[caret-users] Visualization in Caret

Donna Dierker donna.dierker at sbcglobal.net
Wed Feb 5 17:54:05 CST 2014


On Feb 5, 2014, at 4:19 PM, Eshita Shah <eshshah at ucla.edu> wrote:

> Hi Donna, 
> 
> I have tried changing the user threshold in the Metric Settings menu, but nothing seems to change beyond +/- 0.05. There are a few blotches of orange and yellow when it is at 0, and many sub-threshold regions (green) show up when I put it up to 0.05, but nothing else changes as I move beyond 0.05 to higher values (the orange slowly disappears, and it's all green). 

At least in Caret5 (less sure about workbench), thresholding won't work properly on the p-value, because thresholding assumes more extreme values -- further from zero -- are the more exceptional ones, whereas the opposite is true with p-values, where the closer to 0, the more rare.  Since q is 1-p it should behave better in caret5 thresholding.  If you threshold at q=.95, you should see less than if you threshold at q=.90.  Like percentiles.

> Is the value I'm changing the p or the q value? Or does that depend on what column I have loaded in the "Threshold Adjustment" section?

I'd display and threshold on both, for now, while you are trying to understand what the data shows.

> If I am changing the q value, then does it mean that the regions that are showing up have a p-value greater than 0.95 (since nothing changes after 0.05) and thus they're not showing up as significant in my report?

If you threshold at q=.95, you should see vertices colored that have p values of .05 or less, but you know none exist, because nothing survived in your report.  Start at q=0.5.  See some vertices.  Probably lots of them.  Then try q=0.75.  You should see the clusters shrink now.  Now 0.90.  Anything?

> Let me know if I am interpreting this the wrong way. 
> 
> Also, the coloring somewhat changes depending on the color palette I use. I believe the default is PSYCH, but when I change it to PSYCH-NO-NONE, I see orange blots in more regions than before (and of course what was grey earlier turns dark blue). Why is that? The "new" orange blots appear in the same positions as the sub-threshold green color does when I change the user threshold to 0.05. 

It is how the palette is defined.  There is a region of the color scale that blots out coloring near zero, while the NO-NONE removes that gap.  While my memory fails me as to why,, I remember thinking there was something not quite intuitive about the one with the gap. Palettes are a matter of taste to some degree.  Some are better with pos/neg values, while others are better with positive only, which is what you will have with your f-stats.  For figures, I don't use p/q-values typically, but rather t- or f-maps.

But for right now, you're doing a post-mortem on your analysis to see how close you were to having differences, so the q-maps will be useful for this purpose.

> Lastly, how do I know which group is baseline and treatment? Does TFCE automatically output the control group as the baseline, so the yellow would indicate that the sulci are deeper in the treatment group vs. control? Or the other way around? 

You used an ANOVA, which should produce a f-map -- all positive.  There should be no +/- valence to it, unless I'm misunderstanding what you did.

Out of curiosity, how many subjects were in each group?

If you have only two groups and want to see where one group is deeper than the other, you can run a t-test instead of an anova.

