Hi Casper,

Yes: With Caret, the same test/command that creates the *real* t- or 
f-map also creates the randomized ones. The most typical permutation 
method is on the group, but in the case of a paired t-test, for example, 
it's the sign of some randomly determined subjects/columns that gets 
sign-flipped.

Whatever test/statistic you compute for the real t- or f-map gets 
computed for the randomly split groups.

For a cluster test, the user-specified threshold is applied (both 
positive and negative, usually) to both the real and randomized t- or 
f-maps. The size of the largest cluster in each randomized map is stored 
and a distribution of max clusters generated. The 95th percentile is 
typically the significance cut-off.

Note, too, that you can get around the need for group or other 
permutation by using something like False Discovery Rate (FDR), which 
I'd guess AFNI/SUMA can compute for 1D files. FDR is different from FWER 
(family-wise error rate, which is what the cluster method gives you), 
and in my very limited experience it has not proven more sensitive 
(possibly because high-res surface data is smoother, which makes FDR 
more conservative?). But I thought it worth mentioning, since it has the 
benefit of not needing permuations. ;-)

Donna

On 11/19/2009 09:19 AM, Oers, C.A.M. van wrote:
> Hi Donna, 
>
> Thanks for the explanation.
>
> So you are saying that Caret cannot create these permuted f-maps, since
> the statistical model needs to be applied to create the f-map, but also
> to create the permuted f-maps. I would have to create them myself using
> the statistical model, right? 
> How are the permuted f-maps created? What permutation is used?
>
> Thanks again, 
> Casper
>   
>
> -----Oorspronkelijk bericht-----
> Van: [email protected]
> [mailto:[email protected]] Namens Donna Dierker
> Verzonden: maandag 16 november 2009 17:00
> Aan: Caret, SureFit, and SuMS software users
> Onderwerp: Re: [caret-users] multiple comparison correction on SUMA data
>
> On 11/16/2009 05:27 AM, Oers, C.A.M. van wrote:
>   
>> Hi,
>>  
>> Although I saw a post by Donna Dierker regarding the subject, I am 
>> afraid I still have a question.
>>  
>> I have created SUMA files using a mixed model analysis. The resulting 
>> files have an individual statistical threshold (and coefficient). I 
>> would like to use Caret to do a multiple comparison correction with a 
>> cluster size threshold.
>>     
> I'm not sure how straightforward this will be, due to the randomization
> component.  The caret_command utility has these features:
>
>       caret_command -metric-statistics-anova-one-way
>       caret_command -metric-statistics-anova-two-way
>       caret_command -metric-statistics-coordinate-difference
>       caret_command -metric-statistics-interhemispheric-clusters
>       caret_command -metric-statistics-kruskal-wallis
>       caret_command -metric-statistics-levene-map
>       caret_command -metric-statistics-normalization
>       caret_command -metric-statistics-one-sample-t-test
>       caret_command -metric-statistics-paired-t-test
>       caret_command -metric-statistics-shuffled-cross-correlation-map
>       caret_command -metric-statistics-shuffled-t-map
>       caret_command -metric-statistics-subtract-group-average
>       caret_command -metric-statistics-t-map
>       caret_command -metric-statistics-two-sample-t-test
>       caret_command -metric-statistics-z-map
>
> Enter the command for the help for that feature, or enter "caret_command
> -help-ful > caret_command.txt" to get a text file with all the help.
>
> Let's choose the case of a two-sample-t-test.  In this case, your
> caret_command line will look like so:
>
> DISTCOL=1
> ITERATIONS=2500
> THRESHNEG=-2.68
> THRESHPOS=2.68
> P_VALUE=0.05
> VAR_SMOOTH_ITERATIONS=0
> VAR_SMOOTH_STRENGTH=0.0
> DO_TMAP_DOF=true
> DO_TMAP_PVALUE=true
> THREADS=4
>
> FIDUCIAL=Human.PALS_B12.LEFT_AVG_B1-12.FIDUCIAL.clean.align.73730.coord
> OPENTOPO=Human.sphere_6.LEFT_HEM_OPEN.73730.topo
> DISTMETRIC=Human.PALS_B12.B1-12_LEFT_DISTORTION-vs-AVG-FIDUCIAL_ONLY.737
> 30.metric
> COMPOSITE_IN_A=Composite.DEPTH.LEFT.KIDS.73730.surface_shape
> COMPOSITE_IN_B=Composite.DEPTH.LEFT.ADULTS.73730.surface_shape
> OUT_PREFIX=KIDS29vADULTS23_Depth.LEFT
> COMMAND="caret_command -metric-statistics-two-sample-t-test NO_TRANSFORM
> UNPOOLED"
> COMMAND="$COMMAND $COMPOSITE_IN_A $COMPOSITE_IN_B"
> COMMAND="$COMMAND $FIDUCIAL $OPENTOPO $DISTMETRIC $DISTCOL $OUT_PREFIX"
> COMMAND="$COMMAND $ITERATIONS $THRESHNEG $THRESHPOS $P_VALUE"
