Dear Sadhana,

I'm not an expert on Alphasim, and not quite sure what you are asking. Alphasim is a cluster-extent based thresholding approach to the correction of alpha levels. So it does provide you with a cluster extent (given a primary alpha level threshold) that you should use in order to "correct" for multiple comparisons. In this sense, yes the resulting cluster extent reported by Alphasim is FWE corrected. Not sure what you meant by FDR correction.

Importantly though, the smoothness going into Alphasim should be based on the actual smoothness of your data (and not the smoothing kernel size you used during smoothing preprocessing). The smoothness of your images can be seen at the bottom of the SPM results window (FWHM; it will report smoothness in both mm and voxels, but Alphasim requires mm). You should use this value in Alphasim, and not rely on the default value of Alphasim.

Also, for cluster-extent based thresholding, which primary alpha level threshold you set matters (Woo et.al:Cluster-extent based thresholding in fMRI analyses: Pitfalls and recommendations). Did you set the primary threshold at .05? If so, a more conservative level is advisable; you'll also reduce the cluster extent required to achieve a corrected p < .05.

Best,

-Lasse

Hi

I am still confused. Suppose I have used actual smoothness of my data, rmm at 5 and p-threshold of .01 default. As mask I defined the brain mask and run with 10000 iterations. This yields a minimal cluster size of 144 for a corrected alpha value of .05. It means that the minimum clusters which I get is FWE corrected (using alphasim correction). Is it??

Yes, you should then consider all clusters equal to or above 144 voxels as statsitically significant, at the .05 FWE level. All clusters with fewer than 144 voxels is thus non significant.

Thank you so much. My confusion is clear now.

Just an additional tip; the brain mask you use should only contain gray matter voxels. If you use a whole brain mask also containing white matter, your estimations in alphasim will be a little bit too conservative, as you are not really interested in white matter voxels anyway (so correcting for white matter voxels is unneccesary).

Thanks. I know this. I am using only gray matter mask only, not whole brain mask.