Print

Print
Hi Shruti,

That looks correct, although you need to use a brain extracted binary mask of your MNI152 4mm template, not the MNI152 4mm image. From the screenshots you sent earlier, it looks like you have a brain extracted version of the template (bg_image.nii.gz). So you can just binarise this (the -bin flag to fslmaths), and then use it in the fslmaths command that you listed.

Cheers,

Paul

On 17 January 2017 at 12:33, Shruti Narasimham <[log in to unmask]> wrote:
Dear Paul,

Thanks a lot. So I would run : 
fslmaths (MNI152 4mm here) -mas (mask.nii.gz here) 'my_output_mask.nii.gz' 

And then use this output mask in randomise? Sorry, I just wanted to confirm and make sure this is correct before running it again on my data.

Regards,
Shruti

On 17 January 2017 at 12:23, paul mccarthy <[log in to unmask]> wrote:
Hi Shruti,

maskALL.nii.gz is a 4D file containing the individual masks for every subject. mask.nii.gz file is the intersection of these masks - this is the default mask used by the dual_regression script.

You only need to change the mask at the randomise stage - you should use the intersection of the mask generated by dual_regression (mask.nii.gz), and the standard space brain mask (MNI152 4mm - you might need to create this manually if you haven't already).

So you only need to re-run randomise manually using this mask. 

If you wish, you can re-run dual_regression in full - you will need to edit the call to randomise in the dual_regression script, at the point where it says "EDIT HERE", to use your generated mask instead of the one generated by dual_regression.

Cheers,

Paul


On 17 January 2017 at 11:50, Shruti Narasimham <[log in to unmask]> wrote:
Dear Paul,

Thank you for your reply.

I ran the following command for dual_regression & randomise in one : 

dual_regression Dec21.gica/groupmelodic.ica/melodic_IC.nii 1 GLM-Dec20.mat GLM-Dec20.con 5000 Dec21.gica/groupmelodic.ica/DR_Output_Random `cat Dec21.gica/.filelist`

Based on what you are saying, 

  1. Does this mean I just have to run the randomise command separately again using an explicit mask? Or should I use an explicit mask for the dual regression stage also to correct the errors I am getting in my dr output?
  2. Can I supply this mask in the dr command itself and then run dr and randomise together like above? Or does the dr command always need the standard brain mask whereas randomise the explicitly stated subject mask?
  3. What mask image do I need to use in randomise so that it doesn't take the intersection of all the subjects? 
  4. In my dr log, it is written that a mask for each subject (mask_0001...mask0062) was created for each subject in the DR output folder. However, in this folder I can just see mask.nii.gz and maskALL.nii.gz. How are these different? Which one do I use for randomise?

I greatly appreciate your help on this matter. Thank you very much for taking the time to read my queries and patiently answer them.



Regards,
Shruti







On 17 January 2017 at 10:11, paul mccarthy <[log in to unmask]> wrote:
Hi Shruti,

The dual_regression script does not explicitly mask by the standard brain - it uses an intersection, across all subjects, of the brain extracted functional data after registration to standard space. 

This mask may contain voxels outside of the standard brain, due to the warping/smoothing/interpolation that occurs as part of the registration process.

To work around this, you may want to explicitly use a mask image for your call to randomise.

Cheers,

Paul

On 16 January 2017 at 11:59, Shruti Narasimham <[log in to unmask]> wrote:
Yes, please find the images attached below.

In some small activations appear outside the brain, whereas in some tcfe_corrp results, there are bigger clusters that appear outside. 

I have compressed them due to the file size limit.

On 16 January 2017 at 11:56, Shruti Narasimham <[log in to unmask]> wrote:
Yes, please find the images attached below.

In some small activations appear outside the brain, whereas in some tcfe_corrp results, there are bigger clusters that appear outside. 

I have compressed them due to the file size limit.



On 16 January 2017 at 11:54, Shruti Narasimham <[log in to unmask]> wrote:
Yes, please find the images attached below.

In some small activations appear outside the brain, whereas in some tcfe_corrp results, there are bigger clusters that appear outside.

On 16 January 2017 at 11:42, paul mccarthy <[log in to unmask]> wrote:
Hi Shruti,

Could you send some screenshots demonstrating the problem?

Cheers,

Paul

On 16 January 2017 at 11:06, Shruti Narasimham <[log in to unmask]> wrote:
Dear all,

Sorry for posting this again, but I since I didn't get any reply to it, I am looking for any advice on the following.

After running Melodic and dual regression then, I am getting contrasts that show activation outside of the standard brain after dual regression. I have checked the registration and it all looks fine. Could there be any other possible problem (eg. in the pre-processing) with this? if so, is it ok if I do my pre-processing in SPM and input these files to Melodic?

My parameters in the registration step of Melodic are :
  1. Linear Normal Search BBR for the main structural images where I have input BET outputs for each corresponding brain structure
  2. Non-linear Warp resolution 10mm for Standard space where I have input the FSL standard T1 2mm brain image. Does this also have to be brain extracted with BET before?
  3. Resampling resolution of 4mm

Should I change any of these, for the above problem I am running into, to not happen?


Looking for any help and inputs on this.

Thank you.

Regards,
Shruti