Print

Print
Hi Mark and Andreas!

Thank you so much for your very interesting comments.This is all really helpfull!

Andreas:
Actually the reason why I thought about dilating my ROIs is the fact that they are so cortical that I have a hard time getting fibers tracked at all. Thus I wanted to dilate them until they comprise a given number of voxels with a FA value > 0.2 where I have hope that the tracking can begin.
My ROIs are functionally defined (parietal/occipital/temporal, cognitive tasks) and unfortunately not so easy to locate as are M1 or internal capsula. .
Maybe you are very right in saying that functionally defined ROIs have a lot of problems and disadvantage against anatomically defined ones, I am only too avare of this, in my daily data struggle.
Yes my data are smoothed (I got them from a former collegue). And you are right this also influences the localisation of the functional ROI.
However the main problem for me is a acurate registration from the epi and the struct to the DTI space. I have been trying FLIRTing in every direction, but even if it seems quite OK at a first glance... when superposing the anatomical cerebellar exclusion-mask  (in order to prevent occipito-cerebellar shortcuts) and the functional ROIs, eihter one is either to hight or to low and I either get nothing or I get meaningless trackts from the infratemporal lobe to the vermis and the pons to the internal capsule until M1/S1...
I would love to try FNRIT, but I am hesitating with FSL 4.1.0 as I already did a lot in FSL 4.0.4 and the data initially are in ANALYZE.

Mark:
Thank you! I tryed as you said from a single voxel and it works fine!
The -kernel sphere 3 however dilates only in the plain, right, so that the smallest dilatation step is sphere 4 right?

Thank you both so much!
All the best!
Markus


2008/10/21 Andreas Bartsch <[log in to unmask]>
Hi,
my 2 cents on that take:
I absolutely agree with Mark on the possibility to normalize counts by relating them to the number of voxels in the original ROI.
However, shape may matter. Imagine, for example, a tumor pushing the motor strip forward so that it is more curved and bended than on the contralateral side. Or just think of the normal anatomical variability. Of course, you should incorporate that information and try to draw your ROIs to cover M1 as best as you can. Aside from the fact that motor mapping is rarely necessary to identify M1, activation studies are subject to their own localization errors / variability and anatomical differences are "smoothed" (also depending on the preprocessing of your FMRI). Dilated ROIs around activation peaks will completely attenuate / ignore the underlying anatomical shape. Btw - contrary to a widespread belief, it has NOT been shown conclusively that functional is superior to anatomic seeding for pyramidal tractography: As long as anatomic localization is well preserved and not concealed by a lesion's mass effect or abnormalities of gyration, for example, precentral ROI placement according to pure structural criteria remains straight-forward and entirely sufficient. In these cases, benefits of functional ROI definitions cannot be expected. The study of Smits et al. [2007], which suggested a benefit of functional over anatomical guided tractography, was biased in that their FMRI-informed pyramidal tractography took advantage of a two-ROI approach whereas the anatomically guided counterpart employed just a single ROI in the peduncle. The single case illustrating anatomical landmark vs. FMRI-based pyramidal tractography in Schonberg et al. [2006] is not convincing because the anatomically defined ROI is only shown for the posterior limb of the internal capsule. Probably, it was also used in a single-ROI approach – otherwise the fibres rostral to the lesion should not have been extracted. Also, the particular lesion could well be of extra-axial origin, meaning that neither presurgical tractography nor FMRI would have been necessary.
So the question is: does the functional seeding makes sense for your particular tractography? (It may, for example, when you are tracking the arcuate fascicle.) And how do you preprocess your FMRI data? I.e., if you for some reason have to smooth by a rather large kernel, you may be fine dilating the peak activation voxel. It just depends;)
Cheers-
Andreas

-----Ursprüngliche Nachricht-----
Von: FSL - FMRIB's Software Library im Auftrag von Mark Jenkinson
Gesendet: Di 21.10.2008 14:14
An: [log in to unmask]
Betreff: Re: [FSL] How dilate ROI with fslmaths

Hi,

Now I see what you are trying to do.
There are two points worth mentioning:

 - Does it makes sense to have an area around your activation peak
which,
       on average, goes further away from the peak just because it is near
       the edge of the brain mask?  Isn't there another way to normalise for
       what you want in your final output?  For instance, dividing by the
number
       of voxels in the original ROI would normalise counts.  It depends on
       what your final objective is.

 - If you really do want to control the size of the ROI more
sensitively then
       you are better off dilating directly from the original peak, rather
than
       expanding from the boundary of an already large sphere.  This is
because
       you have a large surface area of the 5mm sphere and the dilation cannot
       distinguish between different points on the surface.  However, if you
use a
       large kernel directly on a single voxel (your activation peak) then
you can
       distinguish between points based on how far they are (to sub-voxel
precision)
       away from the original point.  For example, taking a one voxel seed
point
       in 2mm space and dilating by spheres of radii 1,2,3,4,5,6,7,8 mm
gives you
       outputs of size 1,7,19,33,81,123,179,257 voxels.  And you can use
smaller
       changes, for instance a 7.3mm kernel gives 203 voxels and a 7.5mm
kernel
       gives 251 voxels.  So, by trying different values (and masking
afterwards) you
       can probably find a value for the kernel which gets you close.

However, I would think more about why you want the same number of
voxels in
each ROI, and whether there are better ways of controlling for this,
as comparing
things with ROIs of different shape is also probably not what you want.

