An adaptive smoothing technique for random noise suppression in fMRI data

The low signal to noise ratio (SNR) of functional magnetic resonance imaging (fMRI) data necessitates the use of efficient noise filtering techniques to denoise the data while preserving its statistical properties. We propose an adaptive spatial smoothing technique in which we perform weighted-average filtering of fMR images based on correlation of the time courses followed by the voxels. We have tested the technique on simulated fMRI-like data with different SNR values, as well as, on real fMRI data. The results show that the technique effectively filters the random noise while preserving the sharpness of the images, thus, retaining the original shapes of the active regions.