 |
 |
Hi Matthieu, There isn't a good solution for cases as these. It's possible to do sign-flippings for each subject as a whole, to allow inference on within-subject effects and interactions. If the sample size is large (i.e., many subjects), and the residuals are symmetric, this could be a reasonable option that can be combined with (restricted) permutations. All the best, Anderson On 15 November 2017 at 04:31, Matthieu Vanhoutte < [log in to unmask]> wrote: > Hi Anderson, > > Many thanks for this detailed answer, it helps ! > > I would have another question yet. I have to deal with non regular missing > timepoints as in the example below: > > *(subj1, visit1)* > *(subj1, visit3)* > *(subj2, visit2)* > *(subj2, visit3)* > *(subj3, visit1)* > *(subj3, visit2)* > *(subj3, visit4)* > *(subj4, visit2)* > *(subj4, visit3)* > *(subj4, visit4)* > > > Subjects 1 and 2 have two visits each, but subject 1 at 1st and 3rd visits > whereas subject 2 at 2nd and 3rd visits. In the same manner, subjects 3 and > 4 have three visits each but with different numerous of visit. > > In this case, could I as you explained allow exchangeability between > subject 1 and 2, and between subject 3 and 4 even if numerous of the visits > don't correspond ? > > Best regards, > Matthieu > > 2017-11-15 1:27 GMT+00:00 Anderson M. Winkler <[log in to unmask]>: > >> Hi Matthieu, >> >> For the design, have a look in the Example 6 of the randomise paper: >> https://doi.org/10.1016/j.neuroimage.2014.01.060 >> >> In the case of subjects with a different number of visits, only those >> with same number of visits should be permuted with each other, such that >> multi-level exchangeability blocks are required (these are described >> elsewhere, though: https://doi.org/10.1016/j.neuroimage.2015.05.092). >> Define a block per subject and mark these with negative indices (3rd column >> belo). Then one level above (2nd column), define one block for each number >> of visits a subject has, and keep these with positive indices. At the top >> level, a single block with a negative number (to prevent subjects with >> different number of visits from being mixed with each other). At the level >> of observations, unique indices per subject (and be one value per visit). >> Something like this: >> >> *-1,2,-1,1 (subj1, visit1)* >> *-1,2,-1,2 (subj1, visit2)* >> *-1,2,-2,1 (subj2, visit1)* >> *-1,2,-2,2 (subj2, visit2)* >> *-1,3,-3,1 (subj3, visit1)* >> *-1,3,-3,2 (subj3, visit2)* >> *-1,3,-3,3 (subj3, visit3)* >> *-1,3,-4,1 (subj4, visit1)* >> *-1,3,-4,2 (subj4, visit2)* >> *-1,3,-4,3 (subj4, visit3)* >> >> >> Subjects 1 and 2 have two visits each; subjects 3 and 4 have three >> visits. Subjects 1 and 2 can be permuted with each other; subjects 3 and 4 >> can be permuted with each other, but subjects with two visits can't be >> permuted with subjects with three visits (negative indices in the 1st >> column). The visits cannot be permuted with each other (negative indices at >> the 3rd column). >> >> Hope this helps. >> >> All the best, >> >> Anderson >> >> >> >> >> >> On 11 November 2017 at 03:45, Matthieu Vanhoutte < >> [log in to unmask]> wrote: >> >>> Hi Anderson, >>> >>> Thank you for replying. In case I want to study between subject factors, >>> as longitudinal progression in one group vs another one, how should I >>> define the design and contrasts with PALM ? >>> >>> Missing timepoints in some of the subjects will not be problematic ? >>> >>> Best, >>> Matthieu >>> >>> >>> Le 11 nov. 2017 3:17 AM, "Anderson M. Winkler" <[log in to unmask]> >>> a écrit : >>> >>> Hi Matthieu, >>> >>> It is, but