[MINC-users] t threshold for corrected p value of Aston's method

Tue May 6 09:00:55 EDT 2008

Hello,

Greg Ward's scripts will work under linux with some fiddling - they are 
written in perl after all but they also make system calls to a wide 
variety of tools! I have used them under linux but they will 
require more work before making them available for mass comsumption. The 
bulk of the work resides in rewriting the code used for the creation of 
verification images since the current versions use an Irix image toolbox 
based on Irix's ImageVision Library. Also the code using batch needs 
revamping!

So use do_mritotal -noverify -resample all -background ... and verify the 
images in the mrital_rsl directory.

Better yet, do the following:

qbatch do_mritotal -noverify -resample all

qbatch is available on all linux systems and I hopefully fixed all 
the bugs ... but do_mritotal is NOT yet available - I suppose a version 
could be made available with a few options disabled to prevent havoc. I'll 
see what I can do in the next couple of weeks.

do_mritopet needs to be revamped and I don't use do_pettopet anymore as 
mritopet is more robust - hence the reason why do_mritopet needs 
revamping.

Yorick's days are counted and it will be retired in 2008 so I suspect I 
will be tempted at some point to do the work, same goes for dot ... but 
time is a luxury I don't have anymore since I'm a fulltime sys admin.

As for PET motion correction, another option might be available, ie using 
Anthonin Reilhac's Dynamic PET motion correction method. It would have to 
be adapted for static PET images by pooling together a 12-(static)-scan 
study for example and then considering individual scan as time frames. I 
gave that a try last year but the results weren't very good and I failed 
to find time to get back to it. Anthonin seemed confident that it could 
work so there's still some hope.

BUT concerning Ji Hyum KO's data - it is Dynamic PET, so Anthonin's method 
should be preferred ! Under BIC linux systems, you will find
/usr/local/bin/do_PET_frame_realignment to do the job.

As for Keith's fmristat (for PET), I never got around to using/testing it. 
It uses a different approach to estimate the variance - Keith calls it a 
"local form of pooling" and he posted something on the bic-users mailing 
list many years ago ... here it is:

>       Well I must admit PET scanners have changed in the last 10 years 
> and the assumption of equal sd probably needs to be checked. In the mean 
> time, fmristat's multistat program does (almost) everything that DOT's 
> statistical analysis does, but better. Instead of pooling the sd, it 
> smoothes it, i.e. 
> a sort of local pooling. We usually choose the amount of smoothing so 
> that the resulting df is large, say 100. If you set the FWHM of the 
> smoother very large (say 100mm), you get the pooled sd (almost); if you 
> set it to 0 you get the voxel sd.
> I would suggest you look into this - the help of multistat (look under
> INPUT_FILES_SDEFFECT) tells you how to run it for PET data - essentially 
> all you do is leave the input sd files blank -
> 
> Keith Worsley
> Department of Mathematics and Statistics, McGill University
> 805 ouest, rue Sherbrooke, Montreal, Quebec, Canada H3A 2K6
> tel: (514) 398-3842, fax: (514) 398-3899
> e-mail: keith.worsley at mcgill.ca, web: http://www.math.mcgill.ca/keith

HTH,

S

On Mon, 5 May 2008, EJ Nikelski wrote:

> Hi all,
>
>    Reading this e-mail reminded me that I could use some advise on a
> related issue.  Pls forgive me if its a wee bit tangential to the
> subject line.
>
>   Specifically, given that Keith's fmristat (for PET) assumes that
> the input volumes have already been spatially normalized, do we have
> any tools to help with this task, short of doing the minctracc (or
> mritoself, mritotal) calls manually?   In the days of 'dot' (ah, the
> good old days), Greg Ward's scripts (do_mritopet,  do_mritotal,
> do_pettopet) were nicely able to generate the required xfms  --
> however, while still available on Yorick, they do not appear to be
> migrating to the Linux systems.  I suppose that one could use
> fmr_preprocess to do PET motion correction (with some fiddling),
> however, the version of this script that I have seen is buggy (at
> least the version that I found has bugs), and it does not appear to be
> quality controlled (i.e. no official release).
>
>   So in sum, the questions are (1) are the old PET preprocessing
> tools (Greg's stuff)  unsupported and therefore not
> being migrated off of Yorick, (2) and what are other BIC PET
> researchers using to do motion correction, cross-modality alignment,
> and stereotactic transformation? Ideas and suggestions?
>
> Thanks,
>
>
> -Jim
>
>
>
> On Mon, May 5, 2008 at 6:58 PM, Andrew Janke <a.janke at gmail.com> wrote:
>> Hi Ji,
>>
>>  I was hoping Sylvain would answer this one as I am not that familiar
>>  with the stats associated with PET but in any case here goes...
>>
>>
>>  2008/5/1 Ji Hyum KO <kojinet at bic.mni.mcgill.ca>:
>> >  As far as I understand, I should calculate the t threshold based on
>> >  Worsely et al., Human Brain Mapping, 1996.
>> >  I am really sure that it is very well-written famous paper, but I
>> >  couldn't figure out how to calculate the t-threshold in the given
>> >  conditions.
>>
>>  Correct, to calculate this I would suggest that you use Keith Worsleys
>>  MATLAB package (called fmristat) to do this.
>>
>>    http://www.math.mcgill.ca/keith/fmristat/
>>
>> >  Is there any easy way to calculate t threshold for p<0.05 (corrected)?
>>
>>  This will depend on the number of subjects, (and thus Degrees of
>>  Freedom), the number of voxels you are considering and so forth.
>>
>> >  And, is it going to be the same even if I consider the whole brain rather
>> >  than just striatum, e.g., fallypride study?
>>
>>  It will be different, but again this will depend on the method you use
>>  to calculate the t-value. Certainly it will change if you calculate
>>  pooled variance across you data (or mask).
>>
>>  Again though I would suggest that you spend some time getting
>>  acquainted with the fmristat package. Keith has a number of examples
>>  on his webpages that should help you get going.
>>
>>
>>  a
>>
>>  --
>>  Andrew Janke (a.janke at gmail.com || http://a.janke.googlepages.com/)
>>  Canberra->Australia +61 (402) 700 883
>>  _______________________________________________
>>  MINC-users at bic.mni.mcgill.ca
>>  http://www2.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
> Tel: (514) 340-8222 x 2298
> Fax: (514) 340-8295
> _______________________________________________
> MINC-users at bic.mni.mcgill.ca
> http://www2.bic.mni.mcgill.ca/mailman/listinfo/minc-users
>

---
Sylvain Milot (sylvain at bic.mni.mcgill.ca)
               (bicadmin at bic.mni.mcgill.ca)
Brain Imaging Centre
Montreal Neurological Institute
3801 University Street
Webster 2B, Room 212
Montreal, Qc., Canada, H3A 2B4
Phone  : (514) 398-4965, Fax: 398-8948
Mobile : (514) 712-1768
Office : 527 Av Des Pins O., Room 204
          Montreal, Qc., H2W 1S4