Print

Print
Thank you again for your responses- I will change my analysis accordingly but I had some questions for clarification:

1a) For "group effect" I meant testing whether there is group A > group B or group A < group B. How should I approach this? I was thinking of having two between-subjects models such that 1 model would compare group A condition 1 against all conditions of group B and another model comparing group A condition 2 against all conditions of group B. 

So in that case would the command line look like this: palm -i A1.<ext> -i B1.<ext> -i B2.<ext> -i B3.<ext> -d design_a1vsB.csv -t contrasts_a1vsB.csv -ise -n 5000 -logp [blah]

Then design matrix would look like the following:
1 0 0 0 0 0 0  1  1  1 
1 0 0 0 0 0 0  -1  0  0
1 0 0 0 0 0 0  0  -1  0
1 0 0 0 0 0 0  0  0  -1
0 1 0 0 0 0 0  1  1  1 
0 1 0 0 0 0 0  -1  0  0
0 1 0 0 0 0 0  0  -1  0
0 1 0 0 0 0 0  0  0  -1
The last three columns would represent- A1-vs-B1, A1-vs-B2, and A1-vs-B3. 

1b) For the one-sample t-tests command line (in this case, for Group A) you suggested- I just wanted to double-check whether I got the data correctly prepared:

Command line: palm -I A1.<ext> -i A2.<ext> -d design_a1.csv -t contrasts_a1.csv -ise -n 5000 -logp [blah]

For input A1.<ext> and A2.<ext> would they look like the following:

A1.<ext> & A2.<ext>
subject1 bh.thickness file for condition A1/A2
subject2 bh.thickness file for condition A1/A2
subject3 bh.thickness file for condition A1/A2
...and so on

Then for design_a1.csv, the file would look like this:
           
1 0 0 0 0 0 0 -1  1
1 0 0 0 0 0 0  1  -1
0 1 0 0 0 0 0 -1  1
0 1 0 0 0 0 0  1  -1
There would N number of columns corresponding to the number of subjects times 2 (for condition A1 and condition A2) and the last two columns would represent conditions A1 and A2. 

And contrasts_A1.csv would look like this:
0 0 0 0 0 0 0  1  0
0 0 0 0 0 0 0  0  1


Thank you very much for all the information. And I apologize again for the long email. 

Best,
Julia Shin



On Wed, Aug 10, 2022 at 1:50 PM Matthew Webster <[log in to unmask]> wrote:
Dear Julia,
                 If I understand your analysis correctly, you have two groups, each having a different number of conditions, and for each group you want to test if each of its conditions.display a positive or negative effect.

The simplest approach here is to split your data by both group and condition and run a one-sample t-test ( similar to design_a1 from Anderson’s reply in the linked thread ), the command line would then be ( e.g. for group A )

palm -I A1.<ext> -i A2.<ext> -d design_a1.csv -t contrasts_a1.csv -ise -n 5000 -logp [blah]

With contrasts_a1.csv defining both positive and negative contrasts.

Add -corrmod to correct across conditions, -corrcon to correct across contrasts.

I’m not sure how you’re defining “group effect” here, e.g. the average across all conditions?

Note NPC may not explicitly be needed in this case ( Anderson would be able to clarify ).

2) For the second question, the same logic will apply ( it’s just a change in storage format )

Hope this helps,
Kind Regards
Matthew
--------------------------------
Dr Matthew Webster
FMRIB Centre
John Radcliffe Hospital design
University of Oxford

