Hi Saad,
Just looking for an expert confirmation, Will --omatrix1 work fine
with "multiple" seed surfaces (I mean freesurfer and FIRST)? Won't
this need to merge surfaces in a single .vtk or .asc format?
Cheers,
Sourena
On 7/10/12, Saad Jbabdi <[log in to unmask]> wrote:
> Hi
> you can speed things up if you include all the targets into the same
> probtrackx2 call, instead of running target by target.
> You can also do all seeds to all targets in a single call using --omatrix1,
> but as I said you may need to reduce the number of vertices (to around 10K
> to 30K depending on how much RAM you have).
>
> Cheers
> Saad
>
>
> On 5 Jul 2012, at 19:13, bi foc wrote:
>
>>
>> Hi Saad,
>>
>> I think I'm getting there. You warned me in your first reply, but I seem
>> to have seriously underestimated the amount of CPU hours I need for this
>> connectome approach. :-)
>> First is still running, so I'm working with 'only' 150 cortical rois.
>> Assuming I'm doing everything right, right now every region 2 region
>> tracking seems to cost around 40 minutes. I have 100 subjects....
>>
>> This is what I'm doing right now (testing with just two freesurfer labels
>> converted to .asc).
>>
>> probtrackx2
>> -x /home/FreesurferSubjects/1/labels/aparc-lh-043.asc
>> --targetmasks=/home/FreesurferSubjects/1/labels/aparc-lh-041.asc
>> --seedref=/home/FreesurferSubjects/1/mri/orig.nii.gz
>> --xfm=/home/FreesurferSubjects/1/FS2Bedpost.mat
>> -m /home/Bedpost/1.bedpostX/nodif_brain_mask.nii.gz
>> -s /home/Bedpost/1.bedpostX/merged
>> --dir=/home/Bedpost/1.bedpostX/tracks4341
>> --forcedir
>> --meshspace=freesurfer
>> --os2t
>> --s2tastext
>>
>> I am already searching for cluster solutions. I still need to get the time
>> down.
>>
>> I tried your suggestion: reducing the amount of vertices:
>>
>> mris_decimate lh.white lh.white_reduced -d 0.5
>> and then again
>> label2surf --surf=/home/FreesurferSubjects/1/surf/lh.white_reduced.asc
>> --out=/home/FreesurferSubjects/1/labels/aparc-lh-043_reduced.asc
>> --labels=/home/FreesurferSubjects/1/labels/aparc-lh-043.path
>>
>> before:
>> surfer: vertices=138108, faces=276212
>> after:
>> surfer: vertices=69055, faces=138106
>>
>> Half the amount of triangles halves the tracking time. Of course, one
>> would be tempted to bring it even further down. How low could I go?
>>
>> Do you have other tricks for speeding things up? for instance, using the
>> ventricles and grey matter of subcortical areas as exclusion masks?
>>
>>
>> Thank you very much for your help!
>>
>> Bernd
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>> Date: Fri, 29 Jun 2012 14:29:37 +0100
>> From: [log in to unmask]
>> Subject: Re: [FSL] (to saad) some directions for Freesurfer N to N surface
>> tracking with probtrackx2
>> To: [log in to unmask]
>>
>> Hi Bernd
>>
>>
>> Hi Saad,
>>
>> Thank you for your response. I have two additional questions:
>>
>> If I want to have my results as a connectivity matrices as well, would
>> your command work with --omatrix2?
>>
>> No, you will have to produce that matrix as a post-proc. Sorry.
>>
>>
>> I would like to work with subcortical surfaces from first and cortical
>> surfaces from freesurfer. First has been run with an average of two t1's
>> (manually betted). Freesurfer has been run with the same two T1's, but it
>> averaged and stripped itself. This means that the inputs are slightly
>> different off course.
>>
>> I have the feeling that I would need two transformation matrices if I
>> work with both these surface types (first2dti and freesurfer2dti). That
>> wouldn't be possible with the single command you provided would it? Would
>> surf2surf solve this or does this application also 'assume' that the same
>> T1 was source to both First and Freesurfer?
>>
>> You can only use one of the two conventions in probtrackx2. In order to
>> combine FIRST and FREESURFER surfaces, you can use surf2surf and work in
>> one the two spaces. If, for example, you use freesurfer as output for
>> surf2surf, then you should use freesurfer2dti as your transformation in
>> probtrackx2 for all your surfaces/rois
>>
>> cheers
>> saad
>>
>>
>>
>> Thank you!
>>
>> Bernd
>>
>> Date: Tue, 26 Jun 2012 16:34:21 +0100
>> From: [log in to unmask]
>> Subject: Re: [FSL] (to saad) some directions for Freesurfer N to N surface
>> tracking with probtrackx2
>> To: [log in to unmask]
>>
>> Hi Bernd
>> You can use probtrackx2 with all your ROIs (surfaces and volumes) as both
>> seeds and targets.
>> > probtrackx2 -x <asciiListOfROIs> --targetmasks=<asciiListOfROIs> [other
>> > options]
>>
>> The results will be stored as either surface files or volumes depending on
>> the seeds, and will be called seeds_<i>_to_<target>.
>>
>> If you run into RAM issues, which you might do considering how highres the
>> freesurfer surfaces are, you can either reduce the number of surface
>> vertices (can't remember what the freesurfer command is for this), or run
>> the above command for a subset of the seeds.
>>
>> Cheers
>> Saad
>>
>>
>>
>> On 26 Jun 2012, at 13:30, bi foc wrote:
>>
>> Dear Saad and other FSL experts
>>
>> I would like to track between all freesurfer surfaces of the 2009 atlas.
>> I'm very happy with what I've read about the new probtrackx2, but I'm not
>> not sure how to do this as computationally efficient as possible (I have
>> quite some participants). Would you be so kind to give me some
>> directions?
>>
>> I have:
>>
>> coregistered the DTI t2 to freesurfer with the bedpost output using flirt
>> and bbregister and inverted this matrix to get freesurfer2dti.mat
>> converted the freesurfer annotations from the aparc.a2009s.annot files to
>> .label files using mri_annotation2label for the left and right
>> hemispheres
>> converted the subcortical annotations from aseg.mgz to .label files using
>> mri_cor2label
>> converted all .label files to .asc files using mris_convert
>>
>>
>> I feel that I now have all the moving parts (hope I got them right). I
>> would like to track between all the freesurfer surfaces bidirectionally,
>> perhaps only using the ventricles as exclusion masks. average A>B and B>A
>> and normalize the averaged values using the sizes of the two regions.
>> Ideally, I would like to run 'as parallel as possible'. What would be the
>> best approach?
>>
>>
>> Thank you and kind regards,
>>
>> Bernd
>>
>>
>>
>>
>>
>>
>>
>>
>> --
>> Saad Jbabdi
>> University of Oxford, FMRIB Centre
>>
>> JR Hospital, Headington, OX3 9DU, UK
>> (+44)1865-222466 (fax 717)
>> www.fmrib.ox.ac.uk/~saad
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>> --
>> Saad Jbabdi
>> University of Oxford, FMRIB Centre
>>
>> JR Hospital, Headington, OX3 9DU, UK
>> (+44)1865-222466 (fax 717)
>> www.fmrib.ox.ac.uk/~saad
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>
> --
> Saad Jbabdi
> University of Oxford, FMRIB Centre
>
> JR Hospital, Headington, OX3 9DU, UK
> (+44)1865-222466 (fax 717)
> www.fmrib.ox.ac.uk/~saad
>
>
>
>
>
>
>
>
>
>
>
|