Thank you Andeson,
My question is, just, to correct my understanding and I highly apprentice your input!
Can FEAT with the option OLS be used to study the difference between two groups for non-fMRI data? For example if I have PET images for two groups and I want to study the difference in PET signal between the two groups. Can I feed every PET images into FEAT as a cope and the binarized PET image as vacope then run the analysis using FEAT and the option OLS?
I used to use PAM or randomise to to study the difference between the groups. Although these methods are conservative, but I feel more confident with its results (specially with TFCE method). Honestly, I heard that some people are using FEAT/OLS to study the difference between groups for non-fMRI data sets (e.g PET, DTI ,etc) instead of Randomise/PALM, and I am wondering if this is a correct procedure?!
I highly appreciate your input!
John
Hi John,
OLS is a method to estimate parameters. FEAT doesn't do permutation and inference is done using assumptions for these parameters (normality, independence, identical distributions, plus others for the random field theory).
All the best,
Anderson
On 7 February 2017 at 10:53, John anderson <[log in to unmask]> wrote:
Dear Anderson,
Thank you very much for answering my question.
Kindly I have one more question, and I highly appreciate your input!
What is the difference between the permutation analysis that PALM is doing, and the permutation analysis in FEAT with the option OLS? Are the algorithms different ?
Thank you for any comment!
John
Hi John,
The results may differ: To accommodate additional test statistics (including NPC and MANOVA), it is more convenient to work on a common scale, which PALM does internally via conversion to a z-score before computing these. Randomise uses instead the t-stat directly. Additionally, PALM uses a fixed step (the "dh") for the calculation of TFCE, whereas randomise uses 1/100th of the values of the first permutation. Hence the differences, which should generally be small. The test is valid in both cases.
The above is described in the manual, please see here: https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/PALM/FAQ#Why_the_TFCE_results_aren.27t_identical_to_the_ones_produced_by_randomise.3F
All the best,
Anderson
On 5 February 2017 at 21:46, John anderson <[log in to unmask]> wrote:
Dear FSL experts,
I have two groups of subjects. The subjects have PET images. In order to study the difference between the groups I merged the data as a 4D file then I ran Randomise (TFCE and 5000 permutations. I got small significant difference between the groups (i.e. the area of the difference is small).
I repeated the analysis using PALM and the same parameters (TFCE and 5000 permutations). I got the same difference but the area of difference is larger.
My questions is:
Is this difference in the area of the significant difference between the groups is related to the fact that PALM is using parametric and non-parametric stats? if not how can this happen?
Thanks!
John
|