Print

Print
Hi Conor,

I think this error message isn't because of lack of overlap, but because the mask and the zstat image don't have the same size. This error should disappear if registration is performed, or if all images are put into some standard space. But if the masks actually don't overlap, then there will be a different error later as there would be nothing to be tested (i.e., the intersection would be an empty set of voxels).

By other methods I meant either not using a voxelwise test, replacing it for some region-based comparison or analysis, or reformulating the problem. Often if a test becomes impossible it's probably because the original question wasn't meaningful...

All the best,

Anderson


On 2 February 2018 at 01:19, Conor Owens-Walton <[log in to unmask]> wrote:

Hi Anderson,


Thanks again for your continued support.


So I believe that your second idea came to fruition. The masks are all in distinct locations and the higher-level FEAT group analysis breaks down when I use a pre-threshold mask.


--------------


WARNING:: Inconsistent orientations for individual images in pipeline!
          Will use voxel-based orientation which is probably incorrect - *PLEASE CHECK*!

Image Exception : #3 :: Attempted to multiply images/ROIs of different sizes
libc++abi.dylib: terminating with uncaught exception of type RBD_COMMON::BaseException

FATAL ERROR ENCOUNTERED:
COMMAND:
/usr/local/fsl/bin/fslmaths thresh_zstat1 -mas /Users/u4307082/Desktop/rs_preprocessing/generic_files/sum_motor_cortex_thr2.nii.gz thresh_zstat1
ERROR MESSAGE:
child killed: SIGABRT
END OF ERROR MESSAGE

-------------


You mention I might have to resort to other methods, can you possible direct me towards any resources that outline these methods? Or can you give me any direction?


I really want the work to follow an a priori design and thus using a ROI for my search is important to me.


Kind regards,

Conor 

From: FSL - FMRIB's Software Library <[log in to unmask]> on behalf of Anderson M. Winkler <[log in to unmask]>
Sent: Monday, 29 January 2018 3:33:59 PM
To: [log in to unmask]
Subject: Re: [FSL] ROI Higher-level feat analysis
 
Hi Conor,

In the higher level you'd use an average mask that is the same for all subjects. To produce such mask, align all your current lower level mask to standard space (I think featregapply may do this automatically; otherwise you'd register each one by hand or with a short script, using the warp files that Feat leaves in the directories for each 1st level). Then average all aligned masks and threshold at some reasonable level, say, at 50%.

It might be the case that the masks are at the lower level are in such distinct locations for each subject that they may not overlap after being placed in standard space. In that case, the voxelwise strategy at the higher level wouldn't be adequate, and you'd have to resort to other methods.

All the best,

Anderson


On 20 January 2018 at 18:26, Conor Owens-Walton <[log in to unmask]> wrote:

Hi Anderson, thanks so much for getting back to me.


I understand what you mean, and i thought about putting a mask in the 'pre-threshold masking' section of the feat_gui (in my lowres func space) for my higher-level analysis would work, but wouldn't that mask be specific to one subject (when I have n = 99 subject specific lowres masks that i wanted to use as cortical ROIs)?


Is it fine to just use one target cortical ROI mask here, and hope that it is aligned correctly for all subjects?


Sorry if I am missing a simple point here.


Kind regards,

Conor

From: FSL - FMRIB's Software Library <[log in to unmask]> on behalf of Anderson M. Winkler <[log in to unmask]>
Sent: Sunday, 21 January 2018 1:04:53 AM
To: [log in to unmask]
Subject: Re: [FSL] ROI Higher-level feat analysis
 
Hi Conor,

Is there a reason why you couldn't do the 1st level analysis (step 2) for the whole brain, and at the higher level (step 3) provide a mask for the ROIs? This shouldn't be a problem...

All the best,

Anderson


On 18 January 2018 at 00:06, Conor Owens-Walton <[log in to unmask]> wrote:
Hi everyone,

I have had a pretty thorough look through FSL JISCMAIL but to no avail, so thought I would ask the group.

I want to compare functional connectivity of a PD cohort with controls using a parcellation of the thalamus and confine my search to target ROIs in the cortex. I am fine with the creation and registration of masks etc.

I have performed the following:

1) Preprocessing using MELODIC and ICA denoising - via FIX
2) Voxelwise individual GLM using timseries extracted from parcellations - via FEAT (I tried using pre-threshold cortical masks here)
3) Higher-level statistics on Feat directories from 2) controlling for (demeaned) age and sex

I am concerned that the data that gets passed from 2) up to 3) is the total cope image and is not confined to the pre-threshold mask region that I incorporated.

Is there anyway I can perform 3) just with the thresh_zstat images produced from step 2)?

Or can someone suggest a better approach to compare my groups confining my search to certain cortical regions (which I have masks for in example_func space)?

Thanks for your time,
Conor