Hi Doug,
I am still trying to resolve the different results for GLMs completed in
Freesurfer and SPSS. Interestingly, I get the same p-values for each
method when I remove my continuous covariate (age) from the model
(leaving gender, group, and genderXgroup effects). This suggests that
the two methods are treating covariates in a different manner.
Unfortunately, SPSS does allow you to print out the GLM design matrix or
use the design matrix from mri_glmfit as an input. However, I am able to
print out the contrast matrices for the analysis, which look like this:
Contrast Intercept Age [Gender=Female] [Gender=Male]
[diagnosis=Case] [diagnosis=Control] [diagnosis=Case] *
[Gender=Female] [diagnosis=Case] * [Gender=Male] [diagnosis=Control] *
[Gender=Female] [diagnosis=Control] * [Gender=Male]
Intercept 1 0 0.5 0.5 0.5 0.5 0.25 0.25
0.25 0.25
Age 0 1 0 0 0 0 0 0 0 0
gender 0 0 1 -1 0 0 0.5 -0.5 0.5
-0.5
diagnosis 0 0 0 0 1 -1 0.5 0.5
-0.5 -0.5
gender*diagnosis 0 0 0 0 0 0 1
-1 -1 1
While for mri_glmfit, the contrast vectors look like this:
Intercept Intercept Intercept Intercept Age Slope Age Slope Age
Slope Age Slope
contrast Male-Case Female-Case Male-Control Female-Control
Male-Case Female-Case Male-Control Female-Control
age 0 0 0 0 0.5 0.5 0.5 0.5
gender 0.5 -0.5 0.5 -0.5 0 0 0 0
group 0.5 0.5 -0.5 -0.5 0 0 0 0
gender*group 0.5 -0.5 -0.5 0.5 0 0 0 0
From this information, do you know if there is a way to work out why
the two methods are producing different results when a continuous
covariate is included in the model? I am fairly stumped and would great
appreciate any further insight you can provide.
Compared to
Kind regards,
Bronwyn Overs
Research Assistant
Neuroscience Research Australia
Neuroscience Research Australia
Margarete Ainsworth Building
Barker Street Randwick Sydney NSW 2031 Australia
*M* 0411 308 769 *T* +61 2 9399 1883 *F* +61 2 9399 1265
neura.edu.au <http://neura.edu.au>
Follow @neuraustralia on twitter
<https://twitter.com/neuraustralia>Follow NeuRA on facebook
<https://www.facebook.com/NeuroscienceResearchAustralia>Subscribe to the
NeuRA Magazine <http://www.neura.edu.au/help-research/subscribe>
On 9/02/2015 9:47 am, Bronwyn Overs wrote:
Thanks.
Kind regards,
Bronwyn Overs
Research Assistant
Neuroscience Research Australia
Neuroscience Research Australia
Margarete Ainsworth Building
Barker Street Randwick Sydney NSW 2031 Australia
*M* 0411 308 769 *T* +61 2 9399 1883 *F* +61 2 9399 1265
neura.edu.au <http://neura.edu.au>
Follow @neuraustralia on twitter
<https://twitter.com/neuraustralia>Follow NeuRA on facebook
<https://www.facebook.com/NeuroscienceResearchAustralia>Subscribe to
the NeuRA Magazine <http://www.neura.edu.au/help-research/subscribe>
On 7/02/2015 3:16 am, Douglas N Greve wrote:
It should be 10^-abs(sig)
On 02/06/2015 12:53 AM, Bronwyn Overs wrote:
Hi Doug,
Thanks for all of your help. I am investigating the differences
between the design matrices now.
