See inline replies/comments below.
On Jul 15, 2013, at 12:55 PM, Michael Cohen <[email protected]> wrote: > Thanks for the feedback on this. Just a few follow-up questions/issues... > > About AFM vs. MFM: because I'm using functional volumes that have already > been thresholded, I think AFM is better. (MFM looked to be much noisier.) > But even with AFM, the areas of activation look somewhat different in the > Caret rendering than in other programs. > > Also, I did try what you suggested, but I'm not quite sure what to make of > it. Assuming that I did everything right, I assume that the green line in > the attached image represents the surface. It represents a surface. Did you make sure that the surface contour that is toggled on in the D/C: Volume Outline window is the same surface you used as the mapping substrate? I assume this is a PALS average surface (e.g., FLIRT, SPM5, or MRITOTAL?). What is the anatomical volume underlay? What was the spatial normalization (and atlas target) used to get the subjects into atlas space for the group analysis? The alignment looks reasonable, but the green outline deviates from the mean T1 in places, so it made me ask. > Then, should I assume that the surface rendering just looks at whether there > is activation in the voxels that cross that line (with some variation based > on the algorithm chosen, of course)? If you chose enclosing voxel, yes, but as you say, it depends on the algorithm. The gaussian algorithm lets you build an ellipsoid earound the vertex. > The strange thing here is that in fact, some activation that appears to be > on the IFG in the flat coronal slice is on the part of the green line that > represents the MFG. But I'm not sure why that would be, based on what I did. > Looking at the attached image, does it look like I did something wrong? Not really, although I did wonder, based on the fact that your activation map's lateral extent seems medial to both the green contour and the gray matter in the mean T1 underlay whether your subjects' brain extraction might have been overly aggressive, or whether you otherwise masked out activation near the edges. But it could also be a consequence of doing your analysis in volume-land. Maybe nothing is wrong at all. You seem surprised, perhaps based on your paradigm, to see activation in both IFG and MFG. Looking at your capture, it is clear you have an extended cluster. This need not mean something is amiss. These are not the regions I'd expect in a finger-tapping task, but in the absence of more information, I wouldn't find this to be evidence you did something wrong -- unless your Caret rendering doesn't match it. In that case, the volume you mapped and the volume displayed in your capture are not the same, or one had palette/thresholding that the other did not. > Finally, to make sure I understand Matt's point: I take it that your answer > to my original question is that we generally should trust Caret to give a > more accurate representation of which areas are "active", as compared to a > volume-based tool like MRIcron? That's good to know--thanks. I think Matt was urging you to do your analysis on the surface, rather than on the volume. Matt does very fine-grained work for which he finds volumetric analysis/display inadequate. As a complement to surface-based display and analysis, he uses volumes all the time. > --Michael > <Caret volume_surface sample.png> > > > On Sat, Jul 13, 2013 at 4:57 PM, Donna Dierker <[email protected]> > wrote: > Hi Michael, > > If you are using multi-fiducial mapping (MFM), you will get a broader > representation than if you use average fiducial mapping (AFM). This is > explained in section 1.22.2 of the Sept 2006 tutorial > (http://brainvis.wustl.edu/wiki_linked_files/documentation/Caret_Tutorial_Sep22.pdf), > which says: > >> In AFM, each node is assigned the value of the voxel in which it resides (or >> an interpolated value, depending on the specific algorithm chosen). In MFM, >> each node takes the average value after mapping the volume to each of the 12 >> contributing hemispheres. MFM gives a smoother map and the best estimate of >> spatial localization; AFM gives the most likely peak value. > > Depending on your goals, sometimes AFM will be the best choice; other times, > MFM will be the better choice. > > Trying both AFM and MFM and comparing may be enough to sort your query. If > not, then view your activation map over the anatomical template that best > represents your study participants (e.g., avg152T1 if this was the atlas > target used for spatial normalization). Over this anatomical volume, view > the contour of the Caret average fiducial surface that most closely > represents your spatial normalization method, i.e. if PALS, then: > > http://brainvis.wustl.edu/help/pals_volume_normalization/ > > The methods shown on that page link to pages that detail how the warp was > made, but near the end, they also tell you how to check the alignment between > surface and volume. Do this with the Caret average fiducial, your atlas > target T1, and overlay your activation map. Open a Window 2 in Caret with an > inflated view that shows your metric overlay. Click near the IFG/MFG in > Window 2 (inflated), and observe the alignment between the volume and surface > in the main window. (Switch the volume view from H (horizontal) to All, so > you can see all three views at once.) It should look kind of like the > attached capture, but the palette on the inflated surface should match the > palette in the volume window. > > This will help you sort out which rendering best represents your results. > > Cheers, > > Donna > <fmri.tiff> > > On Jul 13, 2013, at 4:43 PM, Michael Cohen wrote: > >> Hi, >> I am new to Caret, and I have an activation image that I've tried rendering >> both in Caret and in MRIcron. One large cluster of interest from our study >> is largely in the left inferior frontal gyrus (VLPFC), but the cluster seems >> to extend into middle frontal gyrus (DLPFC). >> >> The thing that seems somewhat strange is that in the MRIcron rendering, the >> cluster appears to only barely crosses the border from IFG into MFG, with >> the vast majority of the cluster appearing in IFG. In the Caret rendering, >> the cluster is shown as extending through a fairly large portion of the >> ventral surface of the MFG. These different renderings would lead to >> somewhat different interpretations of our results, and I'm not sure at this >> point which rendering is a better representation of our data. >> >> So I guess my question for this list is about what the goal of Caret is. >> Specifically, is it actually intended to provide a more accurate mapping >> from the slices to the brain surface, compared to a tool like MRIcron (which >> seems to use a simpler algorithm in rendering)? Or is it just intended to >> provide more realistic-looking renderings, without necessarily having a >> better correspondence between the raw image and the 3-D rendering than >> MRicron? >> >> Any advice or suggestions that you have on this would be very helpful. >> >> Many thanks, >> Michael >> >> -- >> Michael S. Cohen, M.S., C. Phil >> Graduate Student >> Department of Psychology >> University of California, Los Angeles >> >> >> _______________________________________________ >> caret-users mailing list >> [email protected] >> http://brainvis.wustl.edu/mailman/listinfo/caret-users > > > _______________________________________________ > caret-users mailing list > [email protected] > http://brainvis.wustl.edu/mailman/listinfo/caret-users > > > _______________________________________________ > caret-users mailing list > [email protected] > http://brainvis.wustl.edu/mailman/listinfo/caret-users _______________________________________________ caret-users mailing list [email protected] http://brainvis.wustl.edu/mailman/listinfo/caret-users
