Hi --
We did a bit of a comparison of subject-to-atlas registration for different
atlases in a recent paper by comparing the groupwise alignment of
pre-segmented images:
T. Rohlfing, N. M. Zahr, E. V. Sullivan, and A. Pfefferbaum, “The SRI24
multi-channel brain atlas: Construction and applications,” in Medical
Imaging 2008: Image Processing, J. M. Reinhardt and J. P. W. Pluim, Eds.,
Bellingham, WA, 2008, vol. 6914 of Proceedings of SPIE, p. 691409.
The bottom line is that the difference in registration accuracy between
atlases acquired on different equipment is smaller than the difference
between a fuzzy and a crisp atlas. The MNI152, for example, is a rather
fuzzy atlas, whereas the ICBM452_warp5 is a rather crisp atlas.
The most interesting outcome was that our own atlas ("SRI24"), which is
based on data acquired at 3T, did not perform worse than all other atlases,
which were based on 1.5T data, even though the test data was all from 1.5T.
In fact, by some standards the 3T-based atlas outperformed the 1.5T atlases,
likely due to the fact that it was the only one created using fully nonrigid
unbiased groupwise alignment.
Best,
TR
--
Torsten Rohlfing, PhD SRI International, Neuroscience Program
Research Scientist 333 Ravenswood Ave, Menlo Park, CA 94025
Phone: ++1 (650) 859-3379 Fax: ++1 (650) 859-2743
[log in to unmask] http://www.stanford.edu/~rohlfing/
"Though this be madness, yet there is a method in't"
|