I have one more query about the sig.table.dat file put out by the GLM
using the parcellated ROIs. When I transform each of the values in
this file to p values using 10^"value", some of the resulting p-value
are >1. Do you know why this would be happening? Here is an example of
one of my sig.table.dat files, where 10^.799 = 6.295:
lh.aparc.thickness me_gender_ageRem me_group_ageRem
lh_bankssts_thickness 0.107 -2.752
lh_caudalanteriorcingulate_thickness -2.616 -0.190
lh_caudalmiddlefrontal_thickness -0.701 -4.258
lh_cuneus_thickness 0.799 -1.178
lh_entorhinal_thickness 1.669 -4.129
lh_fusiform_thickness -0.088 -6.808
lh_inferiorparietal_thickness -0.665 -1.477
lh_inferiortemporal_thickness 0.149 -7.985
lh_isthmuscingulate_thickness -0.476 -2.393
lh_lateraloccipital_thickness 0.212 -1.189
lh_lateralorbitofrontal_thickness 0.288 -7.657
lh_lingual_thickness 1.148 -1.594
lh_medialorbitofrontal_thickness 1.405 -4.461
lh_middletemporal_thickness 0.727 -7.215
lh_parahippocampal_thickness -1.059 -2.854
lh_paracentral_thickness -0.514 -0.282
lh_parsopercularis_thickness 0.444 -3.541
lh_parsorbitalis_thickness -0.110 -7.075
lh_parstriangularis_thickness 0.244 -4.769
lh_pericalcarine_thickness 0.376 -0.218
lh_postcentral_thickness -0.485 -0.832
lh_posteriorcingulate_thickness -0.135 -1.241
lh_precentral_thickness 0.196 -2.102
lh_precuneus_thickness 0.018 -1.361
lh_rostralanteriorcingulate_thickness -1.208 -2.073
lh_rostralmiddlefrontal_thickness 0.437 -4.470
lh_superiorfrontal_thickness -0.141 -2.743
lh_superiorparietal_thickness -0.288 -0.365
lh_superiortemporal_thickness 0.227 -4.646
lh_supramarginal_thickness -1.185 -1.354
lh_frontalpole_thickness -0.325 -0.521
lh_temporalpole_thickness -0.067 -4.074
lh_transversetemporal_thickness -0.009 -1.618
lh_insula_thickness 0.553 -6.063
Kind regards,
Bronwyn Overs
Research Assistant
Neuroscience Research Australia
Neuroscience Research Australia
Margarete Ainsworth Building
Barker Street Randwick Sydney NSW 2031 Australia
*M* 0411 308 769 *T* +61 2 9399 1883 *F* +61 2 9399 1265
neura.edu.au<http://neura.edu.au>
Follow @neuraustralia on twitter
<https://twitter.com/neuraustralia>Follow NeuRA on facebook
<https://www.facebook.com/NeuroscienceResearchAustralia>Subscribe to
the NeuRA Magazine<http://www.neura.edu.au/help-research/subscribe>
On 5/02/2015 10:46 am, Douglas N Greve wrote:
How sure are you that you are using the exact same model? Can you output
the design matrix from SPSS? Can you input the FS design matrix into
SPSS? Are you sure you are using the exact same input data?
On 02/04/2015 06:32 PM, Bronwyn Overs wrote:
Ahh right. I have just understood which part of the output I needed to
look at. However, for the ROI GLM there are only 1-2 regions that were
significantly different between groups, while the SPSS ANCOVA showed
significant group differences for the majority of parcellated regions.
I have confirmed that I am using the exact same model for each, so it
is only the analysis method that differs. Do you know why these two
methods would produce such disparate results?
Kind regards,
Bronwyn Overs
Research Assistant
Neuroscience Research Australia
Neuroscience Research Australia
Margarete Ainsworth Building
Barker Street Randwick Sydney NSW 2031 Australia
*M* 0411 308 769 *T* +61 2 9399 1883 *F* +61 2 9399 1265
neura.edu.au<http://neura.edu.au>
Follow @neuraustralia on twitter
<https://twitter.com/neuraustralia>Follow NeuRA on facebook
<https://www.facebook.com/NeuroscienceResearchAustralia>Subscribe to
the NeuRA Magazine<http://www.neura.edu.au/help-research/subscribe>
On 5/02/2015 3:16 am, Douglas N Greve wrote:
I'm not sure that I understand what I'm looking at. If it is an ROI
analysis, then there is no surface, it should be about 40 numbers, one
for each ROI.
