Perfect. Thanks
I just tried with and without -cost mutualinfo
No detectable difference (on one case)
Sven
On 16.03.2011, at 21:30, Gwenaëlle DOUAUD wrote:
> As MJ said, this is a cost function that helps for registration across different modalities... You can set this in the flirt advanced options.
>
> Cheers,
> Gwenaelle
>
>> De: Sven Haller <[log in to unmask]>
>> Objet: Re: [FSL] Normalize DATSCAN to MNI (ideally non-linear)
>> À: [log in to unmask]
>> Date: Mercredi 16 mars 2011, 19h56
>> PS
>> What is mutual information?
>>
>> On 16.03.2011, at 12:54, Gwenaëlle DOUAUD wrote:
>>
>>>
>>>
>>> Hi Svan,
>>>
>>> my 2p on this...
>>>
>>> When I had to deal with F-Dopa in controls and
>> patients, I found that a linear registration with mutual
>> information to the T1w, followed by the application of the
>> warp derived from the optimised VBM was working very well...
>> The other advantage is that you will also place DAT and VBM
>> in the same space, so that if you want to regress out
>> voxel-wise the VBM "confound", you'd be able to do so.
>>>
>>> Cheers,
>>> Gwenaelle
>>>
>>>
>>>
>>>> De: Sven Haller <[log in to unmask]>
>>>> Objet: Re: [FSL] Normalize DATSCAN to MNI (ideally
>> non-linear)
>>>> À: [log in to unmask]
>>>> Date: Mercredi 16 mars 2011, 10h55
>>>> Dear Mark, Cornelius and Andreas
>>>>
>>>> Thanks a lot for you input!
>>>>
>>>> So far I have no experience with DatScans, and I
>> highly
>>>> appreciate your comments!
>>>>
>>>> My idea is to do a multi-modal analysis of GM
>> (VBM), WM
>>>> (TBSS), and DatScan
>>>>
>>>> Concerning DatScan, I fully agree with the
>> concerns of
>>>> Cornelius. The pattern is quite different from
>> the
>>>> anatomical boundaries of the basal ganglia, in
>> particular in
>>>> patients.
>>>> In a previous study, we (together with Andreas)
>> used TBSS
>>>> preprocessing of DTI to derive a non-linear
>> transformation
>>>> matrix dominantly driven by the deep white matter
>>>> structures. We then applied this matrix to SWI
>> images
>>>> (linear SWI to DTI, non linear to MNI using the
>> TBSS
>>>> transformation matrix). The results looked quite
>> nice. So I
>>>> was hoping to do the same thing i.e. linear
>> DatScan to FA.
>>>> The problem here in my view is the very peculiar
>> image
>>>> contrast of the DatScan images. My hope is that by
>> this
>>>> approach I could avoid some problems of
>> normalization
>>>> related to the decreased uptake predominantly in
>> the
>>>> posterior parts in patients.
>>>>
>>>> Concerning the other question by Cornelius, I want
>> to use a
>>>> control ROI in occipital area (after spatial
>> registration)
>>>> and calculate a ratio for each voxel.
>>>>
>>>> In the literature, I now found this link
>>>> http://jnm.snmjournals.org/cgi/reprint/48/9/1459.pdf
>>>>
>>>> The authors however use essential tremor (ET)
>> patients. In
>>>> my limited experience of DatScans, I understand
>> that the
>>>> variability of uptake in this pathology is
>> considerably less
>>>> pronounced as compared to Parkinson patients -
>> hence less
>>>> problems for the spatial registration. I am thus
>> not sure
>>>> whether this approach can be applied to my dataset
>> ( also it
>>>> was done in SPM2 :-( )
>>>>
>>>> I would like to do randomize analyses of GM (VBM),
>> WM
>>>> (TBSS) and DatScan in the same subjects, therefore
>> I would
>>>> like to pre-process all data in FSL and
>> co-register into MNI
>>>> space
>>>>
>>>> Thanks a lot for your help
>>>>
>>>> Sven
>>>>
>>>>
>>>> On 16 mars 2011, at 11:18, Mark Jenkinson wrote:
>>>>
>>>>> Dear Cornelius and Sven,
>>>>>
>>>>> OK, I've seen Sven's images now and have a
>> much better
>>>> feel for it.
>>>>> The segmentation approach for an individual
>> structural
>>>> sounds flawed
>>>>> based on Cornelius's observation that the
>> DaTScan does
>>>> not show
>>>>> all the structure borders clearly.
>> However, the
>>>> good news is that from
>>>>> the images it is clear that you can find the
>> whole
>>>> brain in the scan
>>>>> and there main boundaries (at least
>> brain/non-brain)
>>>> are shown OK.
