voxel significance vs whole brain pattern significance

[ab698]

Hi Chris
I know we've had this conversation before but can you unpick the
following statement from the help manual for me

"Generally, if the loading of a voxel in a covariance pattern is
robustly positive ( > 2 on the ICV map), its activity in the brain is
localized and increases with task difficulty level. If the loading is
robustly negative ( < ‐2 on the ICV map), its activity is localized
and decreases with task difficulty. If the loading is not robust,
there is no localization at the voxel but the voxel could nonetheless
contribute to the overall covariance pattern across task levels in a
dispersive fashion in conjunction with other voxels within the
pattern."

When the bootstrapping finishes and there is a statement saying
'extrema on the current ICV map is -1.88 +1.78' what does that mean
exactly ? I understand that the maximum loading on some voxel within
the image is +1.78 but I'm not entirely sure how you get that
'loading' measure from the resampling that you do and what it is a
measure of exactly (amount of covariance with other voxels?) and
anyway why would we want a significance measure for a single voxel
since its the whole pattern we're interested in isn't it ?

Many thanks
Anna

[Anna Barnes]

Thanks Chris.

The reason I ask is that it looks like I've got some interesting results for my genes and brain structure analysis but I'm not sure how to display them now.

Do I use the behavioural best fit images and no threshold at all but then show the R2 graph figure as well?

Sorry I've got myself a bit confused now :(

cheers

A

On 25 Nov 2008, at 04:42, Christian Habeck wrote:

 --- the forwarded message follows ---

From: "Christian Habeck" <cha...@sergievsky.cpmc.columbia.edu>

Date: 25 November 2008 04:41:09 GMT

To: gc...@googlegroups.com

Subject: Re: voxel significance vs whole brain pattern significance

Hi Anna,

I think you are on the money, and this issue needs clarification. I'll try to improve the documentation and make clear that

(1) the Z-map from the bootstrap is for visualization purposes only,

(2) all information about covariance and pattern loadings are contained in the point estimate image. This image also, can be prospectively applied to other data sets to obtain a network score.

------------------------

You are right, the point estimate of the covariance pattern consists of ALL loadings. The bootstrap procedure is only a ill-fitting attempt to use parametric maps and come up with a inferential statistical statement on a voxel-by-voxel basis. The question to be answered is: does this voxel's loadings have a positive (or negative) value in more 95% of the bootstrap samples. This is tailored for statements about single brain regions and the desire to reduce the pattern to some few key regions, similar to an SPM{T} map. (Are those regions reliably identified with a consistent sign in their loadings??)

Overall, the bootstrap should be de-emphasized for analysis purposes. The important covariance pattern consists of ALL voxels and loadings, whether they have a low Z-value or not.

The reduction to a few areas in a sense runs totally counter to the whole idea of covariance or multidimensional analysis in general.

I'll try to make this distinction between the covariance pattern and the bootstrap map clearer- they're different things. At the moment the aspect of prospective application is missing. Also, my RA is a very smart guy, but sometimes has a stilted English that's worse than mine.

Thanks for your input and maybe you could keep reading and pointing out passages that need clarification.

Chris

PS: More tomorrow

ch...@columbia.edu

Hi Chris

I know we've had this conversation before but can you unpick the

following statement from the help manual for me

"Generally, if the loading of a voxel in a covariance pattern is

robustly positive ( > 2 on the ICV map), its activity in the brain is

localized and increases with task difficulty level. If the loading is

robustly negative ( < ?2 on the ICV map), its activity is localized