> Thanks for your help, 
> Eshita 
> 
> 
> 
> On Wed, Feb 5, 2014 at 5:58 AM, Donna Dierker <donna.dierker at sbcglobal.net> wrote:
> Use the D/C: Metric Settings menu to adjust the threshold to .90, .85, etc. until you start seeing something.  If you see nothing, set it to zero and start cranking up in larger increments.  Q=1-p.
> 
> 
> On Feb 4, 2014, at 8:09 PM, Eshita Shah <eshshah at ucla.edu> wrote:
> 
> > Hi Donna,
> >
> > What file specifically outputs the q-values and how far they are from significance? I think I am able to load the Q statistic column from the f-map onto the Conte69 atlas, but where should I be looking if I want to know what to change the threshold to?
> >
> > Thank you,
> > Eshita
> >
> >
> > On Tue, Feb 4, 2014 at 8:32 AM, Donna Dierker <donna at brainvis.wustl.edu> wrote:
> > Yes, pretty much:  I usually have a study directory into which I copy the Conte69 files.  Then I rename the Conte69 spec to something more study-specific.  I usually use the Conte69 inflated and very inflated for t-map visualization, along with mean group mid thickness (both medial/lateral surface views, but also overlaid as contours on volume slices).
> >
> > I don't usually use the TFCE column for visualization, and if I recall correctly, there might be p-value and q-value (1-p, which works better with the Caret thresholding) columns.  This can tell you how close to significance you got.
> >
> > And yes:  You use the D/C Overlay/Underlay surface menu to control what is displayed, which column, etc.
> >
> >
> > On Feb 3, 2014, at 6:10 PM, Eshita Shah <eshshah at ucla.edu> wrote:
> >
> > > Yes, that's what I was afraid of. I was expecting significant differences between the two groups. But thanks for clarifying.
> > >
> > > I am still a bit confused on how exactly to load the metric files on the Conte69 atlas. Do I open up the Conte69 spec and "add data files" in the menu to open up TFCE files? And then do I overlay it using D/C --> Overlay/Underlay Surfaces --> Primary Overlay, etc.?
> > >
> > > Again, thank you for all your help.
> > >
> > > Eshita
> > >
> > >
> > > On Mon, Feb 3, 2014 at 3:49 PM, Donna Dierker <donna.dierker at sbcglobal.net> wrote:
> > > No, I think the problem is that nothing survived TFCE thresholding.  If it had, you would see an entry (or more) under the column heads (Column, Thresh, Num-Nodes, etc.).  There is no entry, which means nothing survived.
> > >
> > > Column    Thresh  Num-Nodes          Area  Area-Corrected     COG-X     COG-Y
> > >   COG-Z   P-Value
> > >
> > > TFCE           P
> > >
> > > You can try loading your f-map (ANOVA_29-01-14.OCD_CTRL.Depth.LH.fmap.significant.tfce.1.0E.2.0H.73730.metric) and switch to the TFCE column, and apply thresholds corresponding to the list of values right under the column heads, so you can see how close/far you were.
> > >
> > > I am under the weather right now, so I will have another look at this tomorrow, but I honestly think you are interpreting it correctly.  If you are like me, you probably are disappointed with these results.  (There are exceptions, of course.)
> > >
> > >
> > > On Feb 3, 2014, at 4:37 PM, Eshita Shah <eshshah at ucla.edu> wrote:
> > >
> > > > Donna,
> > > >
> > > > Thank you so much for your thorough response. What I'm worried about as of now is the significance.report.txt file. I have uploaded it using the link you provided, please let me know if there is anything unusual. When I ran ANOVA without TFCE, I had rows of information right below the header, as you mentioned. But for the TFCE report, I don't see anything similar. Maybe I am interpreting it incorrectly?
> > > >
> > > > Thank you,
> > > > Eshita
> > > >
> > > >
> > > > On Fri, Jan 31, 2014 at 1:15 PM, Donna Dierker <donna.dierker at sbcglobal.net> wrote:
> > > > On Jan 31, 2014, at 2:17 PM, Eshita Shah wrote:
> > > >
> > > >> Hi Donna,
> > > >>
> > > >> Yes! I was able to successfully get past the issue of JRE halting-- I just installed the latest JRE as Tim suggested, and added some options for garbage collection so that it would optimize memory use. Thank you for all your help!
> > > >>
> > > >> I have computed one mean midthickness for all my subjects, but specifically how do I overlay that onto an anatomical template? Would there be any advantage of using the NIFTI volume vs. using an average volume created from my subject pool?
> > > >
> > > > One advantage of using the template used for stereotaxic/volumetric registration, if any was done, is that it is standard.  Reviewers and readers are more familiar with it, and don't have to understand how it was generated.  This is just for display/orientation -- not for analysis.
> > > >
> > > > Another is that you don't have the extra step of computing a mean volume.
> > > >
> > > >> f so, how would I be able to generate that average volume?
> > > >
> > > > I usually use AFNI's 3dMean when I need to do this, but FSL, SPM, and other packages have similar features.  Maybe wb_command supports it now.  You can probably do it in multiple steps with caret_command, but it's a pain.
> > > >
> > > >> I am also a bit unclear on how to interpret and draw conclusions from the outputs of TFCE. I understand that TFCE creates many .metric files including one that indicates all the significant differences between the two groups. How can I overlay that (along with the .label file) onto a surface in Caret?
> > > >
> > > > I usually generate a border about the cluster in the label.gii file and overlay it on the unthresholded t-map, so that users can see subthreshold diffs.  I display the t-map on the inflated atlas surface (Conte69, if I recall correctly here).  If there are diffs in the insula/operculum, i use the very inflated surface, which shows them more clearly.
