$0.52\pm 0.05$	$0.18\pm 0.03$	$0.26\pm 0.03$	$\mathbf{0.04\pm 0.00}$
	Graph	✗	✗	✗	✗	$0.70\pm 0.05$	$0.46\pm 0.06$	$0.56\pm 0.05$	$0.36\pm 0.01$
	Graph	✓	✗	✗	✗	$0.72\pm 0.05$	$0.50\pm 0.06$	$0.59\pm 0.05$	$0.39\pm 0.01$
	Graph	✗	✓	✗	✗	$0.73\pm 0.06$	$0.48\pm 0.06$	$0.58\pm 0.06$	$0.38\pm 0.01$
	Graph	✗	✗	✓	✗	$0.71\pm 0.05$	$0.50\pm 0.04$	$0.59\pm 0.03$	$0.38\pm 0.00$

Table 3: Backbone and controller ablation. All variants use the same normalized-Cholesky validity chart and a 6-step sampling budget. Controller columns denote class-anchored conditioning (Subtype), amortized bridge conditioning (Bridge), relevance-preserving noise (Noise), and diversity-preserving classifier-free guidance (Guidance). The vector row tests whether the validity chart alone is sufficient with a flat denoiser. The graph row restores the region-level graph denoiser but disables the controller. The individual controller rows activate one mechanism at a time on top of the graph denoiser. The penultimate row uses the integrated controller without diversity preserving classifier-free guidance, and the final row is full MAGNET.

Atlas	Method	$\alpha$ -Prec. $\uparrow$	$\beta$ -Rec. $\uparrow$	$\alpha,\beta$ -F1 $\uparrow$	ROC-AUC $\uparrow$	F1 $\uparrow$	Time (s) $\downarrow$
AAL116	Real Data	0.78 $\pm$ 0.06	0.77 $\pm$ 0.09	0.77 $\pm$ 0.05	0.72 $\pm$ 0.09	0.63 $\pm$ 0.07	N/A
	TriangCFM	0.34 $\pm$ 0.09	0.02 $\pm$ 0.01	0.04 $\pm$ 0.02	0.49 $\pm$ 0.05	0.42 $\pm$ 0.19	3.80 $\pm$ 0.21
	TriangDDPM	0.41 $\pm$ 0.05	0.12 $\pm$ 0.07	0.37 $\pm$ 0.08	0.53 $\pm$ 0.08	0.48 $\pm$ 0.12	3.21 $\pm$ 0.32
	DiffeoGauss	0.55 $\pm$ 0.05	0.35 $\pm$ 0.07	0.43 $\pm$ 0.05	0.53 $\pm$ 0.07	0.32 $\pm$ 0.12	1.94 $\pm$ 0.39
	DiffeoCFM	0.58 $\pm$ 0.12	0.62 $\pm$ 0.08	0.59 $\pm$ 0.05	0.65 $\pm$ 0.08	0.54 $\pm$ 0.13	4.16 $\pm$ 0.24
	GDSS-proj	0.05 $\pm$ 0.02	0.21 $\pm$ 0.01	0.08 $\pm$ 0.03	0.51 $\pm$ 0.03	0.34 $\pm$ 0.09	1.43 $\pm$ 0.12

Table 4: Generalization to higher-resolution ABIDE parcellations. We evaluate MAGNET and the competing generators on AAL116, an anatomically defined 116-region atlas, and CC200, a functionally derived parcellation with approximately 200 regions. Both settings use the same quality-controlled 831-scan ABIDE cohort and 10-split group-aware evaluation protocol. We report class-conditional support fidelity, downstream classifier-based utility, and wall-clock sampling time. Results are mean

\pm

standard deviation across splits. The Real Data row provides the empirical reference, and bold denotes the best generative model within each atlas.

Appendix figures & tables13 assets

Supplementary material from the paper’s appendix.

