On 10/15/2010 11:00 AM, Bruno L. Giordano wrote: > Hello Donna, > > yes, I see the trade-offs, and I have just tried registering the > surfaces from the T1s of the subjects onto those from the > group-average T1: although the main landmarks are there, the resulting > surfaces are horrible! What were the original (pre-registration) surfaces like? Were they not horrible?
It isn't clear to me why post-registration surfaces would be horrible, unless the pre-registration ones were. Or there were lots of crossovers or torquing during registration, but this is pretty rare. > > The way I see it, what is important is that I check the projections of > the functionals onto the average of the surfaces of my subjects, > whichever target I might choose for the spherical registration. See, the thing is that you should not be mapping individual functionals onto a group average surface. Here are the typical routes: 1) Quick and dirty, low effort/time, less fidelity: Map your statistical output maps (e.g., contrasts across groups during same task) onto one of our PALS-B12 average fiducial surfaces (or MFM). 2) Greater effort/time, higher fidelity: Generate surfaces for each subject; map individual fMRI results to individual's surface; register surface to PALS-B12 atlas; do group statistics on PALS-B12 standard mesh; render group results on standard PALS-B12 visualization substrates It feels like you're somehow mixing these two. > > Thank you for your time and for your pro bono tutoring :-) > > Bruno > > > > > > On 15/10/2010 11:23 AM, Donna Dierker wrote: >> On 10/14/2010 03:29 PM, Bruno L. Giordano wrote: >>> Hi Donna, >>> >>>>> 1. create a target surface from the SPM-normalized T1s of each of the >>>>> participants: this is not a very nice representation, but it shows >>>>> very clear landmarks (with the possible exception of the ventral >>>>> medial wall); >>>> So far, so good. What software will you use to generate the surfaces? >>> My description was incomplete: I meant average the >>> participant-specific SPM-normalized T1s to create a group-average T1. >>> Then create a target surface by segmenting this group T1 with Caret. >>> This surface will obviously look bad. However, it still has each of >>> the default landmarks very clear. This target surface will be improved >>> at point 3. >> Segmenting group averaged MRI is usually a bad idea. >>>>> 2. project the surfaces of the participants (derived from >>>>> SPM-normalized T1s) onto this target surface; >>>> This isn't really clear. As best I can discern, you mean register >>>> your >>>> subjects to the PALS-B12 target atlas, at which time your various >>>> coord >>>> files, including fiducial/midthickness coord files, will get resampled >>>> onto the PALS-B12 73730 standard mesh. >>> See above. >>>>> 3. create a new improved target surface by averaging those coming out >>>>> from 2. (eventually include the additional temporal landmarks at this >>>>> point and not at 1.). >>>> Why? What is your motivation for creating a customized target, >>>> instead >>>> of using PALS-B12? What is different about your subjects? >>> My thought is that by using a target surface that actually comes out >>> from my subjects the quality of the functional-to-surface mapping will >>> be improved, especially if I do the points 4-6. below: >>> >>>>> 4. project the surfaces of the participants used at 2. on the new >>>>> target; >>>>> 5. average surfaces coming out at 4; >>>>> 6. project my functionals (group results) onto the average surface >>>>> from 5. >> No, I am not convinced you will significantly improve your outcome >> this way. >> >> If I understand you correctly -- and I'm not sure I do -- you still plan >> to segment your individuals and register their surfaces to some target. >> >> The question is what the target should be. I remain unconvinced that >> there is benefit to be gained by generating this customized target. I >> don't think tightened correspondence across nodes will offset the loss >> of benefit of using the PALS-B12 target (e.g., availability of several >> published parcellation schemes, borders, fMRI results from published >> studies, left-right correspondence). >> >> Now if you were not segmenting/registering your subjects, but rather >> just mapping their group results on some average fiducial surface, then >> I do acknowledge that a surface coming out of step 1 above would be >> closer to your subjects than the available PALS-B12 surfaces. But then >> your results would be less comparable to other studies than ones that >> mapped to these standard average fiducial surfaces. You see the >> trade-offs? >> >> Honestly, I don't think it's worth the effort, but I encourage my >> colleagues to speak up if they disagree. >>> Please correct me if I'm wrong. Thank you again, >>> >>> Bruno >> . >> _______________________________________________ caret-users mailing list [email protected] http://brainvis.wustl.edu/mailman/listinfo/caret-users
