[MINC-users] 6-parameter fitting with minctracc

[EJ Nikelski]

Hi all,

   Jason, the problem that I'm trying to address is the same problem
that you mentioned in your MACACC paper -- poor gray/white tissue
contrast in the dorsal motor and somatosensory cortex. As the
thickness values are ... ummmm .... of questionable value, processing
them separately won't really help.  I need to increase my tissue
contrast in that region in order to extract anything that looks like
reasonable thickness values (reasonable = motor cortex almost twice as
thick as the somatosensory).

  Vlad and Jason (and others) :  So in the above scenario, you would
recommend non-linear fitting?  Would this be (1) fitting both scans to
a model (like the ICBM symmetrical), or (2) nl-fitting a given
subject's scan #1 against their scan #2?  I have to admit that using a
non-linear fit when fitting a given subject's scan to his/her own scan
-- possibly acquired during the same scanning session -- seems
counter-intuitive. I would have thought that rigid-body would be the
way to go ...

Jason: Yes, please send me your nonlinear-mritoself perl script ...
it's certainly worth a shot.

Vlad: Do you mean the "-normalize" switch on mincaverage, or something
more diabolical? For my initial tests, both scans were acquired
back-to-back, during the same scanning session, so I didn't do
intensity normalization.  Once I'm able to get the 2 within-scan scans
to align properly, I plan to add in the additional scans ... and use
intensity normalization during averaging.  BTW, as these are ADNI
scans, they've also already had one run through of N3 (off topic, but
related).

Any additional ideas would certainly be welcome ... and I'd be willing
to test all reasonable suggestions (I have 5 test subjects that I'm
currently using).

Thanks,

-Jim


On Sat, Feb 20, 2010 at 11:13 AM, Vladimir Fonov
<vladimir.fonov at gmail.com> wrote:
> Hello,
>
> I also suggest using nonlinear registration ( 8mm step size should be
> enough) or at least lsq12 when averaging scans acquired during
> different scanning sessions. Also, do you normalize intensities
> between scans ?
>
>
> On Sat, Feb 20, 2010 at 12:56 PM, Jason Lerch <jason at bic.mni.mcgill.ca> wrote:
>> In my limited experience you're much better off processing each of the scans independently and then either averaging the thickness maps at the end or, even better, using mixed effects models and all your data for the analysis.
>>
>> Alternately, if you really want to go the higher SNR route, I have a nonlinear-mritoself perl script that might help.
>
>> On 2010-02-20, at 12:36 PM, EJ Nikelski wrote:
>>>  I need some validation and/or suggestions (or both?).  I have
>>> subject scans (T1) for which each subject was scanned multiple times
>>> (same scanner), at difference points in time.  For each subject, some
>>> of the scans were acquired within the same scanning sessions, some
>>> were acquired within a 6-month window.
>>>
>>>     I would like to combine the scans to give me the best SNR for
>>> Civet processing.  I've tried specifying one of the scans as the
>>> "target" and then using "mritoself" to align all others to it; all
>>> aligned scans and the "target" would then be mincaverage'd.  The
>>> results are OK, but not splendid, as Civet spits up during tissue
>>> classification on one of the averaged volumes, although the
>>> non-averaged scans go through without error.  I'm using defaults for
>>> the mritoself call, plus "-nocrop".
>
>
>
>
> --
> Best regards,
>  Vladimir S. Fonov ~ v.s.fonov <@> ilmarin.info
> _______________________________________________
> MINC-users at bic.mni.mcgill.ca
> http://www.bic.mni.mcgill.ca/mailman/listinfo/minc-users
>



-- 
=================================
Jim Nikelski, Ph.D.
Postdoctoral Research Fellow
Bloomfield Centre for Research in Aging
Lady Davis Institute for Medical Research
Sir Mortimer B. Davis - Jewish General Hospital
McGill University