Hi Bas,
> That's interesting, I am actually trying this with a new dataset right
> now. It is high resolution 3T EPI data (2x2x2), eg with considerable
> distortion, and besides our time series data concerning a smaller FOV we
> also acquired one identically angulated 2x2x2 whole brain EPI. The
> contrast in that image is pretty good, it approaches T1 contrast at
> 1.5T. So for 9 out of 10 subjects it can be used for unified
> segmentation normalization. And, importantly, it has the same spatial
> distortion as the time series data (because we took grate care the
> angulation was identical), greatly improving coregistration.
> I am not sure however how well segmetation works on more common spatial
> resolution EPI data (eg 4x4x4), I think it is more problematic there.
Well, my idea would be to only use the bias field estimation which is
part of unified segmentation. In other words, I don't care about the
segmentation results as long as the bias field is properly estimated.
This likely is also better in hi-res/hi-contrast data but it may be
worth a try. However, unified segmentation of course takes too long to
run it on every single image of a timeseries, which would be optimal,
and to estimate a bias field from, say, a mean EPI and then apply it to
the other images in the timeseries disregards motion * B0 effects, among
others... tricky. One could also try to tweak segmentation to not do
segmentation but only bias correction, which I guess would speed things
up considerably. So much to do, never enough time :)
Best,
Marko
--
=====================================================================
Marko Wilke (Dr.med./M.D.)
[log in to unmask]
Universitäts-Kinderklinik University Children's Hospital
Abt. III (Neuropädiatrie) Dept. III (Pediatric neurology)
Hoppe-Seyler-Str. 1, D - 72076 Tübingen
Tel.: (+49) 07071 29-83416 Fax: (+49) 07071 29-5473
=====================================================================
|