 |
 |
Hi Matthew, I've just found out that it works if I don't use variance smoothing... any ideas? Let me know if you still want the data (i.e. if you can get "-v 3 --vxl=3 --vxf=confounds.nii" to work on any other data). Many thanks, Ged On 6 May 2010 13:53, Matthew Webster <[log in to unmask]> wrote: > Hi Ged, > The versions of randomise from 4.1.3 onwards should work correctly with confound EV's. If you're > getting a persistent error message, I would be happy to look over the data if you upload it to our site. > > Many Regards > > Matthew >> Hi Matthew, >> >> Has this internal version with better handling of voxelwise nuisance >> covariates made it out yet? It's not obvious to me from the comments >> here: >> http://www.fmrib.ox.ac.uk/fsl/fsl/whatsnew.html#revisions >> >> If it's not yet released, please could you give me a link to a binary? >> I'm currently sat with: >> Linux 2.6.21.5-smp i686 Intel(R) Xeon(TM) >> but have access to a couple of other Linux machines (32 or 64 bit), >> and might be able to borrow a Mac if easier. >> >> With version 2.1 of randomise, I seem to be getting errors about >> incompatible dimensions (see below), even though my fMRI and VBM data >> have the same x, y, z, and t dimensions, and my design.mat has the >> same t dimension. Is it possible this is a bug that is fixed in the >> newer versions, or do you think I have miscounted somewhere? As a >> sanity check, I have tried using just the fMRI data, or just the VBM >> data (with the same design, contrast, and mask) and both work without >> any errors, which would seem to imply that the dimensions should be >> consistent for fMRI adjusted for VBM, right? (the only change to the >> apparently successful fMRI analysis is to add --vxl=3 and >> --vxf=vbm.nii, does that sound right?). >> >> I briefly wondered if EVs were counted from zero (like voxels and >> times in fslview, etc.) but using --vxl=2 for my third EV also results >> in the same error. >> >> Many thanks, >> Ged >> >> P.S. The not entirely helpful error message: >> >> ERROR: Program failed >> An exception has been thrown >> Logic error:- detected by Newmat: incompatible dimensions >> Trace: SubMatrix(=). >> >> >> On 9 April 2009 10:32, Matthew Webster <[log in to unmask]> wrote: >>> Hi, >>> It looks like you are using the voxelwise EV as a confound, rather than >>> an EV of interest, this is a situation which is handled a lot better in our >>> internal build of randomise - if you let me know what architecture you are >>> using ( Mac, Linux 32/64 etc ) I can send you a link to newer binary. >>> >>> Many Regards >>> >>> Matthew >>> >>>> I tested with few images to understand. I used a simple regression >>>> model: >>>> >>>> Constant Variable >>>> 1.000000e+00 8.000000e+00 >>>> 1.000000e+00 1.000000e+01 >>>> 1.000000e+00 1.200000e+01 >>>> 1.000000e+00 1.800000e+01 >>>> 1.000000e+00 2.000000e+01 >>>> 1.000000e+00 2.200000e+01 >>>> >>>> I used Glm to generate my .mat and .con files, so the voxel-dependent >>>> EV column was filled with mean values across all voxels for each image. >>>> >>>> Constant Variable voxel-dependent >>>> 1.000000e+00 8.000000e+00 0.000000e+00 >>>> 1.000000e+00 1.000000e+01 0.000000e+00 >>>> 1.000000e+00 1.200000e+01 0.000000e+00 >>>> 1.000000e+00 1.800000e+01 1.704375e-04 >>>> 1.000000e+00 2.000000e+01 1.704375e-04 >>>> 1.000000e+00 2.200000e+01 1.704375e-04 >>>> >>>> With this type of analysis, I noticed that my statistical map (contrast >>>> 0 -1 0) was different from the analysis without voxel-dependent EV even >>>> in voxels where there was no variation in the other imaging modality. >>>> After that, I tested putting only 1 in the voxel-dependent EV column: >>>> >>>> Constant Variable voxel-dependent >>>> 1.000000e+00 8.000000e+00 1 >>>> 1.000000e+00 1.000000e+01 1 >>>> 1.000000e+00 1.200000e+01 1 >>>> 1.000000e+00 1.800000e+01 1 >>>> 1.000000e+00 2.000000e+01 1 >>>> 1.000000e+00 