Mark,
Please see inline responses below. Sorry for the cross-posting on both lists.
On Tue, Feb 19, 2013 at 8:26 PM, Mark <[log in to unmask]> wrote:
> Hi,
>
> I don’t know where to ask this question to report uncorrected maps and I hope some experts here could help me....
>
> I have 10 subjects in my experiment. The fmri image of each subject is 64x64x35 with 3.75x3.75x4 mm voxel size. In my 1st-level analysis, images were coregistered to the corresponding T1-weighted image (image 256x256x192; voxel size 0.98 x 0.98 x 1 mm) and then normalized to the standard MNI template. The resulting resampled voxel volume of the normalized images was 2 x 2 x 2 mm. Data were then smoothed using an 8-mm FWHM. First-level t-contrasts were then entered into a second-level 1-sample t-test to produce an average group activation map. Results showed that no voxels survived corrected threshold, but I’d like to make the correction at the cluster level by 3dClustSim in AFNI.
>
> <Q1>
> To determine the minimal cluster size with a cluster threshold of 0.05, I tried individual voxel p = 0.005 using the command below:
>
> 3dClustSim -nxyz 64 64 35 -dxyz 3.75 3.75 4 -fwhm 8 -iter 10000 -pthr 0.005 –athr 0.05
>
> Am I correct? Or should I use:
>>>> No, this would be incorrect. There are a number of issues: (a) The voxel size needs to be that of the data analyzed, not that of the acquired data; (b) the fwhm should be the smoothness of the residuals of your model and not the smoothing kernel; (c) you didn't specify the connectivity criteria - what defines if a voxel is to be included in a cluster (1 2 3 for face, edge, corner connected voxels); (d) I would limit it to the voxels analyzed not the entire cube.
>
> <Q2>
> Actually we also did voxel-based morphometry analysis and thus I made an average gray matter mask (gm) of all subjects. I wanted to confine the analysis in gray matter voxels only, not the whole brain. I then did:
>
> 3dClustSim -mask gm –iter 10000 -pthr 0.005 –athr 0.05
>>> In this case, you'd need to make sure you mask your results when you apply the threshold. Additionally, you need to specify the smoothness of the data and the connectivity criteria.
>
> Is this correct? (I think maybe not because their image and voxel dimensions are different from fmri but I need your comments).
>>>> You need to run 3dClustSim for each analysis you want to correct.
Here is an example command:
3dClustSim -mask mask.hdr -fwhmxyz 8.80992 8.60894 7.73012 -niter
10000 -prefix SPMmask -NN 2 -nodec
In this case, I used the mask.hdr image from SPM analysis to limit the
number of voxels being tested. This is highly recommend as the entire
cube reduces the detection ability. Secondly, I used the FWHM from SPM
in mm to tell the program the smoothness of the data. The NN option
tells the program that a cluster is formed from face and edge
connected voxels. The -nodec option tells the program to round the
answers up to the nearest voxel.
Hopefully these answers and example will help.
>
> I do appreciate any help.
>
> Mark
|