Dear All,
I am facing the exact same problem. I have used a smoothing of 6mm for mwp1 images and after performing TFCE (with default settings), the entire brain lights up as a single cluster (after FWE correction). Is there any suggestion on dealing with this?
Thanks and Regards
PB
On Saturday, April 3, 2021, 12:18:53 AM GMT+5:30, victor altmayer <[log in to unmask]> wrote:
Dear SPM Community,
I need some help on the use of Christian Gaser's TFCE Toolbox, that I use in SPM12 for VBM analysis
I am running correlation analyses between behavioural variables and gray matter atrophy in patients with behavioural variant fronto-temporal dementia.
After running the statistical model on the swmp1 images, that is modulated nomalized gray matter segmented images, smoothed with a 8-mm Gaussian kernel, I use the TFCE toolbox with the following parameters :
- 5000 permutations
- Permutation method : Smith
- E = 0.5
I then apply the TFCE results to my data using "TFCE results" on the TFCE GUI,
- Type of statistics : TFCE (non-parametric)
- p-value adjustment to control : FWE, p < 0.05
The problem is, whatever variable I am looking at, the results displayed always show a giant cluster of more than 50000 voxels, encompassing brain large brain regions that tend to overlap to the general atrophy pattern of the patients.
See the attached figures for a clearer overview,
Figure 1 shows results without TFCE, pFWE < 0.05,
Figure 2 results with TFCE, pFWE < 0.05.
I can manage to get more focal results by using more severe p-value thresholds, but the threshold needed is very dependent on the variable analyzed, and it seems to me that arbitrarily setting a different threshold for every variable so that it fits well the data is far from correct in terms of methodology.
Any ideas to help me with this ?
Is it possible that the atrophy of the patients analysed is too important for the use of TFCE, resulting in the clustering of all atrophied regions ?
Thanks in advance,
Victor Altmayer,
PhD Student,
Institut du Cerveau,
Paris, France
|