Print

Print
Hi Chao-Gan,

Please see below:

On 7 February 2017 at 21:52, YAN Chao-Gan <[log in to unmask]> wrote:
Thanks, Anderson!

OK. I thought through the null distribution of the maximum cluster size, we can define a cluster size threshold at p < 0.05. Anyway, _clustere_tstat_fwep.nii can work by its own.

I thought you meant minimum cluster size as threshold (i.e., a cluster with, say, at least 10 voxels). That isn't available, but if it were, it wouldn't make any difference on the p-values.

At any rate, in the permutation case, the p-value doesn't directly map to a test statistic that could be used as threshold to the statistical map. Often 2.3 or 3.1 are used but these are just rough heuristics. To have a p-value threshold we'd have to either run the permutation test twice, or store and revisit all the permutations, both of which are impractical and with little or no benefit.

 

Another question regards to the option "-twotail".

For _vox_tstat_fwep.nii, the null distribution is built by the maximum abs(T) value across all the voxels for each permutation?

Yes.
 

For _clustere_tstat_fwep, the null distribution is built by the cluster size defined at abs(Z) > 3.1 for each permutation?

Yes, the maximum cluster size at abs(Z) > 3.1 (or whatever other threshold).

All the best,

Anderson

 

Thanks,

Chao-Gan



On Tue, Feb 7, 2017 at 7:24 PM, Anderson M. Winkler <[log in to unmask]> wrote:
Hi Gao-Chan,

There is no explicit cluster extent threshold in either PALM or randomise, but even if there were one, I think it wouldn't change the corrected p-values in the permutation case (smaller clusters would be removed, but these don't contribute to the distribution of the maximum anyway; and for the rare cases in which a small cluster would be the maximum of a particular permutation, it would still be smaller than the clusters of interest that are above the threshold, thus with no influence on their p-values).

All the best,

Anderson


On 6 February 2017 at 22:46, YAN Chao-Gan <[log in to unmask]> wrote:
Dear Anderson,

PALM outputs a NIfTI file of _clustere_tstat_fwep for cluster extent after FWE correction. We have p values for each cluster.

However, can we find anywhere the cluster size threshold for FWE p < 0.05? Thus we can compare the cluster size threshold with GRF correction.

Thanks,

Chao-Gan


On Fri, Dec 9, 2016 at 7:11 PM, Anderson M. Winkler <[log in to unmask]> wrote:
Hi Chao-Gan,

Please, see below:


On 9 December 2016 at 03:50, YAN Chao-Gan <[log in to unmask]> wrote:
Dear Anderson and PALM experts,

After reading more materials and the code, I am revising my questions as below. Are they correct?

1. _vox_tstat.nii is the T value of a voxel.

Yes.

 
    _vox_tstat_uncp.nii is the p value corresponds to the rank of the observed T value within the permutations FOR A GIVEN VOXEL (the null distribution is the permuted T values of that given voxel).

Yes. Computing the rank is one of the ways in which the p-value can be obtained (it's then divided by the number of permutations). For the uncorrected, PALM increments a counter at every permutation, though.

 
    _vox_tstat_fwep.nii is the p value corresponds to the rank of the observed T value within the permutations of maximum T values across all the voxels (the null distribution is composed by the maximum T value across all the voxels for each permutation)  

Yes. For the corrected, the distribution of the maximum is used as reference, and the rank (or quantile) of a given voxel in relation to that distribution is used to obtain p-values. In PALM this is done with the function palm_datapval.m, which on its turn uses palm_competitive.m, which is described in this blog post: https://brainder.org/2012/11/28/competition-ranking-and-empirical-distributions/
 

2. _clustere_tstat_fwep.nii: is this the p value of the rank of a given cluster size within the permutations? How was the _clustere_tstat.nii calculated? Is it inversely calculated from the _clustere_tstat_fwep values?

The clustere_tstat.nii is simply the size (in voxels) of the cluster. This number acts as the test statistic. The corrected p-values are computed in the same way as above, i.e., using the distribution of the maximum cluster size.
 

3. The TFCE maps are similar to Point 1.

Works in the exact same way as for the t-stat above.

All the best,

Anderson


 

Thank you very much!

Best,

Chao-Gan

On Fri, Dec 9, 2016 at 5:48 AM, YAN Chao-Gan <[log in to unmask]> wrote:
Dear Anderson and PALM experts,

Recently I am using the PALM package. I have several questions regarding to the output of PALM, but I am not sure if I understand the manual correctly.

1. _vox_tstat.nii is the T value of a voxel. _vox_tstat_uncp.nii is the p value corresponds to the rank of the observed T value within the permutations. How was the _vox_tstat_fwep.nii calculated? Is this Bonferroni corrected across voxels (e.g., uncp divided by the number of voxels)?  

2. _clustere_tstat_fwep.nii: is this the p value of the rank of a given cluster size within the permutations? How was the _clustere_tstat.nii calculated? Is it inversely calculated from the _clustere_tstat_fwep values?

3. The TFCE maps are similar to Point 1?  _tfce_tstat.nii is the TFCE value. _tfce_tstat_uncp.nii is the p value corresponds to the rank of the observed TFCE value within the permutations. How was the _tfce_tstat_fwep.nii calculated?

Thank you very much!

Best,

Chao-Gan

--
Chao-Gan YAN, Ph.D.
Principal Investigator
Deputy Director, Magnetic Resonance Imaging Research Center
Institute of Psychology, Chinese Academy of Sciences
16 Lincui Road, Chaoyang District, Beijing 100101, China
-



--
Chao-Gan YAN, Ph.D.
Principal Investigator
Deputy Director, Magnetic Resonance Imaging Research Center
Institute of Psychology, Chinese Academy of Sciences
16 Lincui Road, Chaoyang District, Beijing 100101, China
-




--
Chao-Gan YAN, Ph.D.
Principal Investigator
Deputy Director, Magnetic Resonance Imaging Research Center
Institute of Psychology, Chinese Academy of Sciences
16 Lincui Road, Chaoyang District, Beijing 100101, China
-




--
Chao-Gan YAN, Ph.D.
Principal Investigator
Deputy Director, Magnetic Resonance Imaging Research Center
Institute of Psychology, Chinese Academy of Sciences
16 Lincui Road, Chaoyang District, Beijing 100101, China
-