Scalable Diffusion Transformer for Conditional 4D fMRI Synthesis

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Seo, Jungwoo, Park, David Keetae, Yoo, Shinjae, Cha, Jiook
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866917111131013120
author Seo, Jungwoo
Park, David Keetae
Yoo, Shinjae
Cha, Jiook
author_facet Seo, Jungwoo
Park, David Keetae
Yoo, Shinjae
Cha, Jiook
contents Generating whole-brain 4D fMRI sequences conditioned on cognitive tasks remains challenging due to the high-dimensional, heterogeneous BOLD dynamics across subjects/acquisitions and the lack of neuroscience-grounded validation. We introduce the first diffusion transformer for voxelwise 4D fMRI conditional generation, combining 3D VQ-GAN latent compression with a CNN-Transformer backbone and strong task conditioning via AdaLN-Zero and cross-attention. On HCP task fMRI, our model reproduces task-evoked activation maps, preserves the inter-task representational structure observed in real data (RSA), achieves perfect condition specificity, and aligns ROI time-courses with canonical hemodynamic responses. Performance improves predictably with scale, reaching task-evoked map correlation of 0.83 and RSA of 0.98, consistently surpassing a U-Net baseline on all metrics. By coupling latent diffusion with a scalable backbone and strong conditioning, this work establishes a practical path to conditional 4D fMRI synthesis, paving the way for future applications such as virtual experiments, cross-site harmonization, and principled augmentation for downstream neuroimaging models.
format Preprint
id arxiv_https___arxiv_org_abs_2511_22870
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Scalable Diffusion Transformer for Conditional 4D fMRI Synthesis
Seo, Jungwoo
Park, David Keetae
Yoo, Shinjae
Cha, Jiook
Computer Vision and Pattern Recognition
Neurons and Cognition
Generating whole-brain 4D fMRI sequences conditioned on cognitive tasks remains challenging due to the high-dimensional, heterogeneous BOLD dynamics across subjects/acquisitions and the lack of neuroscience-grounded validation. We introduce the first diffusion transformer for voxelwise 4D fMRI conditional generation, combining 3D VQ-GAN latent compression with a CNN-Transformer backbone and strong task conditioning via AdaLN-Zero and cross-attention. On HCP task fMRI, our model reproduces task-evoked activation maps, preserves the inter-task representational structure observed in real data (RSA), achieves perfect condition specificity, and aligns ROI time-courses with canonical hemodynamic responses. Performance improves predictably with scale, reaching task-evoked map correlation of 0.83 and RSA of 0.98, consistently surpassing a U-Net baseline on all metrics. By coupling latent diffusion with a scalable backbone and strong conditioning, this work establishes a practical path to conditional 4D fMRI synthesis, paving the way for future applications such as virtual experiments, cross-site harmonization, and principled augmentation for downstream neuroimaging models.
title Scalable Diffusion Transformer for Conditional 4D fMRI Synthesis
topic Computer Vision and Pattern Recognition
Neurons and Cognition
url https://arxiv.org/abs/2511.22870