>>>>> There is clearly a hot-spot in the middle of
>> the scan,
>>>> but that can be
>>>>> dealt with by (a) hoping that the multi-modal
>> cost
>>>> functions (mutual
>>>>> information, etc.) will just be OK with this,
>> (b)
>>>> "inverse" thresholding
>>>>> the image so that intensities over a certain
>> threshold
>>>> are capped at
>>>>> that value (thus eliminating the bright
>> spots), or (c)
>>>> using a basal
>>>>> ganglia drived mask (from an individual
>> segmentation
>>>> or a standard
>>>>> space) to do cost-function weighting and set
>> these
>>>> areas to zero
>>>>> in the weight (thus ignoring their
>> contribution to the
>>>> overall alignment).
>>>>>
>>>>> I think that there is a good chance that one
>> of these
>>>> options would work.
>>>>>
>>>>> All the best,
>>>>> Mark
>>>>>
>>>>> P.S. I wrote this before seeing Cornelius's
>> linked
>>>> images, which are
>>>>> worse than the one Sven sent wrt brain
>>>> boundaries. Hopefully
>>>>> most data looks like Sven's images, although
>> it still
>>>> might be possible
>>>>> to register the images that Cornelius
>> sent. The
>>>> outer border of the
>>>>> brain is a very good boundary for
>> registration
>>>> within-subject.
>>>>>
>>>>>
>>>>> On 16 Mar 2011, at 10:10, Cornelius Werner
>> wrote:
>>>>>
>>>>>> Dear Mark,
>>>>>>
>>>>>> that's precisely why I think this will not
>> work -
>>>> the boundaries are
>>>>>> NOT equal. In pathological DatScans the
>> volume
>>>> typically gets smaller
>>>>>> (particularly the "tail" towards the
>> occiput),
>>>> while the anatomical
>>>>>> structure stays the same. See this link
>> for an
>>>> example:
>>>>>>
>>>>>> http://www.medscape.com/viewarticle/735875
>>>>>>
>>>>>> or simply google DaTScan in an image
>> search. While
>>>> I am certainly not
>>>>>> an PET expert, I seem to remember that
>> DaTScans
>>>> are somewhat difficult
>>>>>> to analyze automatically. In our routine,
>> we get
>>>> semiquantitative
>>>>>> results by our nuclear medicine staff at
>> best.
>>>>>>
>>>>>> Cheers,
>>>>>> Cornelius
>>>>>>
>>>>>> On Wed, Mar 16, 2011 at 10:50 AM, Sven
>> Haller
>>>> <[log in to unmask]>
>>>> wrote:
>>>>>>> Dear Mark
>>>>>>>
>>>>>>> Thanks a lot
>>>>>>> Try this link. I hope that it works
>>>>>>> http://gallery.me.com/sven.haller/100063
>>>>>>>
>>>>>>> Sven
>>>>>>>
>>>>>>> On 16 mars 2011, at 10:20, Mark
>> Jenkinson
>>>> wrote:
>>>>>>>
>>>>>>>> Dear Sven,
>>>>>>>>
>>>>>>>> It will all be about whether you
>> can see
>>>> details in the images of
>>>>>>>> anatomical boundaries like in the
>> MRI or
>>>> not. Certainly the more
>>>>>>>> "different" the images are and the
>> less
>>>> details you have in them
>>>>>>>> the more it will be absolutely
>> essential
>>>> to register them to another
>>>>>>>> within-subject image (using 6 dof
>> or
>>>> similar) as non-linear registration
>>>>>>>> requires highly detailed images
>> that show
>>>> the same structures in
>>>>>>>> each.
>>>>>>>>
>>>>>>>> One approach might be (but I can't
>> say for
>>>> sure without seeing the
>>>>>>>> images) to segment the basal
>> ganglia in
>>>> the T1-weighted image and
>>>>>>>> use this as a registration target.
>>
>>>> However, this would only be appropriate
>>>>>>>> if the DATSCAN clearly showed only
>> these
>>>> boundaries and had them
>>>>>>>> close to the anatomical
>> boundaries.
>>>>>>>>
>>>>>>>> As for showing an image - can you
>> post it
>>>> somewhere on the web and
>>>>>>>> put the link to it in the
>> email. It
>>>> is not possible to attach anything but the
>>>>>>>> smallest attachments to the list
>> (to avoid
>>>> everyone's inbox getting swamped).
>>>>>>>>
>>>>>>>> All the best,
>>>>>>>>
>> Mark
>>>>>>>>
>>>>>>>>
>>>>>>>> On 16 Mar 2011, at 09:10, Sven
>> Haller
>>>> wrote:
>>>>>>>>
>>>>>>>>> Dear Mark
>>>>>>>>>
>>>>>>>>> Thank you very much for your
>> e-mail
>>>>>>>>>
>>>>>>>>> In fact the DaTScans are very
>>>> different to "normal" SPECT or PET, e.g. FDG PET.