On 02/03/2015 06:26 PM, Bronwyn Overs wrote:
Hi Doug,
Thanks for your correction.
I have now completed the FDR for my case-control comparisons, and it
appears that none of the regions survived. This is again quite
confusing given the large number of parcellated regions that survived
FDR in the SPSS ANCOVA. Can you confirm that this screenshot of the
sig.mgh file from ROI analysis looks as you would expect (it looks
very strange to me):
Kind regards,
Bronwyn Overs
Research Assistant
Neuroscience Research Australia
Neuroscience Research Australia
Margarete Ainsworth Building
Barker Street Randwick Sydney NSW 2031 Australia
*M* 0411 308 769 *T* +61 2 9399 1883 *F* +61 2 9399 1265
neura.edu.au<http://neura.edu.au>
Follow @neuraustralia on twitter
<https://twitter.com/neuraustralia>Follow NeuRA on facebook
<https://www.facebook.com/NeuroscienceResearchAustralia>Subscribe to
the NeuRA Magazine<http://www.neura.edu.au/help-research/subscribe>
On 4/02/2015 9:35 am, Douglas Greve wrote:
Yes, it should be .^
On 2/3/15 5:33 PM, Bronwyn Overs wrote:
Hi Doug,
I am having a problem with the line in Matlab 2014b:
p = 10^-abs(sigmat);
I keep getting the following error:
Error using ^
Inputs must be a scalar and a square matrix.
To compute elementwise POWER, use POWER (.^)
instead.
Do you know why this would be?
Kind regards,
Bronwyn Overs
Research Assistant
Neuroscience Research Australia
Neuroscience Research Australia
Margarete Ainsworth Building
Barker Street Randwick Sydney NSW 2031 Australia
*M* 0411 308 769 *T* +61 2 9399 1883 *F* +61 2 9399 1265
neura.edu.au<http://neura.edu.au>
Follow @neuraustralia on twitter
<https://twitter.com/neuraustralia>Follow NeuRA on facebook
<https://www.facebook.com/NeuroscienceResearchAustralia>Subscribe to
the NeuRA Magazine<http://www.neura.edu.au/help-research/subscribe>
On 4/02/2015 2:22 am, Douglas Greve wrote:
There is not a way to do it from the command line. You can do it in
matlab, something like
sig = MRIread('sig.mgh');
sigmat = fast_vol2mat(sig);
p = 10^-abs(sigmat);
fdr = .05;
pthresh = fast_fdrthresh(p,fdr);
ind = find(p < pthresh); % This will be a list of ROI indices that
survive FDR
doug
On 2/2/15 10:26 PM, Bronwyn Overs wrote:
Hi Doug,
That makes perfect sense. Just one more thing then, how do you
conduct an FDR via the command line for an ROI mri_glmfit analysis?
Kind regards,
Bronwyn Overs
Research Assistant
Neuroscience Research Australia
Neuroscience Research Australia
Margarete Ainsworth Building
Barker Street Randwick Sydney NSW 2031 Australia
*M* 0411 308 769 *T* +61 2 9399 1883 *F* +61 2 9399 1265
neura.edu.au<http://neura.edu.au>
Follow @neuraustralia on twitter
<https://twitter.com/neuraustralia>Follow NeuRA on facebook
<https://www.facebook.com/NeuroscienceResearchAustralia>Subscribe
to the NeuRA Magazine
<http://www.neura.edu.au/help-research/subscribe>
On 3/02/2015 2:00 pm, Douglas Greve wrote:
This is not something you would run the MC sim on because there
is no clustering, it is just a list of ROIs with their p-values.
The p-values are uncorrected. You can do bonferoni correction
across all the ROIs (or just the ones you are interested in). You
could do FDR too.
doug
On 2/2/15 9:39 PM, Bronwyn Overs wrote:
Thanks Doug, that worked well.
