High-resolution fMRI is largely hindered by random thermal noise. In this study, we propose a denoising method to reduce such noise in magnitude fMRI data. The proposed method synergistically combines: 1) variance-stabilizing transformation to convert Rician data to Gaussian-like data, 2) principle-component-analysis-based denoising algorithm with optimal singular value shrinkage to remove noise, and 3) patch-based implementation with tunable Gaussian weighting to tradeoff between functional sensitivity and specificity. Our results using synthetic and in-vivo cat task fMRI data show that the denoising method can effectively remove Rician noise, promoting functional fidelity and sensitivity in comparison to existing denoising methods.
This abstract and the presentation materials are available to members only; a login is required.