open-discussion > Randomise in DTI analysis
Showing 1-4 of 4 posts
Sep 23, 2019 12:09 PM | Beatriz Simões
Randomise in DTI analysis
Hello,
I used FSL to conduct DTI statistical analysis. In more detail, I used randomise (FSL function) to perform statistical analysis after using TBSS (tract-based spatial statistics). As output randomise produces corrected and uncorrected p-value images (_tfce-corrp.nii.gz and _tfce_p.nii.gz) using TFCE correction. After this, I extracted the significant effects using the FSL function Cluster.
I would like to extract the correspondent t-value for each one of the effects I found. However, I can't find a way to produce a tfce t-value image using randomise.
Can someone help me with this problem?
Thank you in advance.
Best regards,
Beatriz Simões
I used FSL to conduct DTI statistical analysis. In more detail, I used randomise (FSL function) to perform statistical analysis after using TBSS (tract-based spatial statistics). As output randomise produces corrected and uncorrected p-value images (_tfce-corrp.nii.gz and _tfce_p.nii.gz) using TFCE correction. After this, I extracted the significant effects using the FSL function Cluster.
I would like to extract the correspondent t-value for each one of the effects I found. However, I can't find a way to produce a tfce t-value image using randomise.
Can someone help me with this problem?
Thank you in advance.
Best regards,
Beatriz Simões
Sep 24, 2019 07:09 AM | Jens Schwarzbach - Center for Mind Brain Science, University of Trento
Randomise in DTI analysis
Dear Beatriz
The t-values remain the same. The TFCE procedure just changes which voxels’ t-values you report (compared to a different procedure for familywise error correction).
Best,
Jens
The t-values remain the same. The TFCE procedure just changes which voxels’ t-values you report (compared to a different procedure for familywise error correction).
Best,
Jens
Sep 24, 2019 11:09 AM | Beatriz Simões
RE: Randomise in DTI analysis
Dear Jens,
Thank you very much for your help.
Now it is clear that the t-values images obtained using randomise are the same irrespective of the correction perfomed on the p-value images.
Based on your answer, I have another question. To conduct some additional analysis I extracted the peak t-value (for each effect) from the t-value image using fslmeants function (fslmeants - _tstat.nii.gz -o output -c $x $y $z).
However, I noticed that there were some incongruities between the p-values and t-values. In more detail, for one of the images the smaller p-values did not correspond to larger t-values (as expected based on statistical concepts). Do you know what could be the cause for this behavior?
I was hoping this was due to differences in the t-value images, but I guess that is not the problem.
Thank you very much.
Best regards,
Beatriz Simões
Thank you very much for your help.
Now it is clear that the t-values images obtained using randomise are the same irrespective of the correction perfomed on the p-value images.
Based on your answer, I have another question. To conduct some additional analysis I extracted the peak t-value (for each effect) from the t-value image using fslmeants function (fslmeants - _tstat.nii.gz -o output -c $x $y $z).
However, I noticed that there were some incongruities between the p-values and t-values. In more detail, for one of the images the smaller p-values did not correspond to larger t-values (as expected based on statistical concepts). Do you know what could be the cause for this behavior?
I was hoping this was due to differences in the t-value images, but I guess that is not the problem.
Thank you very much.
Best regards,
Beatriz Simões
Sep 24, 2019 08:09 PM | Jens Schwarzbach - Center for Mind Brain Science, University of Trento
Randomise in DTI analysis
>
Dear Beatriz,
with TFCE correction the p value is not alone a function of t (as in univariate parametric statistics), but now also depends on the size of the cluster a given voxel is part of (within a large cluster moderate t-values can yield very low p-values; the same t-value in a smaller cluster would yield a higher p-value).
Best,
Jens
Dear Beatriz,
with TFCE correction the p value is not alone a function of t (as in univariate parametric statistics), but now also depends on the size of the cluster a given voxel is part of (within a large cluster moderate t-values can yield very low p-values; the same t-value in a smaller cluster would yield a higher p-value).
Best,
Jens