Hi Antonin,
This makes a good sense now. Now, my results are making more sense
with clustere_tstat_fwep.mgz,
where the cluster is very close to higher PCC values.
Thank you so very much Antonin all your help, patience and time in figuring
out all these issues :).
Thanks,
Sahil
On Fri, Mar 10,
Dear Sahil,
-approx tail and -twotail are completely different things:
-approx tail switch on the tail approximation which speeds up the permutations.
-twotail switch on two-sided hypothesis, without -twotail the hypothesis is
only one-sided.
It is reassuring that you see some clusters in
Dear Sahil,
thinking of that, the comparisons I suggested in previous mail are probably
valid only for the case of orthogonal design. And even with the orthogonal
design as Anderson said, the values can a bit differ:
Dear Sahil,
I would suggest at first to make check of your GLM PALM setup by comparing
mri_glmfit and PALM output files (using the same design and contrast):
1. values of gamma.mgh should correspond to values of *cope.mgh
2. values of F.mgh should be (*dpv_tstat.mgh) ^2.
3. values of sig.mgh
Dear Sahil,
to assure that there is no other issue with your setup, I would recommend to
obtain cluster-wise p-pvalue of that big cluster to see if it is reasonable,
i.e. if it is somewhat close to the significance.
Therefore, I would load *_clustere_tstat_fwep and click to the area of big
Dear Antonin,
Setting minimum to 1.3 for *_clustere_tstat_fwep doesn't show any
significant cluster and similarly thresholded map *dpv_tstat.mgz also
doesn't show anything at -log10(0.05) = 1.3. As you said, it seems like
even though there is big cluster but after FWER correction, the
Dear Sahil,
If you used -logp as Anderson suggested, you should set your min to 1.3 to
threshold your *_clustere_tstat_fwep map and see the clusters.
What is the value of *_clustere_tstat_fwep in the region of the big cluster
seen at thresholded map *dpv_tstat.mgz ? This should correspond to
Dear Sahil,
could you send the full command-line and unthresholded view of
*_clustere_tstat_fwep ?
How the data showed in screenshot 1 were produced?
How are the actual p-values of your clusters in *_clustere_tstat_fwep?
You can also use -saveglm and inspect the files containing values of GLM
Hi all,
That's exactly as Antonin says -- I have very little to add :-)
Only a few suggestions:
- With surfaces, both cluster and TFCE statistics tend to be slow. Consider
using the tail approximation ("-approx tail -n 500 -nouncorrected")
- Include -logp, so that the p-values are in log-10
Dear Sahil,
I suppose, for qcache 1.3 the equivalent cluster-forming threshold z-value is
two-tailed test:
qnorm(1-10^-1.3/2)=1.958949
for one-tailed test:
qnorm(1-10^-1.3)=1.643704
(qnorm is R function call for quantile function of normal distribution, you can
compute this by using other
Dear Martin,
thinking of that further, my advice for negative cluster-forming threshold was
probably wrong in this context. Instead, I think that you can achieve negative
one-side hypothesis by specifying positive cluster-forming threshold and
inverting the sign of your contrast vector.
Hello Martin and Antonin,
I was following this conversation very closely to understand how to use
PALM in FreeSurfer.
Can any of you please confirm in case I am interested in checking
correlation between gyrification index (LGI) and behavioral measure using
two tailed, p < 0.05:
Step 1: I used
Dear Martin,
after -s option, there have to be 2 arguments, as I specified in my previous
mail:
-s fsaverage/surf/lh.white fsaverage/surf/lh.white.avg.area.mgh
And beware that -C has to have negative sign, if your hypothesis is one-tailed
negative.
Antonin
Hi Antonin,
Thank you so much
Hi Antonin,
Thank you so much for this detailed explanation, that's really useful.
Following your instructions, I ran:
palm -i lh.MEQ_LGI.10.mgh -s fsaverage/surf/lh.white.avg.area.mgh -d
check.csv -t Contrast_MEQ.csv -n 5000 -m lh.MEQ_LGI.glmdir/mask.mgh -o
myresults -Cstat extent -C 3.719016
Dear Martin,
input -i input file is
lh.MEQ_LGI.10.mgh file in your glmdir directory (for left hemisphere).
As you could read in following messages in the referenced thread in FSL
discussion forum, cluster-forming threshold need to be specified in z, not in t.
Therefore, you would have to
Hi Antonin,
As you suggested in discussion forum, I tried to run following command
after mri_glmfit:
palm -s fsaverage/surf/lh.white -n 1 -m mask.mgh -Cstat extent -C
1.974975 -pmethodp none -pmethodr none -twotail -d Design_MEQ.txt -t
Contrast_MEQ.txt
Running PALM alpha104 using MATLAB
Hi Antonin,
Thanks a lot for your reply.
Somehow, in the link you sent, I could not find any response to your email.
But I can see your email to Anderson and command line parameters.
As I am not an expert in using FreeSurfer, so would it be possible for you
to share detailed step-by-step guide
Dear Martin,
I think yes, you can use PALM with FreeSurfer surfaces, see my conversation
with Anderson on FSL list:
https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=ind1604=FSL=0&1=FSL&9=A=on=No+Match%3BMatch%3BMatches=4=239088
but beware not to forget to include average the vertex area
If you don't have an orthogonal design, then you can't use
mri_glmfit-sim. I think you can use PALM:
https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/PALM
I have not tried it yet.
Anderson, can you use PALM with surfaces?
On 03/06/2017 05:23 PM, Martin Juneja wrote:
> Hi Dr. Greve,
>
> I tried to
Hi Dr. Greve,
I tried to run: mri_glmfit-sim --glmdir lh.MEQ_LGI.glmdir --sim perm 1000 3
permcsd --sim-sign abs --cwpvalthresh .05
It gives error that ERROR: design matrix is not orthogonal, cannot be used
with permutation.
But when I run: mri_glmfit-sim --glmdir lh.MEQ_LGI.glmdir --sim perm
This is a problem with using LGI in that it is already extremely smooth
that the smoothness exceeds the limit of the look up table that we
supply. I recommend that you not use a gaussian-based correction for
LGI. Instead, use permutation (see mri_glmfit-sim --help).
On 03/06/2017 01:36 PM,
Hello everyone,
I am trying to extract clusters showing significant correlation between LGI
and a behavioral measure. I am able to extract PCC and sig.mgh but at the
last step when I try to run simulation command to view corrected results
and I run:
mri_glmfit-sim --glmdir lh.MEQ_LGI.glmdir
22 matches
Mail list logo