Appendix

Dataset	Method	Quality	CAS Metrics		Bridge Distillation Error $\downarrow$	Time (s.)
Dataset	Method	$\alpha,\beta$ - F1 $\uparrow$	ROC-AUC $\uparrow$	F1 $\uparrow$	Bridge Distillation Error $\downarrow$	Sampling $\downarrow$
ABIDE	Real Data	0.80 $\pm$ 0.01	0.66 $\pm$ 0.04	0.58 $\pm$ 0.06	N/A	N/A
	No bridge	0.58 $\pm$ 0.05	0.66 $\pm$ 0.06	0.57 $\pm$ 0.09	-	0.10 $\pm$ 0.00
	Self-conditioning	0.51 $\pm$ 0.06	0.66 $\pm$ 0.06	0.56 $\pm$ 0.11	-	3.01 $\pm$ 0.01
	Teacher bridge	0.64 $\pm$ 0.06	0.66 $\pm$ 0.06	0.57 $\pm$ 0.09	-	3.00 $\pm$ 0.01
	Amortized bridge ( $L_{bridge}$ )	0.65 $\pm$ 0.05	0.66 $\pm$ 0.05	0.57 $\pm$ 0.08	0.48 $\pm$ 0.04	0.12 $\pm$ 0.00

Appendix

Table 5: Bridge ablations across ABIDE, ADNI, and OASIS-3. We compare six structural-conditioning variants: no bridge, self-conditioned bridge, teacher bridge at inference, amortized bridge without

L_{\mathrm{bridge}}

, amortized bridge with

L_{\mathrm{bridge}}

, and the full MAGNET objective. We report

\alpha,\beta

-F1, ROC-AUC, F1, bridge distillation error where applicable, and sampling time. Lower bridge error tracks stronger structural fidelity, while removing

L_{\mathrm{bridge}}

causes severe bridge-tracking failure. Gray rows denote the real-data reference. Bold highlights the best generative model for each metric.

Dataset	Configuration	Decay ( $p$ )	Quality	CAS Metrics		Sampling Time (s) $\downarrow$
Dataset	Configuration	Decay ( $p$ )	$\alpha,\beta$ -F1 $\uparrow$	ROC-AUC $\uparrow$	F1 $\uparrow$	Sampling Time (s) $\downarrow$
ABIDE	Isotropic (Baseline)	N/A	$0.62\pm 0.03$	$0.60\pm 0.07$	$0.50\pm 0.09$	$0.36\pm 0.00$
	Low Density	1	$0.68\pm 0.03$	$0.64\pm 0.07$	$0.55\pm 0.06$	$0.44\pm 0.12$
	Adaptive (Ours)	3	$\mathbf{0.68\pm 0.03}$	$\mathbf{0.64\pm 0.07}$	$\mathbf{0.56\pm 0.06}$	$\mathbf{0.38\pm 0.02}$
	High Density	5	$0.69\pm 0.03$	$0.64\pm 0.07$	$0.55\pm 0.05$	$0.39\pm 0.03$
ADNI	Isotropic (Baseline)	N/A	$0.69\pm 0.09$	$0.52\pm 0.07$	$0.55\pm 0.05$	$0.73\pm 0.01$

Appendix

Table 6: Relevance-preserving corruption ablation across ABIDE, ADNI, and OASIS-3. We compare standard isotropic diffusion against three temporal variations of our adaptive protection schedule (Low, Adaptive, and High density decay). Performance remains stable across all adaptive regimes, demonstrating that structural recovery is invariant to the exact decay curve provided early-stage topological protection is active.

Figure 5: Classifier-free guidance scale ablation for MAGNET. We vary the guidance scale $\gamma$ from $0.5$ to $3.0$ and plot class-conditional fidelity ( $\alpha$ -Precision, $\beta$ -Recall) together with downstream utility (CAS F1) on ABIDE, ADNI, and OASIS-3. Across datasets, performance is strongest in the moderate-guidance regime, with $\gamma=1.5$ giving the most stable fidelity–utility trade-off. Larger guidance scales tend to increase precision at the expense of recall, consistent with overspecialization to a narrower region of the class manifold.

Figure 6: Efficacy of Class-Anchored Conditioning against class-only, random-subtype, and global clustering baselines. The performance gain of Class-Anchored Conditioning strictly scales with structural heterogeneity $\mathbf{H}$ .

