Reproduce “T2I-Diff: fMRI Signal Generation via Time-Frequency Image Transform and Classifier-Free Denoising Diffusion Models” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Hwa Hui Tew, Junn Yong Loo, Yee-Fan Tan, Xinyu Tang, Hernando Ombao, Fuad Noman, Raphael C.-W. Phan, Chee-Ming Ting
T2I-Diff generates realistic fMRI BOLD signals by transforming them into time-frequency images for a classifier-free denoising diffusion model. This method effectively augments fMRI datasets, leading to improved Major Depressive Disorder classification accuracy and producing synthetic data with biologically plausible functional connectivity patterns.
Key insights
- Representing fMRI BOLD signals as time-frequency images allows denoising diffusion models to effectively capture and generate complex, non-stationary temporal and spectral characteristics.
- Denoising Diffusion Probabilistic Models (specifically EDM) offer a stable and robust framework for synthesizing high-dimensional, intricate biological time-series data, overcoming limitations of prior generative models.
- Classifier-free guidance streamlines the generation of class-conditioned synthetic data, enabling targeted data augmentation for specific neurological conditions without additional classifier training.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.