Dear Steven
> When I realign, I usually write only the mean image, so that
> resampling/reslicing is done only once, at the spatial normalization stage.
Fair enough. But please be careful when using words like 'reslicing' and 'resampling'. I use the term 'reslicing' for writing a new file, so that a .mat files are not necessary anymore. I use 'resampling' for accessing a volume via a newly produced .mat
file.
> However, my impression is that to create the mean image, *all* volumes must
> be resliced, even if they're not individually written out.
If you want all images to be aligned, you'll need to register all of them individually. This doesn't mean you have to write all the resliced images, it's enough to produce .mat files.
> Is this so? If it is, will using trilinear interpolation be much faster
Trilinear is faster than sinc, but also less accurate
> than sinc? My idea: sinc is preferable to trilinear, but here I'm only
> creating a mean image in order to determine parameters for normalization,
> and not as much accuracy is needed for that. (I plan on using sinc for the
> reslicing of the individual images after spatial normalization.)
This is an idea. But indeed, it only makes sense if you want to use the coregistration to provide an estimate for normalisaton. In that case, you might make a mean image by adding up the volumes, and do a registraion of that image. That is, if the
individual images do not differ too much in orientation. And of course, if that is the case, you might as well use 1 (representative) image to get that estimate.
I hope this answers your question.
Best regards,
Alle Meije Wink
--
Alle Meije Wink
Institute for Mathematics and Computing Science, room 105
University of Groningen P.O. Box 800
9700 AV Groningen
Telephone +31 50 363 71 27 The Netherlands
Fax: +31 20 875 48 00
E-mail: [log in to unmask]
WWW: http://www.cs.rug.nl/~wink
|