Where is physiological noise lurking in k-space?

We analyze the structure of physiological noise in the k-space of BOLD fMRI. We use DRIFTER which is an algorithm based on optimal Bayesian smoothing techniques for separation of the fMRI signal to a BOLD signal component and physiological noises. DRIFTER is run independently for each spatial frequency and it is shown that the physiological noise lies in the k-space points with low spatial frequency and that its amplitude is proportional to the BOLD signal. This result suggests that we can lower the computational burden without losing estimation accuracy by running DRIFTER only on a subset of k-space points.