> COMMAND="$COMMAND $VAR_SMOOTH_ITERATIONS $VAR_SMOOTH_STRENGTH"
> COMMAND="$COMMAND $DO_TMAP_DOF $DO_TMAP_PVALUE $THREADS"
> $COMMAND
>
> These sample scripts (e.g., tstt.sh and gen_composites.sh) are here:
>
> http://brainmap.wustl.edu/pub/donna/WUSTL/BURTON/SCRIPTS/SBM/
> login pub
> password download
>
> But note that this takes two composite shape files (imagine your 1D
> files concatenated together in one file, with one 1D column per subject,
> with the node number preceding all os them).  I don't think generating
> composites is hard (see below about adding header).
>
> Rather, the tricky bit here is that you already have your stat -- not
> raw scores/activations.  The example above uses sulcal depth scores, but
> this could be any measure.  But if it is already a stat, then it's less
> clear to me how to randomize it.  With the two-sample t-test, a t-map is
> generated using the two groups' composites.  Then, the composites are
> combined and group memberships randomized to create permuted t-maps.  
> The max size of each permuted map is stored to create a distribution,
> and the size of the 95th percentile is used as the significance cut-off.
>
> If you've got stats already, how are we randomizing?
>
> If you can create your own permuted t- or f-maps, as well as your own
> real t- or f-map, then there is a newer caret_stats feature that can
> take these real and permuted maps and do either cluster operations or
> TFCE on them.
>
> But in your case, it isn't clear to me where/how the randomizing comes
> in.
>
>   
>>  
>> How can I convert a 1D surface file into a metric or surface_shape
>>     
> file?
> This part is easy:
>
> 1.  Add a node number in column 1, using matlab or some other tool.
>
> 2.  Slap on a header as shown in
> http://brainvis.wustl.edu/CaretHelpAccount/caret5_help/file_formats/file
> _formats.html#metricFile:
>
> metric-version 2
> tag-number-of-nodes 71723
> tag-number-of-columns 2
> tag-title untitled
> tag-column-name 0 Depth
> tag-column-name 1 Smoothed Depth
> tag-column-color-mapping 0 -1.000000 1.000000 tag-BEGIN-DATA 0 1.803019
> -0.045549
> 1 1.704132 -0.007309
> 2 1.523951 0.026103
> ...
> 71722 .822899  0.112485
>
> If you can generate metric/surface_shape files from all your 1D files,
> then it's not hard to generate a composite using something like the
> gen_composites.sh script mentioned above).
>   
>> Which way to perform the correction?
>>  
>> I appreciate the help,
>> Casper
>> ----------------------------------------------------------------------
>> --
>>
>> /De informatie opgenomen in dit bericht kan vertrouwelijk zijn en is 
>> uitsluitend bestemd voor de geadresseerde. Indien u dit bericht 
>> onterecht ontvangt, wordt u verzocht de inhoud niet te gebruiken en de
>>     
>
>   
>> afzender direct te informeren door het bericht te retourneren. Het 
>> Universitair Medisch Centrum Utrecht is een publiekrechtelijke 
>> rechtspersoon in de zin van de W.H.W. (Wet Hoger Onderwijs en 
>> Wetenschappelijk Onderzoek) en staat geregistreerd bij de Kamer van 
>> Koophandel voor Midden-Nederland onder nr. 30244197. /
>>
>> / Denk s.v.p aan het milieu voor u deze e-mail afdrukt. /
>>
>> ----------------------------------------------------------------------
>> --
>>
>> /This message may contain confidential information and is intended 
>> exclusively for the addressee. If you receive this message 
>> unintentionally, please do not use the contents but notify the sender 
>> immediately by return e-mail. University Medical Center Utrecht is a 
>> legal person by public law and is registered at the Chamber of 
>> Commerce for Midden-Nederland under no. 30244197. /
>>
>> / Please consider the environment before printing this e-mail. /
>>
>> ----------------------------------------------------------------------
>> --
>>
>> _______________________________________________
>> caret-users mailing list
>> [email protected]
>> http://brainvis.wustl.edu/mailman/listinfo/caret-users
>>   
>>     
>
> _______________________________________________
> caret-users mailing list
> [email protected]
> http://brainvis.wustl.edu/mailman/listinfo/caret-users
>
> _______________________________________________
> caret-users mailing list
> [email protected]
> http://brainvis.wustl.edu/mailman/listinfo/caret-users
>   

_______________________________________________
caret-users mailing list
[email protected]
http://brainvis.wustl.edu/mailman/listinfo/caret-users

Reply via email to