However, is it possible to run a monte-carlo simulation with
this GLM ROI analysis? I attempted to run it using the following
command...
mri_glmfit-sim --glmdir
DesikanROIAnal_case-control.thick.lh.glmdir --cache 1.3 abs
--cwpvalthresh 0.05 --2spaces
and received the following error:
ERROR: could not determine file for
DesikanROIAnal_case-control.thick.lh.glmdir/mask
Kind regards,
Bronwyn Overs
Research Assistant
Neuroscience Research Australia
Neuroscience Research Australia
Margarete Ainsworth Building
Barker Street Randwick Sydney NSW 2031 Australia
*M* 0411 308 769 *T* +61 2 9399 1883 *F* +61 2 9399 1265
neura.edu.au<http://neura.edu.au>
Follow @neuraustralia on twitter
<https://twitter.com/neuraustralia>Follow NeuRA on facebook
<https://www.facebook.com/NeuroscienceResearchAustralia>Subscribe to
the NeuRA Magazine
<http://www.neura.edu.au/help-research/subscribe>
On 3/02/2015 11:03 am, Douglas Greve wrote:
Even easier. Run aparcstats2table, then run mri_glmfit passing
the output of aparcstats2table with --table (instead of --y).
There's something on the wiki about it, also look for the ROI
tutorial.
doug
On 2/2/15 6:20 PM, Bronwyn Overs wrote:
Hi Doug,
I am not sure how to run an ROI analysis using mri_glmfit. Is
there a wiki page detailing this method (I was unable to find
one)? Is the first step to map lh.aparc.label and
rh.aparc.label from fsaverage to each of my individual
subjects using mri_label2label? When I do so and then view the
mapped label for an individual subject in freeview, it appears
to be a continuous label for all of the parcellated regions
combined. Is this correct?
Kind regards,
Bronwyn Overs
Research Assistant
Neuroscience Research Australia
Neuroscience Research Australia
Margarete Ainsworth Building
Barker Street Randwick Sydney NSW 2031 Australia
*M* 0411 308 769 *T* +61 2 9399 1883 *F* +61 2 9399 1265
neura.edu.au<http://neura.edu.au>
Follow @neuraustralia on twitter
<https://twitter.com/neuraustralia>Follow NeuRA on facebook
<https://www.facebook.com/NeuroscienceResearchAustralia>Subscribe
to the NeuRA Magazine
<http://www.neura.edu.au/help-research/subscribe>
On 3/02/2015 3:31 am, Douglas Greve wrote:
First, I would run the ROI analysis in mri_glmfit to see if
you get the same results as in SPSS. In the handfull of these
cases, no one has been able to correctly replicate the FS
design matrix in SPSS, so I suspect that is part of the
discrepancy. The other thing is that ROI and vertex-wise
analyses are simply different. As an extreme example, if some
of the vertices are pos and some are neg then they would
cancel out when you average them in an ROI but individually
could be significant at the vertex level. If you analyze the
average over the cluster then that should come out as
significant.
doug
On 2/1/15 11:36 PM, Bronwyn Overs wrote:
Dear FreeSurfer Mailing List,
I have a sample of schizophrenia and control subjects for
whom I have run a case-control analysis of cortical
thickness using two separate methods (GLM vertex-wise
analysis in freesurfer, repeated measures ANCOVA analysis of
parcellated data in SPSS). However, for each methods of
analysis I am getting extremely different results. For the
GLM in Freesurfer I have only 1 small cluster in the frontal
lobe that differs between cases and controls (controlling
for all other IVs, FWMH = 10mm, cluster-forming threshold=
.05, cluster-wise pval=.05), while for the ANCOVA method all
but 8 of the parcellated regions differ significantly
between groups (p<.05). For both methods I have used the
same model of predictors (independent variables = gender,
group, scanning site; covariate = age) and exactly the same
sample of participants. I have also replicated the GLM
analysis using the QDEC GUI to ensure that I had no made any
mistakes.
Can you provide any insight into why I would be seeing such
different results for each method using the same data set?
My findings using the ANCOVA analysis make much more sense
to me, given previous findings of reduced cortical thickness
in schizophrenia subjects. I was surprised not to find the
same pattern of effects using the GLM analysis.