this requires either compound symmetry, or that only >>> between-subject factors are investigated. >>> >>> All the best, >>> >>> Anderson >>> >>> >>> On 8 November 2017 at 09:34, Matthieu Vanhoutte < >>> [log in to unmask]> wrote: >>> >>>> Hi Anderson, >>>> >>>> I have 1 to 4 timepoints per subject, depending on the availability and >>>> quality of the processed data. In my case, is this possible to deal with >>>> PALM ? >>>> >>>> Best regards, >>>> Matthieu >>>> >>>> >>>> 2017-11-08 14:11 GMT+00:00 Anderson M. Winkler <[log in to unmask]> >>>> : >>>> >>>>> Hi Matthieu, >>>>> >>>>> Regarding the longitudinal aspect, in PALM compound symmetry has to be >>>>> assumed. If there are 2 timepoints, this is fine. With more than 2, it >>>>> becomes progressively more difficult. How many timepoints do you have? >>>>> >>>>> About the 2 modalities, these can be entered in the same call as two >>>>> separate inputs (two "-i"), and the option "-corrmod" to correct over them, >>>>> or "-npc" for a joint analysis via NPC. >>>>> >>>>> All the best, >>>>> >>>>> Anderson >>>>> >>>>> >>>>> On 8 November 2017 at 03:44, Matthieu Vanhoutte < >>>>> [log in to unmask]> wrote: >>>>> >>>>>> Dear Anderson, >>>>>> >>>>>> Thank you for your answer. My question concerns vertex-wise cortical >>>>>> surface data I have within 2 modalities over time. And I would like to >>>>>> correlate this two modalities across entire cortical surface longitudinally. >>>>>> >>>>>> Could you indicate me a process or how to do it with PALM ? >>>>>> >>>>>> Best regards, >>>>>> Matthieu >>>>>> >>>>>> >>>>>> 2017-11-08 2:41 GMT+00:00 Anderson M. Winkler <[log in to unmask] >>>>>> >: >>>>>> >>>>>>> Hi Matthieu, >>>>>>> >>>>>>> Just adding to Niels' answer: for other types of vertexwise data, >>>>>>> it's possible to do in PALM (i.e., it will read surface formats such as >>>>>>> FreeSurfer's or CIFTI). >>>>>>> >>>>>>> All the best, >>>>>>> >>>>>>> Anderson >>>>>>> >>>>>>> On 7 November 2017 at 15:55, Niels Bergsland <[log in to unmask]> >>>>>>> wrote: >>>>>>> >>>>>>>> Hi Matthieu, >>>>>>>> There isn't a direct longitudinal FIRST pipeline, but you can do it. >>>>>>>> >>>>>>>> What you need to do is run the concat_bvars command with all of >>>>>>>> your data. Then you run first_utils like in the guide. At this point, you >>>>>>>> need to use fslsplit on the output and then perform the subtractions with >>>>>>>> fslmaths and then merge the difference images back together with fslmerge. >>>>>>>> You probably know this already, but the order of the inputs to concat_bvars >>>>>>>> corresponds to their order in the 4D output from first_utils. So when you >>>>>>>> do the fslmaths and fslmerge, just make sure that you have things in the >>>>>>>> right order. >>>>>>>> Good luck! >>>>>>>> Niels >>>>>>>> >>>>>>>> On Tue, Nov 7, 2017 at 9:43 PM, Matthieu Vanhoutte < >>>>>>>> [log in to unmask]> wrote: >>>>>>>> >>>>>>>>> Dear FSL's experts, >>>>>>>>> >>>>>>>>> Do you know if there would be a mean to compare longitudinal >>>>>>>>> evolution of two vertex-wise data (i.e. kind of longitudinal correlation) >>>>>>>>> with PALM or others ? >>>>>>>>> >>>>>>>>> Best regards, >>>>>>>>> Matthieu >>>>>>>>> >>>>>>>> >>>>>>>> >>>>>>>> >>>>>>>> -- >>>>>>>> Niels Bergsland >>>>>>>> Integration Director / Research Assistant Professor of Neurology >>>>>>>> Buffalo Neuroimaging Analysis Center / University at Buffalo >>>>>>>> 100 High St. Buffalo NY 14203 >>>>>>>> <https://maps.google.com/?q=100+High+St.+Buffalo+NY+14203&entry=gmail&source=g> >>>>>>>> [log in to unmask] >>>>>>>> >>>>>>> >>>>>>> >>>>>> >>>>> >>>> >>> >>> >> >