Hi,
I wanted to re-post my questions from a couple of days ago below, but with some more specific questions following a search through the archives.
I want to be able to extract the beta values from a cluster identified using mc-z in a group by behavioral variable interaction so that I can 1) plot the relationship of thickness to behavioral variable data by group in that cluster, 2) conduct post hoc tests to examine the interaction, and 3) calculate the Rsquare and partial correlations for each variable in the glm (i.e. how much variation in thickness is explained by my behavioral variable).
To extract the beta values from a cluster identified by mc-z would I treat the cluster like a label and use mri_segstats to extract the beta weights from the cluster? Would I need to make a label of all the clusters that I want to do this for first?
Is there a way to calculate the basic statistics for the glm and extract in table form? i.e. Fs and ps for peaks of each cluster? What about Rsquare, or the correlation between thickness and my behavioral variable in the clusters? or would I need to compute these outside of freesurfer using the extracted betas?
Thank you!
Laura.
On Mon, Mar 18, 2013 at 5:05 PM, Laura M. Tully tully.laura@googlemail.comwrote:
Hi Freesurfer experts,
I'm hoping you can help me understand how to interpret interactions in clusters identified in whole brain analysis using glmfit and glmfit-sim. Below I describe what I've done and what I'd like to be able to do. Any suggestions would be most appreciated!
- I have two groups (patients, controls) and a behavioral variable of
interest (social functioning). I am interested in cortical thickness differences between groups (main effect of group), whether cortical thickness relate to social functioning across the group (main effect of social functioning), and whether this relationship differs by group (group x social functioning interaction).
- I ran whole brain analysis using mri_glmfit with group and
functioning as variables of interest whilst controlling for/regressing out gender, age, and mean thickness. i.e. 4 classes (conmale,confemale, ptmale, ptfemale) and 3 continuous variables (age, AvgThickness, Functioning) = 16 regressors.
- I tested the group x functioning interaction with the following
contrast - is it correct?
0 0 0 0 0 0 0 0 0 0 0 0 0.5 0.5 -0.5 -0.5
- I then ran mri_glmfit-sim to identify clusters that survive multiple
comparisons. This revealed 4 clusters (3 in LH; 1 in RH) that represent regions showing significant group x functioning interaction.
- I visualized the clusters in tksurfer, and by loading the y.fsgd
file was able to visualize the plotted data to get a sense of the interaction, but this is as much as I know in terms of how to examine interactions in the cluster data......
My specifc questions include:
- I understand that the values in xxx.sig.cluster.mgh overlay reflect
log10 p values, the signs of which indicate the direction of the relationship (i.e. -3 = negative correlation between thickness & variable) but I'm not sure how to interpret this in the context of an interaction with group?
- I understand that the values in xxx.y.ocn.dat contain the average
thickness value for each subject in that cluster and that in a simple between groups test this data could be used to conduct post hoc t-tests to show the direction of the difference, but again I'm not sure how to use this data in the context of the interaction. What do the values represent in a group x variable interaction?
Ideally, I'd like to extract the contrast estimates for each subject in the group x functioning contrast and plot it in another program and conduct pairwise comparisons (t-tests) in order to get a better understanding of the interaction). I'm not sure how to do this - is it possible? My thinking is that I do something similar in fMRI analysis in spm where I can plot the contrasts in a significant cluster and then extract both the average contrast estimates for each group and the contrast estimates for each individual subject.
Thanks in advance!
Laura.
--
Laura M. Tully, MA Social Neuroscience & Psychopathology, Harvard University Center for the Assessment and Prevention of Prodromal States, UCLA Semel Institute of Neuroscience ltully@mednet.ucla.edu ltully@fas.harvard.edu 310-267-0170 -- My musings as a young clinical scientist: http://theclinicalbrain.blogspot.com/ Follow me on Twitter: @tully_laura
On 03/20/2013 10:17 AM, Laura M. Tully wrote:
Hi,
I wanted to re-post my questions from a couple of days ago below, but with some more specific questions following a search through the archives.
I want to be able to extract the beta values from a cluster identified using mc-z in a group by behavioral variable interaction so that I can
- plot the relationship of thickness to behavioral variable data by
group in that cluster, 2) conduct post hoc tests to examine the interaction, and 3) calculate the Rsquare and partial correlations for each variable in the glm (i.e. how much variation in thickness is explained by my behavioral variable).
