Scalable Diffusion Transformer for Conditional 4D fMRI Synthesis

2025-11-27

AI Summary

Addressing the challenges of high dimensionality, heterogeneity in fMRI data, and lack of neuroscientific validation, this paper introduces the first whole-brain 4D BOLD sequence generation framework conditioned on cognitive tasks. Methodologically, it innovatively integrates a diffusion Transformer for voxel-level conditional generation, coupled with 3D VQ-GAN latent-space compression, a CNN-Transformer hybrid architecture, AdaLN-Zero normalization, and cross-attention mechanisms to enable robust, task-driven dynamic modeling. Evaluated on the Human Connectome Project (HCP) dataset, generated sequences achieve a correlation of 0.83 with ground-truth task activation maps and a representation similarity of 0.98—both metrics improve steadily with model scale and significantly outperform U-Net baselines. This work establishes the first interpretable and neuroscientifically verifiable paradigm for cognitive-response generation, bridging computational neuroscience and generative neuroimaging with a principled, biologically grounded framework.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Cognitive Modeling & Cognitive Systems: Neural Spike Coding
• Natural Language Processing: Generation

Application Category

• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Web Mining and Content Analysis: Web data generation and simulation

Abstract

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.

Problem

• Generating whole-brain 4D fMRI sequences conditioned on cognitive tasks
• Addressing high-dimensional heterogeneous BOLD dynamics across subjects
• Lacking neuroscience-grounded validation for fMRI synthesis models

Innovation

• Diffusion transformer with 3D VQ-GAN latent compression
• CNN-Transformer backbone with AdaLN-Zero conditioning
• Scalable architecture for conditional 4D fMRI synthesis