> > > >
> > > >> (Where does the mean midthickness come into play?)
> > > >
> > > > Sometimes it is evident just by comparing the mean midthickness surfaces that there is a difference.  Other times, you need to look at a slice view of the template with group contours overlaid at a slice that best shows the diffs.  Could be coronal, axial, or sagittal.
> > > >
> > > >> Also, how do I interpret the results written in the significance.report text file?
> > > >
> > > > If you upload your report, I can tell you the lines to focus on:
> > > >
> > > > http://brainvis.wustl.edu/cgi-bin/upload.cgi
> > > >
> > > > They should be near the top, just below a header that lists the column, number of nodes, corrected and uncorrected areas, x, y, z, etc.  I'm psyched you got this far!  I was feeling frustrated after you ran into the JRE problem.  I'm glad you got past it.
> > > >
> > > >> Thank you so much.
> > > >>
> > > >> Sincerely,
> > > >> Eshita
> > > >>
> > > >>
> > > >> On Thu, Jan 30, 2014 at 5:17 PM, Donna Dierker <donna.dierker at sbcglobal.net> wrote:
> > > >> Wow, does this mean you got past the grind-to-a-halt JRE problem?  Excellent!
> > > >>
> > > >> Here is a script I used to compute mean midthickness surfaces for two groups:
> > > >>
> > > >> http://brainmap.wustl.edu/pub/donna/US/UCLA/ESHITA/gen_mean_fiducials.pared.sh
> > > >> login pub
> > > >> password download
> > > >>
> > > >> But the main command is this one:
> > > >>
> > > >> caret_command -surface-average $OUTCOORD $COORD1 $COORD2 … $COORDn $SHAPE
> > > >>
> > > >> The $SHAPE is a vertex:scalar mapping identical in format to a metric, but it stores the 3D variability for each vertex.
> > > >>
> > > >> You can visualize multiple mean coord files (e.g., one for each DX group) overlaid on the same anatomical volume (e.g., avg152T1) and click on hot spots on your metric, to see if the contours diverge there.  You can also compute the distance between the two surfaces directly on the Surface: Measures menu (if I recall correctly).
> > > >>
> > > >> Sounds like you're making great progress!
> > > >>
> > > >>
> > > >> On Jan 30, 2014, at 5:27 PM, Eshita Shah <eshshah at ucla.edu> wrote:
> > > >>
> > > >> > Hello,
> > > >> >
> > > >> > I have created metric files from my TFCE statistical analysis that I wish to view on my own study-specific generated average coordinate file. How would I go about doing so? I do have the Conte69 Visualization Atlas, but I am not sure how to overlay the metric files generated by TFCE to visualize significant clusters. I would eventually like to do this overlay on my own average file, not the 164k averages.
> > > >> >
> > > >> > Thank you,
> > > >> > Eshita
> > > >> >
> > > >> > --
> > > >> > Eshita Shah
> > > >> > University of California, Los Angeles | 2014
> > > >> > B.S. Neuroscience
> > > >> > eshshah at ucla.edu
> > > >> >
> > > >> >
> > > >> > _______________________________________________
> > > >> > caret-users mailing list
> > > >> > caret-users at brainvis.wustl.edu
> > > >> > http://brainvis.wustl.edu/mailman/listinfo/caret-users
> > > >>
> > > >>
> > > >> _______________________________________________
> > > >> caret-users mailing list
> > > >> caret-users at brainvis.wustl.edu
> > > >> http://brainvis.wustl.edu/mailman/listinfo/caret-users
> > > >>
> > > >>
> > > >>
> > > >> --
> > > >> Eshita Shah
> > > >> University of California, Los Angeles | 2014
> > > >> B.S. Neuroscience
> > > >> eshshah at ucla.edu
> > > >>
> > > >>
> > > >> _______________________________________________
> > > >> caret-users mailing list
> > > >> caret-users at brainvis.wustl.edu
> > > >> http://brainvis.wustl.edu/mailman/listinfo/caret-users
> > > >
> > > >
> > > > _______________________________________________
> > > > caret-users mailing list
> > > > caret-users at brainvis.wustl.edu
> > > > http://brainvis.wustl.edu/mailman/listinfo/caret-users
> > > >
> > > >
> > > >
> > > >
> > > > --
> > > > Eshita Shah
> > > > University of California, Los Angeles | 2014
> > > > B.S. Neuroscience
> > > > eshshah at ucla.edu
> > > >
> > > >
> > > > _______________________________________________
> > > > caret-users mailing list
> > > > caret-users at brainvis.wustl.edu
> > > > http://brainvis.wustl.edu/mailman/listinfo/caret-users
> > >
> > >
> > > _______________________________________________
> > > caret-users mailing list
> > > caret-users at brainvis.wustl.edu
> > > http://brainvis.wustl.edu/mailman/listinfo/caret-users
> > >
> > >
> > >
> > > --
> > > Eshita Shah
> > > University of California, Los Angeles | 2014
> > > B.S. Neuroscience
> > > eshshah at ucla.edu
> > >
> > >
> > > _______________________________________________
> > > caret-users mailing list
> > > caret-users at brainvis.wustl.edu
> > > http://brainvis.wustl.edu/mailman/listinfo/caret-users
> >
> >
> > _______________________________________________
> > caret-users mailing list
> > caret-users at brainvis.wustl.edu
> > http://brainvis.wustl.edu/mailman/listinfo/caret-users
> >
> >
> > _______________________________________________
> > caret-users mailing list
> > caret-users at brainvis.wustl.edu
> > http://brainvis.wustl.edu/mailman/listinfo/caret-users
> 
> 
> _______________________________________________
> caret-users mailing list
> caret-users at brainvis.wustl.edu
> http://brainvis.wustl.edu/mailman/listinfo/caret-users
> 
> _______________________________________________
> caret-users mailing list
> caret-users at brainvis.wustl.edu
> http://brainvis.wustl.edu/mailman/listinfo/caret-users




More information about the caret-users mailing list