T2I-Diff: fMRI Signal Generation via Time-Frequency Image Transform and Classifier-Free Denoising Diffusion Models

2025-09-25

AI Summary

High-cost fMRI acquisition and limited sample availability hinder data-driven brain analysis; existing generative models struggle to capture the inherent non-stationarity and nonlinear dynamics of BOLD signals. To address this, we propose a time-frequency-aware diffusion generative framework: (1) raw BOLD time series are transformed into time-frequency spectrograms via sliding-window Fourier transform to explicitly encode time-varying spectral characteristics; (2) a classifier-free guided denoising diffusion probabilistic model generates high-fidelity spectrograms; and (3) inverse Fourier transform reconstructs realistic time-domain fMRI signals. Our method preserves time-frequency structural consistency while significantly improving generated data fidelity and functional utility. In multi-site brain network classification tasks, synthetic data augmenting training boosts downstream model accuracy by 5.2% on average and markedly enhances generalization—demonstrating its efficacy and practical potential as a low-cost, high-quality data augmentation tool.

Technology Category

• Natural Language Processing: Generation
• Machine Learning: Deep Generative Models & Autoencoders
• Computer Vision: Diffusion Models for Vision

Application Category

• Social Networks and Social Media: Generative AI / large language models and their impact on social systems
• Web Mining and Content Analysis: Web data generation and simulation
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Functional Magnetic Resonance Imaging (fMRI) is an advanced neuroimaging method that enables in-depth analysis of brain activity by measuring dynamic changes in the blood oxygenation level-dependent (BOLD) signals. However, the resource-intensive nature of fMRI data acquisition limits the availability of high-fidelity samples required for data-driven brain analysis models. While modern generative models can synthesize fMRI data, they often underperform because they overlook the complex non-stationarity and nonlinear BOLD dynamics. To address these challenges, we introduce T2I-Diff, an fMRI generation framework that leverages time-frequency representation of BOLD signals and classifier-free denoising diffusion. Specifically, our framework first converts BOLD signals into windowed spectrograms via a time-dependent Fourier transform, capturing both the underlying temporal dynamics and spectral evolution. Subsequently, a classifier-free diffusion model is trained to generate class-conditioned frequency spectrograms, which are then reverted to BOLD signals via inverse Fourier transforms. Finally, we validate the efficacy of our approach by demonstrating improved accuracy and generalization in downstream fMRI-based brain network classification.

Problem

• Generating high-fidelity fMRI data from limited samples
• Capturing complex non-stationary and nonlinear BOLD signal dynamics
• Improving accuracy for downstream brain network classification tasks

Innovation

• Time-frequency transform captures BOLD dynamics
• Classifier-free diffusion generates conditioned spectrograms
• Inverse Fourier transform reconstructs fMRI signals