Dataset	$\lambda_{\mathrm{spec}}$ sensitivity				$\lambda_{\mathrm{bridge}}$ sensitivity
Dataset	$\lambda_{\mathrm{spec}}$	$\alpha,\beta$ -F1 $\uparrow$	ROC-AUC $\uparrow$	F1 $\uparrow$	$\lambda_{\mathrm{bridge}}$	$\alpha,\beta$ -F1 $\uparrow$	ROC-AUC $\uparrow$	F1 $\uparrow$
ABIDE	0	0.51 $\pm$ 0.04	0.52 $\pm$ 0.06	0.43 $\pm$ 0.01	0	0.53 $\pm$ 0.03	0.56 $\pm$ 0.06	0.41 $\pm$ 0.09
	0.025	0.66 $\pm$ 0.06	0.62 $\pm$ 0.09	0.55 $\pm$ 0.03	0.025	0.66 $\pm$ 0.05	0.63 $\pm$ 0.05	0.55 $\pm$ 0.06
	0.05	0.67 $\pm$ 0.01	0.62 $\pm$ 0.06	0.55 $\pm$ 0.09	0.05	0.66 $\pm$ 0.03	0.64 $\pm$ 0.08	0.55 $\pm$ 0.05
	0.1 (default)	0.68 $\pm$ 0.03	0.64 $\pm$ 0.07	0.56 $\pm$ 0.06	0.1 (default)	0.68 $\pm$ 0.03	0.64 $\pm$ 0.07	0.56 $\pm$ 0.06
	0.2	0.67 $\pm$ 0.02	0.63 $\pm$ 0.02	0.56 $\pm$ 0.12	0.2	0.66 $\pm$ 0.03	0.64 $\pm$ 0.02	0.54 $\pm$ 0.01

Appendix

Table 7: Sensitivity to bridge-supervision and spectral-regularization weights. We vary

\lambda_{\mathrm{spec}}

and

\lambda_{\mathrm{bridge}}

independently while fixing the other coefficient at its default value of

0.1

. All remaining training and evaluation settings are unchanged. Results are mean

\pm

standard deviation across the same splits used in the main benchmark. Setting either coefficient to zero measures removal of the corresponding objective, while the nonzero values test sensitivity over a

16\times

range.

Figure 7: Global radial calibration in log-Euclidean geometry. We plot the distribution of distances from the held-out real data center for each class and dataset. A systematic left shift relative to Real (grey) indicates manifold contraction, while a right shift indicates overdispersion. MAGNET (blue) consistently demonstrates the closest alignment to the real distribution across all datasets, while baselines like TriangCFM (insets) exhibit significant overdispersion.

Data	CI	Real	MAGNET	D-CFM	D-Gauss
ABIDE	C0	4.22	4.65 (+.43)	4.76 (+.53)	5.05 (+.83)
ABIDE	C1	4.17	4.70 (+.54)	4.85 (+.68)	5.15 (+.98)
ADNI	C0	5.54	7.15 (+1.6)	7.43 (+1.9)	10.1 (+4.5)
ADNI	C1	5.62	7.80 (+2.2)	7.71 (+2.1)	10.5 (+4.9)
OASIS	C0	4.84	5.27 (+.43)	5.54 (+.70)	5.83 (+.98)
OASIS	C1	4.88	5.35 (+.48)	5.70 (+.82)	5.98 (+1.1)

Appendix

Table 8: Geometric Fidelity Analysis. Median log-Euclidean radial distances to the held-out real class center (Eq. 15 ). Red deltas (

\Delta

) indicate the shift relative to the ground-truth median. TriangCFM is excluded due to severe overdispersion (see Fig. 7 insets). D-CFM is DiffeoCFM; D-Gauss is DiffeoGauss. CI indicates the class index.

Figure 8: Held-out real tangent-manifold occupancy atlas. We project generated samples into a 2D tangent chart defined strictly by held-out real data to evaluate local support occupancy. Visual guide: Gray contours represent the true data distribution (real support, $S_{y}$ ). The colormap indicates the density difference (generated minus real), where red highlights areas of over-generation (excess mass) and blue indicates under-generation (missing mass). Takeaway: MAGNET tightly bounds its generated samples within the valid real contours, consistently achieving the highest Inside-Support Rate (ISR, Eq. 21 ). Conversely, DiffeoCFM leaks mass beyond the true boundaries, TriangCFM exhibits severe mode collapse, and DiffeoGauss hallucinates out-of-distribution samples under the high heterogeneity of the ADNI cohort.