2.200000e+01 1 >>>> >>>> This time, I got the same statistics for both analysis in the voxel >>>> where there was no variation in the other imaging modality but different >>>> statistics in voxel where there was variation. In this case, the results >>>> makes more sense. Basically, you want to correct voxels where the >>>> variation in the other imaging modality explains the results in you >>>> imaging modality of interest. I'm not sure if I'm correct . Maybe, I'm >>>> missing something as I'm combining permutations and voxel-dependent >>>> variables. Could you help me to clarify this thing? >>>> >>>> Herve >>>> >>>> >>>> >>>> -----Original Message----- >>>> From: Matthew Webster [mailto:[log in to unmask]] >>>> Sent: Tuesday, April 07, 2009 12:26 PM >>>> To: [log in to unmask] >>>> Subject: Re: [FSL] randomise and voxel-dependent EVs >>>> >>>> Hi, >>>> The numbers in the column that you replace with the voxelwise-EV >>>> will be used to determine valid permutations - does this explain the >>>> results you're seeing? >>>> >>>> Many Regards >>>> >>>> Matthew >>>> >>>>> It seems to work but I noticed that the numbers in the column I want >>>>> to replace with voxel-dependent EV still matter. Is each images >>>>> weighted by the corresponding number in the column? Should I put a >>>>> column fill with only with 1? >>>>> >>>>> Herve >>>>> >>>>> -----Original Message----- >>>>> From: Matthew Webster [mailto:[log in to unmask]] >>>>> Sent: Thursday, March 26, 2009 6:32 AM >>>>> To: [log in to unmask] >>>>> Subject: Re: [FSL] randomise and voxel-dependent EVs >>>>> >>>>> Hi, >>>>> You need to supply two additional inputs to randomise: A 4D volume >>>> >>>>> file, where each voxel timeseries corresponds to equivalent >>>>> voxel-dependent EV and a number telling randomise which column in your >>>> >>>>> original input design the voxelwise EV replaces: >>>>> >>>>> e.g --vxl=2 --vxf=my_input_EV >>>>> >>>>> tells randomise for each voxel to replace the 2nd EV in your input >>>>> design with the appropriate "timeseries" from my_input_EV. >>>>> >>>>> Many Regards >>>>> >>>>> Matthew >>>>> >>>>>> Thank for your answer but I'm not sure how to do it. When you use >>>>>> randomise, you feed it with your 4D image, the design.mat file and >>>>>> the >>>>> >>>>>> design.con file but none of them contain information about the >>>>>> voxel-dependent EV except one column including mean across all >>>>>> voxels. >>>>>> How can I specify my voxel-dependent EV in the randomise model? >>>>>> >>>>>> Thanks, >>>>>> >>>>>> Herve >>>>>> >>>>>> >>>>>> -----Original Message----- >>>>>> From: Steve Smith [mailto:[log in to unmask]] >>>>>> Sent: Wednesday, March 25, 2009 9:00 AM >>>>>> To: [log in to unmask] >>>>>> Subject: Re: [FSL] randomise and voxel-dependent EVs >>>>>> >>>>>> Hi, >>>>>> >>>>>> See the usage - you should just be able to specify this additional >>>>>> component to your model. >>>>>> >>>>>> Cheers. >>>>>> >>>>>> >>>>>> >>>>>> On 24 Mar 2009, at 14:42, Herve Lemaitre wrote: >>>>>> >>>>>>> Hi FSL experts, >>>>>>> >>>>>>> In a TBSS analysis, I would like to use randomise with a statistical >>>> >>>>>>> design including one voxel-dependent EV. Is it implemented in >>>>>>> randomise and do I have to change the way to run randomise to take >>>>>>> into account this EV? >>>>>>> >>>>>>> Thanks, >>>>>>> >>>>>>> Herve Lemaitre >>>>>>> >>>>>>> >>>>>>> >>>>>> >>>>>> >>>>>> --------------------------------------------------------------------- >>>>>> - >>>>>> -- >>>>>> --- >>>>>> Stephen M. Smith, Professor of Biomedical Engineering Associate >>>>>> Director, Oxford University FMRIB Centre >>>>>> >>>>>> FMRIB, JR Hospital, Headington, Oxford OX3 9DU, UK >>>>>> +44 (0) 1865 222726 (fax 222717) >>>>>> [log in to unmask] http://www.fmrib.ox.ac.uk/~steve >>>>>> --------------------------------------------------------------------- >>>>>> - >>>>>> -- >>>>>> --- >>>>>> >>>>> >>>> >>> >> >