Hi Bruce,

Re your 2nd point: My assumption was that very few FS users are actually creating their own atlases. To do that you need to do (much) more than just run make_average_subject, right? (Namely, create your own .gcs file for annotations, and .tif for registration, correct? And isn't that quite an involved process?)

Re your 1st point: As for a truer anatomical visualization (without deriving your own atlas), it seems that is predicated on the assumption that the buckner40 based registration is representative of the population under study in the first place (so that the subsequent averaging to the buckner40 target is indeed appropriate for the population under study). And, if you're making this assumption, shouldn't the average white, pial, curv, sulc, etc. files derived from buckner40 by extension already be "representative" as well? In which case, I'm still left wondering why bother with making your own average subject? It seems that unless you go all-out and create your own atlases, you'll always be left wondering if the registration was truly appropriate for the population under study, and thus by extension whether the "average" surfaces are truly "representative" as well.

(I guess that technically, the key assumption is that the sulcal pattern, and its variance, are consistent across populations in order for the buckner40 based registration to be appropriate as the basis for creating your own average subject. But again, without ever actually computing your registration atlas, how would you ever know if this was in fact true?)

thanks for helping me understand this, Mike H.

On Thu, 2008-03-27 at 13:16 -0400, Bruce Fischl wrote:

Hi Mike,

a couple of things. One is that creating your own average will give you a better ability to see what the true anatomical localization of your effects are. Another is that you can re-register your subjects to your own atlas, and possibly obtain better registration, particularly if your population is substantially different from the ones in our atlas (e.g. young kids maybe).

cheers, Bruce

On Thu, 27 Mar 2008, Michael Harms wrote:

...

Hello, There have been numerous posts recently related to the creation of a study specific "average subject". I have a question about the purpose/motivation of this for surface-stream analysis/visualization that I was hoping someone could clarify.

Specifically, my understanding is that the annotation/labels for any derived study specific average subject are determined by the "buckner40" gcs file (for the case of the desikan atlas). And similarly, the average surface (e.g., white, pial) and curvature-format files (e.g., curv, sulc, thickness) are registered across subjects via sphere.reg, which is determined by ?h.curvature.filled.buckner40.tif?

So, since the annotation and registration for any study specific "average subject" is going to be determined by the "buckner40" set of subjects regardless of the specific subjects in a particular study, what is the purpose, advantage, or necessity for creating a study specific average subject? Or stated differently, since the whole FS approach for registering surfaces (and by extension parcellating the cortex) is based on the assumption that the "buckner40" set of subjects is an appropriate registration and parcellation target for ALL studies, what is the point of creating a study specific average rather than using the default 'fsaverage' (which has the advantage of being created from the same set of 40 subjects that drive the registration itself)?

Please help me understand what I'm missing here.

thanks, Mike H.

P.S. Am I correct in presuming that the same 40 subjects were used for the Desikan GCS atlas and for deriving the spherical registration (.tif) target?

Good points! I (partly) see the light :)

Still (please correct me if I'm wrong), creating a study specific average subject for visualizing the "average" anatomy doesn't address the issue of whether the method by which those subjects were brought into register (i.e., via sphere.reg using a term in the energy functional based on the buckner40 sulcal patterns) is indeed "appropriate" for the subjects under study.

Which leads me to inquire: Do you know if anyone has looked at the "robustness" of the buckner40 registration target relative to other study populations and demographics? (I imagine that this would have to be an empirical question).

Anyone out there that has created their own registration target? If yes, and you're willing to share, I think this (or any other alternative atlases such as customized volume gca, or surface gcs classifiers) would be a great contribution to the FS community. I know users have queried about the possibility of creating their own targets and/or classifiers, but has anyone actually created one or is in the process of doing so??

cheers, Mike H.

On Thu, 2008-03-27 at 14:21 -0400, Bruce Fischl wrote:

Hi Mike,

imagine you had only on subject your study. Then using the buckner atlas as an overlay (or fsaverage, which is kind of equivalent) would not be representative at all. If the registration works perfectly and you have a fair number of subjects then there probably isn't much difference, but there's really no reason *not* to visualize on the average anatomy of your subjects, except to make cross-study comparisons easier.

cheers, Bruce

On Thu, 27 Mar 2008, Michael Harms wrote:

...

Hi Bruce,

Re your 2nd point: My assumption was that very few FS users are actually creating their own atlases. To do that you need to do (much) more than just run make_average_subject, right? (Namely, create your own .gcs file for annotations, and .tif for registration, correct? And isn't that quite an involved process?)

Re your 1st point: As for a truer anatomical visualization (without deriving your own atlas), it seems that is predicated on the assumption that the buckner40 based registration is representative of the population under study in the first place (so that the subsequent averaging to the buckner40 target is indeed appropriate for the population under study). And, if you're making this assumption, shouldn't the average white, pial, curv, sulc, etc. files derived from buckner40 by extension already be "representative" as well? In which case, I'm still left wondering why bother with making your own average subject? It seems that unless you go all-out and create your own atlases, you'll always be left wondering if the registration was truly appropriate for the population under study, and thus by extension whether the "average" surfaces are truly "representative" as well.

(I guess that technically, the key assumption is that the sulcal pattern, and its variance, are consistent across populations in order for the buckner40 based registration to be appropriate as the basis for creating your own average subject. But again, without ever actually computing your registration atlas, how would you ever know if this was in fact true?)

thanks for helping me understand this, Mike H.

On Thu, 2008-03-27 at 13:16 -0400, Bruce Fischl wrote:

...

Hi Mike,

a couple of things. One is that creating your own average will give you a better ability to see what the true anatomical localization of your effects are. Another is that you can re-register your subjects to your own atlas, and possibly obtain better registration, particularly if your population is substantially different from the ones in our atlas (e.g. young kids maybe).

cheers, Bruce

On Thu, 27 Mar 2008, Michael Harms wrote:

...

Hello, There have been numerous posts recently related to the creation of a study specific "average subject". I have a question about the purpose/motivation of this for surface-stream analysis/visualization that I was hoping someone could clarify.

Specifically, my understanding is that the annotation/labels for any derived study specific average subject are determined by the "buckner40" gcs file (for the case of the desikan atlas). And similarly, the average surface (e.g., white, pial) and curvature-format files (e.g., curv, sulc, thickness) are registered across subjects via sphere.reg, which is determined by ?h.curvature.filled.buckner40.tif?

So, since the annotation and registration for any study specific "average subject" is going to be determined by the "buckner40" set of subjects regardless of the specific subjects in a particular study, what is the purpose, advantage, or necessity for creating a study specific average subject? Or stated differently, since the whole FS approach for registering surfaces (and by extension parcellating the cortex) is based on the assumption that the "buckner40" set of subjects is an appropriate registration and parcellation target for ALL studies, what is the point of creating a study specific average rather than using the default 'fsaverage' (which has the advantage of being created from the same set of 40 subjects that drive the registration itself)?

Please help me understand what I'm missing here.

thanks, Mike H.

P.S. Am I correct in presuming that the same 40 subjects were used for the Desikan GCS atlas and for deriving the spherical registration (.tif) target?