>> In the
>>>> latter you have activity in the entire brain, so I
>> think
>>>> that normalization to a T1 (maybe after BET)
>> should be
>>>> possible (although I have no personal experience
>> here).
>>>>>>>>>
>>>>>>>>> The image of a DaTScan is
>>>> fundamentally different (see attachment). In fact
>> activity
>>>> is present only in the basal ganglia. I think that
>> therefore
>>>> the procedure to normalize the scans should be
>> very
>>>> different.
>>>>>>>>>
>>>>>>>>> I like FSL, and I would like
>> to
>>>> analyze VBM and TBSS. Therefore it would be best
>> to analyze
>>>> the DaTScan also in FSL, followed by a RANDOMISE
>> analysis
>>>>>>>>>
>>>>>>>>> Any help is highly
>> appreciated
>>>>>>>>>
>>>>>>>>> Sven
>>>>>>>>>
>>>>>>>>> PS: The image was refused (too
>> large).
>>>> Maybe I can send it to you in another way??
>>>>>>>>>
>>>>>>>>>
>>>>>>>>> On 16 mars 2011, at 09:29,
>> Mark
>>>> Jenkinson wrote:
>>>>>>>>>
>>>>>>>>>> Dear Sven,
>>>>>>>>>>
>>>>>>>>>> I have no direct
>> experience of
>>>> DATSCANs but I assume they are similar
>>>>>>>>>> to general SPECT or PET
>>>> scans. We have had success registering
>>>>>>>>>> SPECT/PET to MRI
>> before. It
>>>> is always better to register the SPECT/PET
>>>>>>>>>> to that subject's MRI
>> using 6 DOF
>>>> with FLIRT and a cost function like
>>>>>>>>>> mutualinfo or
>> normmi. I
>>>> would normally choose the best T1-weighted
>>>>>>>>>> scan as the reference, but
>> it may
>>>> depend on what features are most
>>>>>>>>>> clearly seen in your
>> DATSCAN.
>>>>>>>>>>
>>>>>>>>>> Once you've got a good
>>>> registration of your DATSCAN to your MRI,
>>>>>>>>>> then you can register the
>> MRI to
>>>> the MNI standard space image.
>>>>>>>>>> This registration can be
>> done with
>>>> non-linear (FLIRT then FNIRT)
>>>>>>>>>> whereas it is usually
>> very, very
>>>> bad to try and register the SPECT/PET
>>>>>>>>>> with non-linear directly
>> as there
>>>> are very few features there to drive
>>>>>>>>>> that registration.
>>>>>>>>>>
>>>>>>>>>> When you have the two
>>>> registrations then you can combine them with
>>>>>>>>>> convertwarp to get a
>> non-linear
>>>> registration from the DATSCAN to
>>>>>>>>>> the MNI standard space.
>>>>>>>>>>
>>>>>>>>>> I do not know what you
>> want in
>>>> terms of a "specific template of the
>>>>>>>>>> basal ganglia" but we have
>> several
>>>> atlases in FSL, and they include
>>>>>>>>>> basal ganglia
>> parcellations.
>>>>>>>>>>
>>>>>>>>>> All the best,
>>>>>>>>>>
>> Mark
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> On 16 Mar 2011, at 07:37,
>> Sven
>>>> Haller wrote:
>>>>>>>>>>
>>>>>>>>>>> Dear all
>>>>>>>>>>>
>>>>>>>>>>> I would like to
>> normalize
>>>> DATSCANs to MNI standard space
>>>>>>>>>>> I also have 3DT1 (easy
>> using
>>>> FSLVBM) and DTI (easy using TBSS).
>>>>>>>>>>>
>>>>>>>>>>> Are there any existing
>> tools
>>>> for FSL for DATSCANs?
>>>>>>>>>>> Is there a specific
>> template
>>>> of the basal ganglia?
>>>>>>>>>>> Any experience whether
>> it is
>>>> better to perform a linear registration DATSCAN to
>> DTI, and
>>>> then use the non-linear registration of TBSS, or
>> better
>>>> directly register DATSCAN to NMI? In that case,
>> how? Linear
>>>> or non-linear?
>>>>>>>>>>>
>>>>>>>>>>> Any help is highly
>>>> appreciated
>>>>>>>>>>>
>>>>>>>>>>> Sven
>>>>>>>>>>>
>>>>>>>>>
>>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>> --
>>>>>> Dr. med. Cornelius J. Werner
>>>>>> Department of Neurology
>>>>>> RWTH Aachen University
>>>>>> Pauwelsstr. 30
>>>>>> 52074 Aachen
>>>>>> Germany
>>>>>>
>>>>
>>>
>>>
>>>
>>
>
>
>
|