Dear Ming-Ting:
> -----------------------------------------
> It is important to note that the contrast images generated by SPM99 are
> already thresholded/masked (not in the statisical sense), but either based
> on the absolute or proportional actial voxel value (PET) OR based on voxels
> that are present in all the images (fMRI).Therefore these imges SHOULD NOT
> be threshodled further in the second level analysis.............
>
> >From the above statement, my got the impression that we need not do hight or
> extent threshold in the 2nd level analysis. Apparently I am not precisely
> understanding the statement. Would you please help me for this point?
Ah: probably not well written on my part. What I meant was that these images
do not need to be masked or globally normalized in any way.
>
> (2) In the paper Brain 2000; 123: 954-67 Participate of the left posterior
> inferior temporal cortex in writing and mental recall of kanji orthography.
> I am confused about the analysis method:
>
> 10 subjects, 4 tasks.
>
> a. p. 957, Data analysis. " for individual analysis, the authors used Z >
> 3.09, = p < 0.001, uncorrected for multiple comparison."
> b. p. 958,
> " For intersubject analysis, a paried t test was applied to the
> condition-specific mean images by the use of the PET routine of SPM96. The
> PET statistics thus comprised eight experimental conditions, i.e.. four for
> the active tasks and four for their respective baselines. A threshold of Z >
> 3.09 ws used to determine the presence of significant activation foci. The
> extent of clusters was corrected at p < 0.05 for multiple comparison."
>
> It is unclear to me that if the intersubject comparison had been corrected
> for multiple comparison, why the Z value is so low?
> Has it doen with small volume correction for 4.6 mkl over BA37??
I have not read that article so I cannot comment on its methods. It sounds
like the authors thresholded the maps at .001 but then relied on a cluster
threshold of .05 corrected for multiple comparisons to determine
significance. That is probably OK.
> (3). In the era of RFX analysis by SPM99, is it acceptable to present
> analysis
> with fixed effect only?
Yes it is acceptable, but you may have to modify what you are going to say.
What is commonly said is that fixed effects only tells you about the
subjects you have studied and not about the population from which those
subjects are drawn. Think of fixed effects for a group as taking an average
at each voxel across the group. Let's say you have 10 people with the
following parameter estimate values at some voxel:
[13 12 20 8 6 7 -1 -10 -4 1]. Their mean value is 5. So in our hypothetical
-analysis we might say this is significant. However, this set would fail a
1-tail one sample t-test for significance. The other issue is that a
fixed-effects analysis across multiple subjects mis-estimates the degrees of
freedom and the variance for what you are really trying to compare. If you
have 10 subjects each with 100 images, you end up with something like 700+
degrees of freedom. However, by examining a group you should not really be
combining within subjects estimates of variance, but that across subjects.
If you present a fixed effects analysis you then have to qualify your
results to say that this may not be true of how the brain functions in
general. Another way of dealing with it is to report how many subjects show
particular activations. If you define your areas of interest fairly
precisely, and count which subjects show activations at an individual level
in those regions the results may be much more convincing to your audience.
It is possible that 7/10 subjects may show some lower level activation that
is not significant with only 9 degrees of freedom, but that might be more
significant with more subjects.
You could also try using the work by Friston/Worsley/Price on conjunctions.
If a conjunction of those 10 subjects did show an activation that may be
another way to report the results.
There is, I am sure, work ongoing to address the issues of significance when
combining imaging information across groups. For now you can try some of the
techniques above though others may want to comment additionally.
Hope this helps,
Darren
-------------------------------------------------------------------
Darren R. Gitelman, M.D. E-mail: [log in to unmask]
Cognitive Neurology and Voice: (312) 908-9023
the Alzheimer's Disease Center Fax: (312) 908-8789
WWW: http://www.brain.nwu.edu
Northwestern Univ., 320 E. Superior St., Searle 11-470, Chicago, IL 60611
-------------------------------------------------------------------
> Thank you for your help in advance.
>
>
> Ming-Ting
>
>
>
>
>
>
> ----- Original Message -----
> From: "Darren Gitelman (FIL)" <[log in to unmask]>
> To: <[log in to unmask]>
> Sent: Wednesday, December 13, 2000 7:36 AM
> Subject: Re: threshold for randome effect analysis
>
>
> Dear Ming-Ting:
>
> Correction for multiple comparisons is still required in random effects
> analyses unless you have somehow restricted the number of voxels. Thus
> unless you have an a priori hypothesis about one particular voxel (based on
> some other study, not the fixed effects analysis) a threshold of 1.75 would
> not, in general, be adequate. The same threshold apply for random as for
> fixed effects.
>
> Doubtless you are noticing the severe penalty to your p-values that random
> effects imposes. This is due in large part to the reduction in degrees of
> freedom (10 subjects is barely adequate).
>
> You can work around this in several ways
> 1) get more subjects (double the number)
> 2) use conjunction analyses as implemented in spm99. this does not say the
> same thing statistically as RanFX but does allow you to say that all
> subjects activated at a particular voxel based on your threshold.