> On 5 Aug 2022, at 08:37, Hye Min Shin <[log in to unmask]> wrote:
>
> Thank you for the response Matthew.
>
> So if I still want to NPC by still correct over both hemispheres, can I merge lh and rh into bh using palm_hemimerge and then split them later using palm_hemisplit as discussed in this thread: https://jiscmail.ac.uk/cgi-bin/webadmin?A2=FSL;42baf4dd.1608
>
> I also have a couple more questions if that is ok:
>
> 1) After going through the repeated measures anova tutorial, I wanted to try to apply PALM NPC in my analysis. I have 2 groups with group A having 2 conditions (A1 and A2) and another group B with 3 conditions (B1,B2,B3). I ran all 5 conditions on each subject. At first I just wanted to analyze them by having 10 t-tests and 2 F-contrasts (looking for the effect of group A and the effect of group B) then correcting for multiple comparisons using PALM/NPC. Would this be a good choice for my analysis?
>
> I also considered using NPC to combine within-subject and between-subject models after reading this thread: https://jiscmail.ac.uk/cgi-bin/wa-jisc.exe?A2=ind1801&L=FSL&P=R60203&K=2
>
> I thought this would be more appropriate for my analysis but with how my conditions are set up, it looks unfeasible since I do not have blocks with the same size for whole-block EB.
>
> My initial idea for within block and whole block EB is the following:
> 1 -subject 1 A1
> 1 -subject 1 A2
> 1 -subject 2 A1
> 1 -subject 2 A2
> 2 -subject 1 B1
> 2 -subject 1 B2
> 2 -subject 1 B3
> 2 -subject 2 B1
> 2 -subject 2 B2
> 2 -subject 2 B3
>
> So would it be better to just stick with 10 t-tests and 2 F-tests using NPC?
>
>
> 2) If I can go ahead with this joint NPC analysis, can I also use this on subcortical data?
>
> Sorry for these additional questions- your responses were very helpful to understand PALM/NPC and how I should analyze my study.
>
> Thank you in advance,
> Julia Shin
>
> On Tue, Aug 2, 2022 at 3:37 AM Matthew Webster <[log in to unmask]> wrote:
> Hello Julia,
> Note NPC should not be used here as it requires all inputs to be aligned - and by definition the left and right hemispheres are un-aligned. The -corrcon and -corrmod options can still be used to correct over both hemispheres ( “modalities” ) and contrasts though.
>
> Kind Regards
> Matthew
> --------------------------------
> Dr Matthew Webster
> FMRIB Centre
> John Radcliffe Hospital
> University of Oxford
>
>> On 29 Jul 2022, at 14:00, Hye Min Shin <[log in to unmask]> wrote:
>>
>> No problem- I was able to successfully run the F-test using PALM with npc.
>>
>> So if I also wanted to correct for both right and left hemispheres while applying npc to correct for contrasts can I add "-corrmod" and "-npccon" to my palm command? For a repeated measures design like the one is FreeSurfer tutorial, would it be ok to only apply npc to correct for contrasts? Or do I also have to add "-corrcon" to correct for contrasts AND add "-npccon".
>>
>> The command would look like the following (as also noted in this tread- https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=FSL;42baf4dd.1608):
>>
>> palm -i lh.thickness.sm15.mgh -i rh.thickness.sm15.mgh -m lh/mask.mgh -m rh/mask.mgh -d design.mat -t design.con -f design.fts
>>           -n 5000 -T -npc -corrmod -npccon -logp -approx tail -nouncorrected
>>
>> Best,
>> Julia Shin
>>
>> On Wed, Jul 27, 2022 at 9:28 PM Matthew Webster <[log in to unmask]> wrote:
>> Hello Julia,
>> Ah - my mistake! The F-test is on the _first_ two t-contrasts so the F-contrast should be 1 1 0 0 as you say.
>>
>> Kind Regards
>> Matthew
>> --------------------------------
>> Dr Matthew Webster
>> FMRIB Centre
>> John Radcliffe Hospital
>> University of Oxford
>>
>>> On 27 Jul 2022, at 13:18, Hye Min Shin <[log in to unmask]> wrote:
>>>
>>> I apologize for sending this email again- I originally responded to Matthew's email from JISCM@il site and since I can't see the message thread I am posting my response through my school webmail.
>>>
>>> Thank you very much for your reply Matthew- I am currently trying to run PALM again with your suggested F-test matrix of 0 0 1 1. My question is what does the F-test matrix 0 0 1 1 mean here. I thought the columns of the F-test matrix would correspond to the rows of the t-test matrix. So if t-test matrix has row 1 with tp1-vs-tp2, row 2 with tp1-vs-tp3 then the F-test matrix would be 1 1 0 0 so that row 3 and row 4 of the t-test matrix which are tp2-vs-tp3 and subject mean would not be chosen. Also in my palm command I gave the option -cmcx and I was wondering if this would be an appropriate flag to use instead of -npccon as suggested in PALM/JointInference user guide. At first I did give -npccon option but I got an error so I switched to -cmcx as suggested in the message. Best, Julia Shin
>>>
>>> On Wed, Jul 27, 2022 at 8:15 PM Matthew Webster <[log in to unmask]> wrote:
