On Mon, 4 Jan 2010 18:42:26 +0000, Amit Etkin <[log in to unmask]>
wrote:
>Hi all-
>
>This is a question that has come up periodically, and I've never seen a
convincing answer
>for....but why isn't something like AFNI's alphasim used more frequently with
SPM analyses?
>...it seems to be pretty intuitive, and an empirically-driven cluster-level
correction, derived
>from simulations that take into consideration voxel size, smoothness, etc.
>
>thanks and happy new year!
(1) Heresy of using a competing neuroimaging software package (mostly
kidding :-) )
(2) User must know enough to be able to glean what inputs to give alphasim
from SPM
My impression is that alphasim gives "better" results (i.e., the thresholds are
more liberal) than the SPM cluster-level correction. It's possible that that's
because alphasim uses a joint peak height-extent threshold, which is perhaps
more powerful. On the other hand, alphasim makes certain assumptions which
might not be correct; in particular, IIRC it uses a Gaussian random field. Some
relevant posts to the SPM mailbase (perhaps somewhat redundant):
https://www.jiscmail.ac.uk/cgi-bin/wa.exe?
A2=SPM;MIgFGQ;20080218113511%2B0000
https://www.jiscmail.ac.uk/cgi-bin/wa.exe?A2=SPM;jqlr6w;20010203003210-
0500;ind01
https://www.jiscmail.ac.uk/cgi-bin/wa.exe?
A2=SPM;MIgFGQ;20080218113511%2B0000
https://www.jiscmail.ac.uk/cgi-bin/wa.exe?A2=SPM;ITML2w;20051122081729-
0600;ind05
https://www.jiscmail.ac.uk/cgi-bin/wa.exe?A2=SPM;ITML2w;20051122081729-
0600;ind05
It's not clear to me how badly off the assumptions in alphasim are.
>Amit
|