Dataset	Class	MAGNET	D-CFM	T-CFM	D-Gauss
ABIDE	Class 0	1.000	0.879	0.008	0.996
	Class 1	1.000	0.787	0.008	0.983
	Avg.	1.000	0.833	0.008	0.990
ADNI	Class 0	0.896	0.762	0.000	0.083
	Class 1	0.821	0.721	0.000	0.029
	Avg.	0.859	0.742	0.000	0.056

Appendix

Table 9: Quantitative tangent-manifold support recovery. Inside-support rate (

\mathrm{ISR}_{m,y}\uparrow

) measures the fraction of generated samples falling within the held-out real support region (Eq. 21 ).This quantifies whether a model accurately allocates probability mass to validation-supported regions of the tangent-manifold atlas (Fig. 8 ) without relying on regions it defined itself.

Figure 9: Dynamic support alignment for healthy controls (Class 0). We track the generated probability mass over the reverse-time denoising trajectory ( $T=5\rightarrow 0$ ). Each checkpoint is projected into the fixed, real-data-defined tangent chart established in our occupancy atlas. Visual Guide: The solid gray contour demarcates the true held-out data support ( $S_{y}$ ); the colormap tracks localized density error. Takeaway: MAGNET strictly bounds its trajectory within the interior of the true held-out data contour, achieving near-perfect inside-support rates without the boundary leakage typical of transport baselines.

Figure 10: Dynamic support alignment for heterogeneous disease cohorts (Class 1). Tracking the reverse-time trajectory ( $T=5\rightarrow 0$ ) against the fixed real-data tangent chart. Visual Guide: The solid gray contour demarcates the true held-out data support ( $S_{y}$ ); the colormap tracks localized density error. Takeaway: The high structural heterogeneity of diseased subjects causes baselines like DiffeoCFM to stagnate early and bleed mass outside valid clinical regions. Conversely, MAGNET actively tightens its alignment during the critical final denoising steps, successfully navigating complex, multi-modal topologies to terminate securely within the true clinical support.

Figure 11: Held-out connectome comparisons for class $0$ . Top to bottom: ABIDE, ADNI, and OASIS-3. For each dataset, we show three held-out scans, comparing the real connectome against DiffeoGauss, TriangCFM, DiffeoCFM, and MAGNET. Each row uses the top- $45$ edges of the corresponding real scan and reuses the same real-defined color scale across methods. $\Delta$ denotes the upper-triangular MAE to the real correlation matrix, with smaller values indicating closer agreement.

Figure 12: Held-out connectome comparisons for class $1$ . Top to bottom: ABIDE, ADNI, and OASIS-3. For each dataset, we show three held-out scans, comparing the real connectome against DiffeoGauss, TriangCFM, DiffeoCFM, and MAGNET. Each row uses the top- $45$ edges of the corresponding real scan and reuses the same real-defined color scale across methods. $\Delta$ denotes the upper-triangular MAE to the real correlation matrix, with smaller values indicating closer agreement.

Functional Connectivity

Functional connectivity measures statistical relationships between brain regions to study how neural systems coordinate.

See all 60 papers

FM-fMRI: Event Conditioned Flow Matching for Rest-to-Task fMRI Time-Series Synthesis
May 26, 2026Peiyu Duan, Jiyao Wang, Nicha C. Dvornek +4Flow Matching Functional Magnetic Resonance Imaging
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
Connectome-to-Function: Conditional Generative Latent Representations for Reservoir Computing
Sep 5, 2026Zhuolin Yu, Xingyu Liu, Yuanhao Jia +4Graph Representation Learning Graph Generation
National Key Laboratory for Multimedia Information Processing, School of Computer Science, Peking University, Beijing, China · School of Electronics Engineering and Computer Science, Peking University, Beijing, China · School of Computer Science, Beijing University of Posts and Telecommunications, Beijing, China +2
BrainLinear: A Linear Model for Brain Network Analysis in Sparse Tangent Subspaces
Aug 15, 2026Sijing Wu, Dongyuan Li, Miaoting Huang +4Brain Network Analysis Functional Connectivity
Guangdong University of Foreign Studies · The University of Tokyo · RIKEN AIP +1

May 26, 2026J/K move · Enter open · S save

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

Peiyu Duan, Jiyao Wang, Nicha C. Dvornek, Junlin Yang+3

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

Flow Matching Functional Magnetic Resonance Imaging Synthetic Data Generation Magnetic Resonance Imaging Medical Image Synthesis Conditional Generative Modeling

Open paper AskPDF arXiv

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.

Abstract

Functional Connectivity

Explore similar work

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

Connectome-to-Function: Conditional Generative Latent Representations for Reservoir Computing

BrainLinear: A Linear Model for Brain Network Analysis in Sparse Tangent Subspaces