>>> Hello Julia,
>>>                  You only need to include ( any ) 2 of the 3 tp_x-vs-tp_y t-contrasts as the 3rd difference is implicitly included in the test - e.g. if there is no difference between tp1,tp2 and tp1,tp3 then by definition there is no difference between tp2,tp3.
>>>
>>> With reference to the Freesurfer link in your email, this would be equivalent to "Contrast 4 (tp-effect.mtx)" which in FSL/PALM terms would be F-contrast 0 0 1 1. The F-contrast in your email ( 1 1 1 1 ) as well as being rank-deficient, also includes the subject-mean t-contrast, which may or may not be what is intended.
>>>
>>> Kind Regards
>>> Matthew
>>>
>>> --------------------------------
>>> Dr Matthew Webster
>>> FMRIB Centre
>>> John Radcliffe Hospital
>>> University of Oxford
>>>
>>> > On 27 Jul 2022, at 11:01, Hye Min Shin <[log in to unmask]> wrote:
>>> >
>>> > Hello experts,
>>> >
>>> > I am trying to run FSL PALM specifically NPC using the example from the freesurfer repeated measures anova tutorial page: https://surfer.nmr.mgh.harvard.edu/fswiki/RepeatedMeasuresAnova#Contrast4.28tp-effect.mtx.29
>>> >
>>> > This freesurfer tutorial has 3 subjects with 3 timepoints with the following design matrix (column1-3 are subjects 1-3 and column 4 is tp1-vs-tp3 and column 5 is tp1-vs-tp3):
>>> >
>>> >     1     0     0     1     1
>>> >     1     0     0    -1     0
>>> >     1     0     0     0    -1
>>> >     0     1     0     1     1
>>> >     0     1     0    -1     0
>>> >     0     1     0     0    -1
>>> >     0     0     1     1     1
>>> >     0     0     1    -1     0
>>> >     0     0     1     0    -1
>>> >
>>> > It also has 5 contrasts, 4 t-tests and 1 F-test.
>>> >
>>> > The design matrix for the t-tests is the following:
>>> >
>>> > 0 0 0 1 0            (tp1-vs-tp2)
>>> > 0 0 0 0 1            (tp1-vs-tp3)
>>> > 0 0 0 -1 1           (tp2-vs-tp3)
>>> > 1 1 1 0 0             (is the subject mean == 0? )
>>> >
>>> > The design matrix for the F-test is the following:
>>> > 1 1 1 1
>>> >
>>> > I wanted to run the F-test to see if there is a group effect or specifically in this case the effect of timepoint. However, whenever PALM tries to run the F-test using NPC it keeps saying the rank is deficient.
>>> >
>>> > I ran PALM using npc with the following command:
>>> >
>>> > palm -i ./lh.thickness.sm15.mgh -m ./a_glmfit-lh/mask.mgh -d ./a_glmfit-lh/palm-onetail-1.3-lh/design.mat -t ./a_glmfit-lh/palm-onetail-1.3-lh/design.con -f ./adni_glmfit-lh/palm-onetail-1.3-lh/design.fts -n 5000 -npc -cmcx -o ./a_glmfit-lh/palm_mv-lh -logp
>>> >
>>> >
>>> > Is there something wrong with my F-test matrix or do I not need to run the F-test if I was able to run the t-tests on paired comparisons?
>>> >
>>> > Thank you in advance for your help
>>> >
>>> > Best,
>>> > Julia Shin
>>> >
>>> > ########################################################################
>>> >
>>> > To unsubscribe from the FSL list, click the following link:
>>> > https://www.jiscmail.ac.uk/cgi-bin/WA-JISC.exe?SUBED1=FSL&A=1
>>> >
>>> > This message was issued to members of www.jiscmail.ac.uk/FSL, a mailing list hosted by www.jiscmail.ac.uk, terms & conditions are available at https://www.jiscmail.ac.uk/policyandsecurity/
>>>
>>> ########################################################################
>>>
>>> To unsubscribe from the FSL list, click the following link:
>>> https://www.jiscmail.ac.uk/cgi-bin/WA-JISC.exe?SUBED1=FSL&A=1
>>>
>>> This message was issued to members of www.jiscmail.ac.uk/FSL, a mailing list hosted by www.jiscmail.ac.uk, terms & conditions are available at https://www.jiscmail.ac.uk/policyandsecurity/
>>>
>>> To unsubscribe from the FSL list, click the following link:
>>> https://www.jiscmail.ac.uk/cgi-bin/WA-JISC.exe?SUBED1=FSL&A=1
>>>
>>
>>
>> To unsubscribe from the FSL list, click the following link:
>> https://www.jiscmail.ac.uk/cgi-bin/WA-JISC.exe?SUBED1=FSL&A=1
>>
>>
>> To unsubscribe from the FSL list, click the following link:
>> https://www.jiscmail.ac.uk/cgi-bin/WA-JISC.exe?SUBED1=FSL&A=1
>>
>
>
> To unsubscribe from the FSL list, click the following link:
> https://www.jiscmail.ac.uk/cgi-bin/WA-JISC.exe?SUBED1=FSL&A=1
>
>
> To unsubscribe from the FSL list, click the following link:
> https://www.jiscmail.ac.uk/cgi-bin/WA-JISC.exe?SUBED1=FSL&A=1
>


########################################################################

To unsubscribe from the FSL list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/WA-JISC.exe?SUBED1=FSL&A=1

This message was issued to members of www.jiscmail.ac.uk/FSL, a mailing list hosted by www.jiscmail.ac.uk, terms & conditions are available at https://www.jiscmail.ac.uk/policyandsecurity/

To unsubscribe from the FSL list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/WA-JISC.exe?SUBED1=FSL&A=1