Hi Sourajit,Please see below:On 19 January 2017 at 23:22, sourajit mitra <[log in to unmask]> wrote:For GLM models; were we are testing correlation of diffusion measure say FA with age (using sex as covariant) using PALM could we get R (correlation coefficient) maps on our skeleton_mask?I have one query in TBSS runs we get corrected (1-p) and also uncorrected (1-p).Dear Anderson,Thank you very much for directing me towards PALM. Our system admin just installed PALM in our system and I am looking at the instructions from web about how to implement it.
It seems that PALM has much more advance features than what traditional TBSS could do !Yes, use the option -pearson.I get (1-p) but R will also be useful!That will help us to directly measure strength of correlation in various regions of brain, rather than extracting ROI based values and then do statistics in other software like SPSS.In fact you wouldn't have to do this anyway. The t-statistic can be converted to a correlation coefficient via:r = sign(t) * sqrt(t^2/(df+t^2))where t is the t-statistic, and df are the degrees of freedom of the model (usually the number of rows minus the number of columns in the design). The sign(t) is positive if the t-stat is positive, or negative otherwise.Once you have the t-stat, you can convert to a correlation using fslmaths.All the best,AndersonsourajitKind regardsIn a nut shell a skeleton map based measurement of correlation coefficient and its representation over skeleton is always useful is always helpful!On Mon, Jan 16, 2017 at 5:27 AM, Anderson M. Winkler <[log in to unmask]> wrote:Hi Sourajit,Please see below:On 14 January 2017 at 00:01, sourajit mitra <[log in to unmask]> wrote:Now ever when I look *_tfce_corrp_tstat1.nii.gz " (A >B) comparison after FWE I get a different cluster C3 cluster 1-p> 0.95 but not C1 and C2. C3 is in a anatomically different region than C1 and C2 with no overlap!Similarily for *_tfce_corrp_tstat5.nii.gz " (A >C) comparison after FWE I get the same C1 and C2 cluster where 1-p > 0.95.Now on doing individual pariwise t-test I get *_tfce_corrp_tstat3.nii.gz " (B >C) comparison after FWE correction I get the same C1 and C2 cluster where 1-p > 0.95In *_tfce_corrp_fstat1.nii.gz i.e FWE corrected f-test result I get two cluster C1 and C2 where 1-p > 0.95Dear Anderson,Thank you for the clarification!
I just finished the TBSS runs and here are the results plus one more question.With TFCE, the usual relationship between t and F tests is lost. A significant F-test may or may not be followed by a significant post hoc t-test.Now my question is what is the best way to represent this result. Would my result be based on tfce_corrp_fstats (corrected group f-test) or tfce_corrp_tstats (for corrected individual pairwise t- test)?I have also saved all the uncorrected-p stats map for each f and t-tests.Which clusters do I report in my results?I would use neither. Instead, I would drop the F-test, and run the pairwise comparisons in PALM with the option "-corrcon". Then the result will be corrected not only within test (across voxels), but also across contrasts (across all the pairwise t-tests).All the best,AndersonSorry for bugging you so much on this but deadline is looming over my head!Kind regardssourajitOn Fri, Jan 13, 2017 at 4:49 AM, Anderson M. Winkler <[log in to unmask]> wrote:Hi Sourajit,Even though there are 3 groups, for the F-test only 2 pairwise comparisons are needed (e.g., A>B and B>C, or A>B and A>C). The reason is that once the test "knows" the difference between A and B, and it also knows the difference between B and C, it automatically knows the difference between A and C. These two pairwise can be in any direction, and don't have to have the signs matching, as the logic remains regardless.Further, the F-test is two-tailed, so using positive or negative (or any combination thereof) will lead to the same result.All the best,AndersonOn 12 January 2017 at 15:04, sourajit mitra <[log in to unmask]> wrote:In such a case wouldn't it be better to have a design.fts likewhile parameters like Dr (A<B<C) progressive increase.In such a way that say diffusion parameter FA (A>B>C) progressive decreaseMy groups A, B and C and basically subjects cohorts belonging to 3 stages of the disease.Dear Anderson,Thank you very much for your comments.
However I have one point to clarify.
/NumWaves 6/NumContrasts 2/Matrix1 0 0 1 1 0 i.e F-test between 1st (A>B) , 4th (B>C) and 5th (A>C) t-tests0 1 1 0 0 1 i.e F-test between 2nd (A<B) , 3rd (B<C) and 6th (A<C) t-testsrather than having/NumWaves 6/NumContrasts 1/Matrix1 0 1 0 0 0 F-test between 1st (A>B) and 3rd (B<C) t-tests.This is off-course with reference to my previous design.con file
EV1(A) EV2(B) EV3(C) CV1 CV21 -1 0 0 0 (A>B)-1 1 0 0 0 (A<B)0 -1 1 0 0(B<C)0 1 -1 0 0(B>C)1 0 -1 0 0(A>C)-1 0 1 0 0(A<C)It would be very helpful if you could give your feedback which design.fts should I use?Kind regardssourajit
On Thu, Jan 12, 2017 at 5:01 AM, Anderson M. Winkler <[log in to unmask]> wrote:Ho Sourajit,Please see below:On 11 January 2017 at 16:01, Sourajit Mitra Mustafi <[log in to unmask]> wrote:Dear FSL users,
I have three groups A, B and C; there are also two co variants (CV1 and CV2). At first step I did pairwise t-test between the groups in TBSS runs by using this contrast file
design.con
EV1(A) EV2(B) EV3(C) CV1 CV2
1 -1 0 0 0
-1 1 0 0 0
0 -1 1 0 0
0 1 -1 0 0
1 0 -1 0 0
-1 0 1 0 0
Now I want to perform F-test followed by pairwise t-test.
How do I construct my fts file like "design.fts" and contrast file like "design.con" to achieve this ????The design.fts would contain:/NumWaves 6/NumContrasts 1/Matrix1 0 1 0 0 0This would indicate that the first and third t-tests would constitute the F-test.
Which statistical procedure is more robust just doing individual t-test as I did eariler or doing F-test followed by pairwise t-test?For only 3 groups it's ok to do the F-test, then the paired t-tests. If more than 2, use the option -corrcon in PALM.Hope this helps!All the best,Anderson
It would be nice if I could get some feedback on this issue