Dear David,
On 26 Feb 2018, at 14:10, Szabolcs David wrote:
> Dear Christian,
>
> On Wed, Feb 21, 2018 at 9:41 AM, Christian Gaser <
> [log in to unmask]> wrote:
>
>> Dear David,
>>
>> On Tue, 20 Feb 2018 01:05:10 +0100, Szabolcs David <
>> [log in to unmask]> wrote:
>>
>>> Dear Dr. Gaser and co.
>>>
>>> I would like to analyse T1s of (mostly) single subjects: monitoring
>>> cancer
>>> patients long term morphological changes after treatment.
>>> For the first try, I processed two images of a subject, but got
>>> stuck
>> after
>>> the 'Segment longitudinal data' step.
>>>
>>> My plan is to show the thickness/gyrification/etc. difference
>>> between the
>>> consecutive scans, for that first resample and smooth the
>>> l(r)h.thickness.data1 and data2 to FS mesh with the 'Resample and
>>> Smooth
>>> surface data' tool. Next: Surface calculator: simply 's1-s2' of the
>>> two
>>> previously merged and smoothed surfaces. With the 'Display surface'
>>> option
>>> I can visualize this resulting difference surface map, but how is it
>>> possible to use the 'Display surface results' tool for that? The
>>> latter is
>>> much fancier as I see and also more tools are available (surface
>>> atlases,
>>> flatmap, flatbrain, etc.)
>>
>> The surface based methods will be not that sensitive to track changes
>> over
>> time compared to DBM or VBM methods. However, you can try to analyze
>> cortical thickness. Folding measures such as gyrification are not
>> meaningful because the folding pattern is quite constant across life
>> in
>> adults.
>> Your steps were quite right: resampling and smoothing (try 8mm or
>> lower),
>> use surface calculator to calculate the difference and the "Display
>> Surface
>> Result" should also work if you don't move the images outside of your
>> analysis folder where the SPM.mat is saved. The function "Display
>> Surface
>> Results" tries to find the underlying surface using the SPM.mat file
>> (if no
>> other data were found) and is also trying to get some side
>> information by
>> using "lh" or "rh" for the hemispheres or "mesh" for the merged
>> hemispheres. Thus, take care that you use an output name for the
>> surface
>> calculator like "lh.diff_thickness.subject.gii" or
>> "mesh.diff_thickness.subject.gii" The 1st name part is for side
>> information, the 2nd for the measure and the 3rd for the subject
>> name.
>>
>
> Apparently the naming I used was not correct, now works
> perfectly!
>
>
>>>
>>> Also pretty much the same, but in DBM style ( showing differences
>>> between
>>> two scans using the swj_*.niis, smoothed with an 8mm kernel). In the
>>> OHBM
>>> educational talks, there is an example ( for example here, slide 4,
>>> https://www.pathlms.com/ohbm/courses/252/sections/1834/slid
>> e_presentations/15621
>>> ) How can one make such a represenation? Is it something like a
>>> percentage
>>> difference between the two swj_*.niis?
>> In CAT12 DBM is only supported for cross-sectional data and the
>> volume
>> changes are always related to the template and not to the baseline
>> scan.
>> Either use the VBM data and estimate the difference or try the
>> longitudinal
>> toolbox in SPM12. Please note, that there are substantial differences
>> in
>> these methods w.r.t. the expected changes (also see the CAT12 manual
>> for
>> more information). The SPM12 Longitudinal Toolbox has its strengths
>> for
>> larger long-time changes while CAT12 is more sensitive for short-time
>> changes. If you apply the SPM12 Longitudinal Toolbox to longitudinal
>> data
>> with very short time differences of a few days or weeks the methods
>> is
>> simply not expecting larger changes due to the short time interval.
>>
>
> Yes, exactly that's the expectation, that if anything changes
> that should
> be small.
> I don't quite get, not supporting longitudinal DBM is not supported as
> a
> concept?
> While for VBM the modulated warped segments are the subjects of the
> analyses (the (s)mwp1r*.nii maps), why we can not do the same with the
> deformations?
> It is not OK to compare the positional differences between every voxel
> and
> a standard brain, in my example, of two consecutive scan? Looking at
> the
> difference of..well different amount of local stretching/compressing
> compared to the template?
Longitudinal DBM is simply not yet implemented in CAT12, but as ai
mentioned you can try the SPM12 long. toolbox that is using that idea.
If you compare deformation differences to the template this will be far
less sensitive and not suitable for long. data.
Best,
Christian
>
>
>>>
>>> On the longer term I would like to do something similar of what is
>>> presented in the Frontiers 2009 article (this guy:
>> https://www.frontiersin.
>>> org/articles/10.3389/neuro.11.025.2009/full ; expecially this
>>> figure:
>>> https://www.frontiersin.org/files/Articles/668/fninf-03-025/
>>> image_n/fninf-03-025-g004.gif ) and I do belive that the OHBM
>>> educational
>>> talks were also presenting the same-ish material, but could not find
>>> any
>>> original researach on that particular data, same link as before
>>> slide 5 -
>>> What is applied test here? A one sample t-test?
>> If you have more time points you can use a polynomial model for
>> increase/decrease with time. In that example I have used a polynomial
>> regression with time (linear + squared effects) and analyzed the data
>> using
>> F-test for any effects due to linear or squared increase or decrease.
>>
>
> Thank you for the replies, very educational!
>
> Best,
> Szabolcs
>
>
>
>>
>> Best,
>>
>> Christian
>>
>>>
>>> Thank you in advance and apologies for this sea of qs.
>>>
>>> Kind regards,
>>> Szabolcs
>>>
>>
>>
>>
|