Gissian random field for multiple correction

Submitted by JiangYuchao on

Dear experts,

I used Gissian random field (GRF) in DPABI  for multiple correction when comparing rs FC maps differences between patient groups.

However, a reviewer suggested "this method (GRF) is based upon the hypothesis that the spatial smoothness of BOLD signal in the whole brain is consistent and the distribution of the spatial autocorrelation function in accordance with specific curve. Practically, the BOLD data is hardly accord with the hypothesis".

I did spatial smoothing in preprocessing,  i want to know whether this smoothing can meet the hypothesis? or should i use another multiple correction method? However, i noted a large number of fmri studies using GRF. In addition, i wonder whether cluster-level FDR or FWE correction in SPM is based on this GRF theory?

Thank you so much for your help.



YAN Chao-Gan

Fri, 02/10/2017 - 07:43


You can cite Eklund's PNAS paper.

In that paper, the simulated data has the same smoothness as the real fMRI data. Even in such a situation, voxel-level p < 0.001 could control the FWE rate at 0.05.