> 3) Since random effects is a grouped analysis, perhaps the data should be
> smoothed more. As an example, if you had looked at the individual analyses
> following 6-7 mm of smoothing you might find that smoothing at 10 mm
> produces more robust maps for the group effects. However, if you try
> several different smoothing kernels then you should report that and correct
> your statistics for the repeated trials.
> 4) If you have a priori hypotheses from other studies from your group or
> the literature this may allow you to perform a small volume correction.
>
> try searching the spm help list for other suggestions.
> Hope this helps,
> Darren
>
> At 06:59 AM 12/13/00 +0800, you wrote:
> >Dear SPM group:
> >Please allow me ask my question again since there is no one answered my
> >previous mail.
> >
> >I am doing random effect analysis on 10 subjects. To my understanding, we
> >don't need to correction for multiple comparison. Is this correct?
> >In this case, is a threshold of p < 0.05 (height threshold T = 1.75)
> without
> >voxel extent threshold can be acceptable for statistical report?
> >Any advice will be deeply appreciated.
> >
> >
> >Sincerely
> >
> >Ming-Ting
> >
> >===================================================================
> >Ming-Ting Wu, MD. Department of Radiology
> >386 Ta-chung 1st Rd.
> >Kaohsiung Veterans General Hospital, Kaohsiung, Taiwan, 813
> >Tel: 886-7-3422121 ext 6205, 6235
> >Fax: 886-7-3468301
> >mobile: 0920040712
> >----- Original Message -----
> >From: "Karsten Specht" <[log in to unmask]>
> >To: <[log in to unmask]>
> >Sent: Tuesday, December 12, 2000 8:08 PM
> >Subject: AW: displaying activation on individual anatomicals
> >
> >
> > > Dear Charvy,
> > >
> > > you are on the right way!
> > > There are several possibilities to display the statistical results onto
> >the
> > > individual brain. First, you can select under the 'Result' Section the
> > > option 'overlays' and you will asked for an image. Here you can select
> the
> > > individual, normalized image.
> > > An other possibility is within the 'Display' Section. Here you can
> select
> > > the option 'Add blobs'. Within this option, you can display more than
> one
> > > contrast in different colours.
> > > To get an individual rendered brain, you should select first the
> 'Segment'
> > > option, then 'Render'-'Xtract Brain' and you will get an individual
> >rendered
> > > brain.
> > >
> > > Good luck
> > >
> > > Karsten
> > >
> > > ----------------------------------
> > > Dipl. Phys. Karsten Specht
> > >
> > > Medizin Center Bonn
> > > Spessartstrasse 9
> > > 53119 Bonn
> > > Germany
> > >
> > > Phone: ++49-(0)228/90 81-178
> > > Fax: ++49-(0)228/90 81-190
> > > E-Mail: [log in to unmask]
> > > WWW: http://www.mcbonn.de
> > >
> > > > -----Ursprüngliche Nachricht-----
> > > > Von: SPM (Statistical Parametric Mapping) refers to functional brain
> > > > mapping ana [mailto:[log in to unmask]]Im Auftrag von Charvy Narain
> > > > Gesendet: Dienstag, 12. Dezember 2000 12:47
> > > > An: [log in to unmask]
> > > > Betreff: displaying activation on individual anatomicals
> > > >
> > > >
> > > > Dear SPM,
> > > >
> > > > Is there any way in spm to display contrasts onto individual subjects'
> > > > anatomicals, rather than on the MNI template (preferably as a colour
> > > > rendered image, but if this is not possible, then just as a
> > > > superimposition)?
> > > >
> > > > The individual anatomicals were normalised into MNI space using the
> > > > following procedure:
> > > >
> > > > - coregister the anatomicals to the mean functional image (coregister
> > > > only: no reslicing)
> > > > - normalise the structural image into MNI space using the *sn3d.mat
> file
> > > > generated from normalising the mean functional to the EPI template.
> > > >
> > > > Is this correct? And if so, what is the next step?
> > > >
> > > > An earlier reply by John Ashburner to a similar question recommended
> > > > displaying the background image, and then superimposing 'blobs' onto
> it
> > > > by using an undocumented feature of spm99b
> > > > (http://www.jiscmail.ac.uk/cgi-bin/wa.exe?S2=spm&q=displaying+acti
> > > > vations&s=&f=&a=&b=.
> > > > However, when I tried this,it didn't work too well. Any ideas would be
> > > > very welcome!
> > > >
> > > > Thanks,
> > > > Charvy
> > > >
> > > >
>
> -------------------------------------------------------------------------
> Darren R. Gitelman, M.D.
> Cognitive Neurology and Alzheimer¹s Disease Center
> E-mail: [log in to unmask]
> WWW: http://www.brain.northwestern.edus
> Voice: (312) 908-9023
> Fax: (312) 908-8789
> Northwestern Univ., 320 E. Superior St., Searle 11-470, Chicago, IL 60611
> -------------------------------------------------------------------------
>
>
>
|