All the best,
       Mark




On 21 Oct 2008, at 12:14, Markus Gschwind wrote:

> Hi Mark!
>
> Thanks a lot for your kind answer! OK, I will be more precise.
>
> I need ROIs (=binarised spheres around the activation peak) which
> have about the same number of voxel (for fibertracking).
>
> The problem is that some peaks are quite close to the brain border
> and thus the sphere (5mm radius) extends beyond the mask and gets
> partially cut, which reduces its number of voxels (3x3x3 image
> reslotion).
>
> I thus would like to dilate those ROIs in direction of inside the
> brain in order to compensate for this.
> Now I found that when I am dilating a sphere of 170 voxels (this is
> what the spm-script produced for that sphere of 5mm) with the
> default kernel of 3, I get a sphere of about 450 voxels and not 200.
> This is due to the small size. If I would be dilating a 1000 voxel
> sphere, the increasing step could be a much smaller percentage.
> Maybe this is a principal problem... and I am thinking the wrong
> direction...
>
> What I would like to achieve is that every ROI has finally about 200
> +/- 10 voxels
>
> Thanks a lot for your help!
> All the best, Markus
>
>
>
>
> 2008/10/21 Mark Jenkinson <[log in to unmask]>
> Hi,
>
> What do you mean by "a tiny bit"?
> Is one voxel too much?
> For one voxel you can just use the default kernel.
>
> If you have a binary mask (or ROI mask) then it isn't
> really well defined to add less than one voxel onto the edge.
>
> Maybe if you describe what you are trying to achieve
> then we can work out what is the best thing to do.
>
> All the best,
>        Mark
>
>
>
>
> On 21 Oct 2008, at 10:25, Markus Gschwind wrote:
>
> Hello again!
>
> It seems that
>
> fslmaths INPUT  -kernel sphere 0.5 -dilD OUTPUT
>
> or
> fslmaths INPUT  -kernel sphere 1 -dilD OUTPUT
>
> is not possible.
>
> It begins at -kernel sphere 3.
>
> Is that right?
>
> How could I dilate my ROI only a tiny bit?
>
> Any hints would be greatly apreciated!
> Thanks,
> Markus
>
>
>
> 2008/10/17 Matthew Webster <[log in to unmask]>
> Hello,
>           The kernel needs to be set before the desired filtering
> operation ( with the default kernel being a 3x3x3 box ).
>
> e.g. to dilate using a 5mm radius spherical kernel around each zero-
> voxel use the following command:
>
> fslmaths INPUT  -kernel sphere 5 -dilD OUTPUT
>
> Unfortunately there is no specific option to dilate such that the
> output image has a specific number of voxels...
>
> Many Regards
>
> Matthew
> Hi Steve!
>
> Thank you for the hint. However it is still not clear to me how to
> use this command.
>
> fslmaths INPUT -dilD OUTPUT  >> dilates + 1 voxel all around
>
> fslmaths INPUT -dilD -kernel 5 OUPUT >> no effect
>
> fslmaths INPUT -kernel sphere 5 OUTPUT >> no effect
>
> How would be the correct command to dilate the input so as the
> OUTPUT finally has 500 voxels?
>
> Thank you so much!
>
> All the best,
> Markus
>
>
>
>
> 2008/10/16 Steve Smith <[log in to unmask]>
> Hi - you can change the dilation kernel with the -kernel option
> (type fslmaths to see the options).
> Cheers.
>
>
>
> On 16 Oct 2008, at 15:00, Markus Gschwind wrote:
>
> Hi experts!
>
> I have to dilate ROIs up to a certain voxelsize.
>
> I think this works with fslmaths -dilD or -dilM
>
> However I would like to know if there is a mean  to stop the
> dilation at a given number of voxels.
> Does anybody know?
>
> Thanks a lot!
> Markus
>
>
>
> --
> Dr. med. Markus Gschwind, M.D.
> Laboratory for Neurology and Imaging of Cognition
> Dept of Neurosciences
> University Medical Center (CMU)
> 1 Michel-Servet - 1211 GENEVA - CH
>
> Tel 0041 (0) 22 379 5324
> Fax 0041 (0) 22 379 5402
> email: [log in to unmask]
> http://labnic.unige.ch
>
>
> ---------------------------------------------------------------------------
> 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
> ---------------------------------------------------------------------------
>
>
>
> --
> Dr. med. Markus Gschwind, M.D.
> Laboratory for Neurology and Imaging of Cognition
> Dept of Neurosciences
> University Medical Center (CMU)
> 1 Michel-Servet - 1211 GENEVA - CH
>
> Tel 0041 (0) 22 379 5324
> Fax 0041 (0) 22 379 5402
> email: [log in to unmask]
> http://labnic.unige.ch
>
>
>
>
> --
> Dr. med. Markus Gschwind, M.D.
> Laboratory for Neurology and Imaging of Cognition
> Dept of Neurosciences
> University Medical Center (CMU)
> 1 Michel-Servet - 1211 GENEVA - CH
>
> Tel 0041 (0) 22 379 5324
> Fax 0041 (0) 22 379 5402
> email: [log in to unmask]
> http://labnic.unige.ch
>
>
>
> --
> Dr. med. Markus Gschwind, M.D.
> Laboratory for Neurology and Imaging of Cognition
> Dept of Neurosciences
> University Medical Center (CMU)
> 1 Michel-Servet - 1211 GENEVA - CH
>
> Tel 0041 (0) 22 379 5324
> Fax 0041 (0) 22 379 5402
> email: [log in to unmask]
> http://labnic.unige.ch



--
Dr. med. Markus Gschwind, M.D.
Laboratory for Neurology and Imaging of Cognition
Dept of Neurosciences
University Medical Center (CMU)
1 Michel-Servet - 1211 GENEVA - CH

Tel 0041 (0) 22 379 5324
Fax 0041 (0) 22 379 5402
email: [log in to unmask]
http://labnic.unige.ch