 |
 |
The coregistration needs to be accurate for the anatomical image to be of any use for spatial normalisation. If it is distorted then a rigid registration will not get a good match. Another reason why the coregistration may not work is that there may be a strong bias artifact in your anatomical image - in which case you can do a bias correction (via the segmentation) and then move all your fMRI data to align with the bias corrected anatomical image (coregister). You may be lucky and have EPI data that have a similar intensity distribution to those of the EPI template. This may mean that least squares may be a good alignment model. In general though, it is hard to guess what works well for other people's data. When we do make recommendations, they are usually on the basis of what we have found to work better here in the FIL. Best regards, -John On Tuesday 30 September 2008 13:37, Lucas Eggert wrote: > Dear List, > > I just wanted to make sure that the solution to my problem is not obvious - > or anybody else already found the answer - before going deeper into this > topic: > > The SPM5 manual explains that the default spatial preprocessing of > fMRI-data should involve coregistering the structural file to the mean > EPI-image. After this step you are suppose to segment the structural image > and subsequently apply the resulting transformation-matrix to all EPI's. > The reason for this procedure should obviously be the better normalization > that one should get. I processed one of the SPM-example-datasets according > to this procedure and compared the normalization quality to that of the > same dataset processed the 'old-fashion'-way, i.e. simply taking the > EPI-time-series and normalize it directly to the EPI-template. Results (in > terms of a correlation-analysis) show nearly no difference in > normalization-quality. Now, applying the same test, i.e. comparing the > SPM5-procedure to the 'old-fashion'-way to one of our own datasets yields > in a big difference, but interestingly the SPM5-procedure is worse! > > My questions are: > (a) > What exacty is the reason, that we should use the normalization parameters > of the structural file to transform the EPI's into standard-space? > > (b) > What are the conditions for this procedure to yield in better results than > the normal normalization procedure? Is it the protocols parameters? Is it > the initial correspondence between structural and functional files, and if > so, which factors do influence this correspondance (protocol again?) and > how could one measure the quality of this correspondance? > > Thank you very much! > -Lucas > > P.S. If needed, I can of course supply more details of my tests so far. > ________________________________________________________________________ > "50 erste Dates" mit Adam Sandler u. Drew Barrymore kostenlos anschauen! > Exklusiv für alle WEB.DE Nutzer. http://www.blockbuster.web.de