--
Kind regards,
Bronwyn Overs
Research Assistant
Neuroscience Research Australia
Neuroscience Research Australia
Margarete Ainsworth Building
Barker Street Randwick Sydney NSW 2031 Australia
*M* 0411 308 769 *T* +61 2 9399 1883 *F* +61 2 9399 1265
neura.edu.au<http://neura.edu.au>
Follow @neuraustralia on twitter
<https://twitter.com/neuraustralia>Follow NeuRA on facebook
<https://www.facebook.com/NeuroscienceResearchAustralia>Subscribe
to the NeuRA Magazine
<http://www.neura.edu.au/help-research/subscribe>
_______________________________________________
Freesurfer mailing list
Freesurfer@nmr.mgh.harvard.edu
https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
_______________________________________________
Freesurfer mailing list
Freesurfer@nmr.mgh.harvard.edu
https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
The information in this e-mail is intended only for the person to whom it is
addressed. If you believe this e-mail was sent to you in error and the e-mail
contains patient information, please contact the Partners Compliance HelpLine at
http://www.partners.org/complianceline . If the e-mail was sent to you in
error
but does not contain patient information, please contact the sender and properly
dispose of the e-mail.
_______________________________________________
Freesurfer mailing list
Freesurfer@nmr.mgh.harvard.edu
https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
_______________________________________________
Freesurfer mailing list
Freesurfer@nmr.mgh.harvard.edu
https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
The information in this e-mail is intended only for the person to whom it is
addressed. If you believe this e-mail was sent to you in error and the e-mail
contains patient information, please contact the Partners Compliance HelpLine at
http://www.partners.org/complianceline . If the e-mail was sent to you in
error
but does not contain patient information, please contact the sender and properly
dispose of the e-mail.
_______________________________________________
Freesurfer mailing list
Freesurfer@nmr.mgh.harvard.edu
https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
_______________________________________________
Freesurfer mailing list
Freesurfer@nmr.mgh.harvard.edu
https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
The information in this e-mail is intended only for the person to whom it is
addressed. If you believe this e-mail was sent to you in error and the e-mail
contains patient information, please contact the Partners Compliance HelpLine at
http://www.partners.org/complianceline . If the e-mail was sent to you in
error
but does not contain patient information, please contact the sender and properly
dispose of the e-mail.
_______________________________________________
Freesurfer mailing list
Freesurfer@nmr.mgh.harvard.edu
https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
_______________________________________________
Freesurfer mailing list
Freesurfer@nmr.mgh.harvard.edu
https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
The information in this e-mail is intended only for the person to whom it is
addressed. If you believe this e-mail was sent to you in error and the e-mail
contains patient information, please contact the Partners Compliance HelpLine at
http://www.partners.org/complianceline . If the e-mail was sent to you in
error
but does not contain patient information, please contact the sender and properly
dispose of the e-mail.
_______________________________________________
Freesurfer mailing list
Freesurfer@nmr.mgh.harvard.edu
https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
_______________________________________________
Freesurfer mailing list
Freesurfer@nmr.mgh.harvard.edu
https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
The information in this e-mail is intended only for the person to whom it is
addressed. If you believe this e-mail was sent to you in error and the e-mail
contains patient information, please contact the Partners Compliance HelpLine at
http://www.partners.org/complianceline . If the e-mail was sent to you in
error
but does not contain patient information, please contact the sender and properly
dispose of the e-mail.
_______________________________________________
Freesurfer mailing list
Freesurfer@nmr.mgh.harvard.edu
https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
_______________________________________________
Freesurfer mailing list
Freesurfer@nmr.mgh.harvard.edu
https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
_______________________________________________
Freesurfer mailing list
Freesurfer@nmr.mgh.harvard.edu
https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
_______________________________________________
Freesurfer mailing list
Freesurfer@nmr.mgh.harvard.edu
https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
The information in this e-mail is intended only for the person to whom it is
addressed. If you believe this e-mail was sent to you in error and the e-mail
contains patient information, please contact the Partners Compliance HelpLine at
http://www.partners.org/complianceline . If the e-mail was sent to you in error
but does not contain patient information, please contact the sender and properly
dispose of the e-mail.