To extract the beta values from a cluster identified by mc-z would I treat the cluster like a label and use mri_segstats to extract the beta weights from the cluster? Would I need to make a label of all the clusters that I want to do this for first?
You can do it label by label. Or, if the annotation created by mri_glmfit-sim has all the clusters you want, you can use that (also in mri_segstats). You can also create an annotation from individual labels with mris_label2annot.
Is there a way to calculate the basic statistics for the glm and extract in table form? i.e. Fs and ps for peaks of each cluster? What about Rsquare, or the correlation between thickness and my behavioral variable in the clusters? or would I need to compute these outside of freesurfer using the extracted betas?
You will need to do it outside of FS. What are Fs and ps? doug
Thank you!
Laura.
On Mon, Mar 18, 2013 at 5:05 PM, Laura M. Tully <tully.laura@googlemail.com mailto:tully.laura@googlemail.com> wrote:
Hi Freesurfer experts, I'm hoping you can help me understand how to interpret interactions in clusters identified in whole brain analysis using glmfit and glmfit-sim. Below I describe what I've done and what I'd like to be able to do. Any suggestions would be most appreciated! * I have two groups (patients, controls) and a behavioral variable of interest (social functioning). I am interested in cortical thickness differences between groups (main effect of group), whether cortical thickness relate to social functioning across the group (main effect of social functioning), and whether this relationship differs by group (group x social functioning interaction). * I ran whole brain analysis using mri_glmfit with group and functioning as variables of interest whilst controlling for/regressing out gender, age, and mean thickness. i.e. 4 classes (conmale,confemale, ptmale, ptfemale) and 3 continuous variables (age, AvgThickness, Functioning) = 16 regressors. * I tested the group x functioning interaction with the following contrast - is it correct? 0000000000000.50.5-0.5-0.5 * I then ran mri_glmfit-sim to identify clusters that survive multiple comparisons. This revealed 4 clusters (3 in LH; 1 in RH) that represent regions showing significant group x functioning interaction. * I visualized the clusters in tksurfer, and by loading the y.fsgd file was able to visualize the plotted data to get a sense of the interaction, but this is as much as I know in terms of how to examine interactions in the cluster data...... My specifc questions include: * I understand that the values in xxx.sig.cluster.mgh overlay reflect log10 p values, the signs of which indicate the direction of the relationship (i.e. -3 = negative correlation between thickness & variable) but I'm not sure how to interpret this in the context of an interaction with group? * I understand that the values in xxx.y.ocn.dat contain the average thickness value for each subject in that cluster and that in a simple between groups test this data could be used to conduct post hoc t-tests to show the direction of the difference, but again I'm not sure how to use this data in the context of the interaction. What do the values represent in a group x variable interaction? Ideally, I'd like to extract the contrast estimates for each subject in the group x functioning contrast and plot it in another program and conduct pairwise comparisons (t-tests) in order to get a better understanding of the interaction). I'm not sure how to do this - is it possible? My thinking is that I do something similar in fMRI analysis in spm where I can plot the contrasts in a significant cluster and then extract both the average contrast estimates for each group and the contrast estimates for each individual subject. Thanks in advance! Laura. -- -- Laura M. Tully, MA Social Neuroscience & Psychopathology, Harvard University Center for the Assessment and Prevention of Prodromal States, UCLA Semel Institute of Neuroscience ltully@mednet.ucla.edu <mailto:ltully@mednet.ucla.edu> ltully@fas.harvard.edu <mailto:ltully@fas.harvard.edu> 310-267-0170 <tel:310-267-0170> -- My musings as a young clinical scientist: http://theclinicalbrain.blogspot.com/ Follow me on Twitter: @tully_laura--
Laura M. Tully, MA Social Neuroscience & Psychopathology, Harvard University Center for the Assessment and Prevention of Prodromal States, UCLA Semel Institute of Neuroscience ltully@mednet.ucla.edu mailto:ltully@mednet.ucla.edu ltully@fas.harvard.edu mailto:ltully@fas.harvard.edu 310-267-0170 -- My musings as a young clinical scientist: http://theclinicalbrain.blogspot.com/ Follow me on Twitter: @tully_laura
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
freesurfer@nmr.mgh.harvard.edu
-
Douglas N Greve -
Laura M. Tully