Thank you so much for your halp and time!
I haven't used Featquery_gui to generate these masks. Instead, I created
the non-linearly transformed mask manually and then fed it into a
modified version of the featquery script (i.e. with the "applywarp" step
deactivated since it had been done already). The linearly transformed
mask was created by hand with flirt for comparison. I had in mind that
the default thresold value was 0.5 - but that was true in featquery, not
in flirt, sorry!
It remains that I can't seem to obtain a similar mask using applywarp
Here's the applywarp command I ran:
applywarp -i mask_standard.nii.gz -o mask_native_space_non_linear.nii.gz
-r example_func.nii.gz -w highres2standard_warp_inv.nii.gz
It's the same command as the one used in the featquery script. I obtain
a mask with 3 voxels with a value of 1 without running any additional
tresholding command or anything. There msut be something obvious that
Stéphane Jacobs - Chercheur post-doctorant / Post-doctoral researcher
Espace et Action - Inserm U864
16 avenue du Doyen Lépine
69676 Bron Cedex, France
Téléphone / Phone: (+33) (0)4-72-91-34-38
Jesper Andersson a écrit :
> Dear Stephane,
> have you run Featquery_gui to obtain these masks? If so, did you keep
> the settings for
> "Do not binarise mask (allow weighting)"
> "Change post-interpolation thresholding of mask"
> constant for the two cases?
> When I look at your masks in native space it appears that the "linear"
> mask hasn't been thresholded at all (i.e. it contains a range of
> values 0 < v <= 1) which when all counted amounts to 72 voxels. The
> non-linear mask on the other hand seems to have been thresholded at a
> very high level, containing a single non-zero voxel with the intensity 1.
> If on the other hand I do a non-linear transform of your
> standard-space mask into native space I get a similarish set of voxels
> to what you have in in your linearly transformed mask. If I then
> threshold both at a 0.5 level I end up with 6 non-zero voxels for the
> linear case and 8 voxels for the non-linear case. I.e. a fairly
> reasonable correspondence.
> Also, the only remaining voxel in your non-linear mask coincides with
> the voxel with the highest intensity after my non-linear transform
> prior to thresholding.
> Hence, is looks as if the difference is due to on the one had no
> post-interpolation thresholding at all (your linear mask) and on the
> other hand very severe post-interpolation thresholding (thr>0.99) for
> your non-linear mask.
> Can you please try to run Featquery_gui again, making sure to keep
> these settings constant, and see if you get the same results?
> On 14 Aug 2009, at 13:23, StéŽphane Jacobs wrote:
>> Hi Jesper,
>> Thanks a lot for your response and your help. The reference number
>> for the upload is 622827.
>> I have included the 2 versions of the mask in example_func space,
>> obtained with the linear and non-linear registrations.
>> Stéphane Jacobs - Chercheur post-doctorant / Post-doctoral researcher
>> Espace et Action - Inserm U864
>> 16 avenue du Doyen Lépine
>> 69676 Bron Cedex, France
>> Téléphone / Phone: (+33) (0)4-72-91-34-38
>> Jesper Andersson a écrit :
>>> Dear Stephane,
>>>> I'm using Featquery to compute percent signal change from various
>>>> regions of
>>>> interest. I've first run Featquery with the default settings, which
>>>> were to
>>>> use the inverse of non-linear registration from highres2standard
>>>> and the
>>>> inverse of the affine transformation from example_func2highres.
>>>> However, I
>>>> noticed that several subjects (I still need to check the others), the
>>>> resulting mask transformed from standard to native EPI space is
>>>> small, 2 or 3 voxels at best, sometimes empty, while the volume of
>>>> the mask
>>>> in standard space is 528 mm2 (66 voxels). I checked the
>>>> example_func2standard registration and it looks fine.
>>>> To compare, I manually registered the standard mask to EPI space
>>>> using only
>>>> the inverse of the affine transformations from example_func2highres
>>>> highres2standard, using the default interpolation method and
>>>> post-interpolation threshold value. Now, I obtain a 72 voxel mask
>>>> (2812 mm2)
>>>> that is even bigger than the original one in standard space...
>>>> At this point I'm a bit confused as to why the mask obtained with
>>>> "back-registration" is so big (might need to adjust the
>>>> threshold?), but what I really don't get is why the results between
>>>> the 2
>>>> registration methods are so different.
>>>> Any hint as to what I might have done wrong would be greatly
>>> I must admit to being confused too. I have had a look at the
>>> scripts, and right now I don't have an idea so I would probably need
>>> to have a look at your data.
>>> Could you please tar up your
>>> example_func image file
>>> image file with mask in standard space
>>> highres2standard_warp image file
>>> highres2standard_warp_inv image file
>>> image file with mask after transformation into example_func space
>>> and then upload them to http://www.fmrib.ox.ac.uk/cgi-bin/upload.cgi
>>> and send me the reference number?