Singular-value truncation techniques have shown promise for reducing thermal noise in fMRI, where singular-values below a certain threshold are assumed to be noise and are discarded. However, this approach could lead to suboptimal signal recovery, since the remaining singular-values could still have variance contributed by noise. Here instead we propose to use a theoretically MSE-optimal function to shrink the remaining singular-values. The proposed method is evaluated using simulations and high-resolution in-vivo human brain data, and is shown to improve signal-to-noise ratio and functional statistics while leaving the spatial precision intact.
This abstract and the presentation materials are available to members only; a login is required.