DSFM: Dual-Spectral Flow Matching for fMRI Time Series Generation
Researchers propose Dual-Spectral Flow Matching (DSFM), a generative framework for synthesizing fMRI time series data. The method addresses the scarcity of high-fidelity fMRI samples by using wavelet decomposition and spectral flow matching to capture non-stationary, multi-scale spatiotemporal dynamics of BOLD signals. DSFM first converts BOLD signals into wavelet decomposition maps via discrete wavelet transform (DWT), then applies spectral flow matching in dual frequency representations. The approach aims to improve brain disorder identification by generating realistic fMRI data for training data-driven models. The paper is available on arXiv under ID 2605.30387.
Key facts
- DSFM stands for Dual-Spectral Flow Matching.
- It generates fMRI time series data.
- Uses wavelet decomposition via discrete wavelet transform (DWT).
- Employs spectral flow matching in dual frequency representations.
- Aims to address scarcity of high-fidelity fMRI samples.
- Targets brain disorder identification.
- Published on arXiv with ID 2605.30387.
- Focuses on non-stationary and multi-scale spatiotemporal dynamics.
Entities
Institutions
- arXiv