Just to be clear, I agree completely with almost all of what you've  
written.  But, for me, I have found it useful to use the T daemon.   
While I hope that someday I have a subtle sense of neuroanatomy, this  
is not a trivial matter to learn, especially for an outsider.  The T  
daemon lets me get a pretty rough sense of what I am looking at so  
that I can show my data to someone who knows the neuroanatomy after I  
have made some attempts at understanding what I am looking at.  My  
experience is that this has made collaborations far more useful than  
if I just bring the images to the neuroanatomist without me having a  
first clue (even if that clue is misguided).

Obviously, there are a lot of people coming to functional imaging who  
are like me and lack a precise training in functional neuroanatomy.   
I feel extremely lucky because I have had at least one formal course  
on the topic.  The ultimate solution is for people like me who are  
experts in other areas to be a part of teams that have experts in  
neuroanatomy.  But, the reason I am guessing that people are  
attracted to the T daemon is because neuroanatomy is very hard to get  
all at once (especially if you are not part of a neuroscience  
program), not because they are per se lazy.

Some of the methodologists I know in my field disdain the "canned  
packages" like SPSS, Stata, etc in favor of roll-your-own models that  
are tailored specifically to the issues raised by each particular  
data set.  And, sometimes they disdain those who use the canned  
models.  But, I would contend that we probably get more science  
accomplished with packaged approaches because they let people focus  
their energies on other issues.  Good scientists don't know  
everything themselves, but they know their limitations and that it is  
important to team with those who know the things that they are  
ignorant about.  The Talairach Daemon certainly has the limitations  
you point out, but it is probably worthwhile to note why people find  
it useful.

     Darren


On Feb 25, 2006, at 12:59 AM, Joseph Devlin wrote:

> Why would you want to convert MNI coordinates into T&T space?   
> Other than for using the Talairach daemon, which I'll return to  
> shortly, there are only disadvantages to T&T space.  First and  
> foremost is that it is not representative as it comes from a single  
> hemisphere of a single elderly woman whereas the MNI coordinates  
> are defined based on a decent sample size (either 152 or 305,  
> depending on the template you use).  The differences between the  
> spaces attest to the poorly representative nature of the T&T brain.
>
> In addition, the internationally accepted standard is the MNI brain  
> (see http://www.loni.ucla.edu/ICBM/).  The advantage to having this  
> standard is that different groups compare like with like, which is  
> why all of the major analysis packages use this template as their  
> default for registration/normalisation.  Converting coordinates to  
> T&T just makes results less comparable, particularly given the non- 
> standard conversions people use (as these don't convert in the same  
> fashion so there is no unambiguous method for recovering the  
> original MNI coordinates).
>
> As far as I can tell, the only "advantage" is using the T daemon to  
> determine the anatomy of individual coordinates and this is really  
> very difficult to justify for many reasons.  The main reason is  
> simply that almost all studies collect structural images for each  
> participant so they have the 100% definite anatomy -- there is no  
> need to look at some probabilistic database.  Best practice is to  
> look at the activations on the individual structurals (or a mean  
> structural in MNI space if looking at group results) and determine  
> the anatomical location based on the data itself.  I've found the  
> Duvernoy atlas to be excellent for helping with this and as someone  
> else mentioned, when in doubt talking to a colleague with greater  
> anatomical experience is also helpful.  In my opinion, the T daemon  
> (and other "automatic" labeling options) are fundamentally  
> inaccurate and basically lazy anatomy.  Since the purpose of fMRI  
> is to identify functional anatomy, it is crucial to the endeavor to  
> get the anatomy right.  *Rant over.*
>
> I'm not trying to harass anyone about this, but this topic comes up  
> again and again and I think it is worth considering these issues  
> very carefully when doing imaging.  fMRI is anatomical tool and  
> thus requires taking the anatomy seriously.  The T daemon (and  
> other similar approaches) are all grossly inferior to learning some  
> anatomy.  And besides, getting out the atlas and learning an area  
> seen in functional results is immensely useful.  It builds up  
> anatomical knowledge surprisingly quickly.
>
> Obviously, this is a position I feel strongly about but I'd be  
> curious to hear other opinions, particularly if there are any other  
> reasons for using T&T coordinates.
>
>
> - Joe
>
> -------------
> Joseph T. Devlin, Ph. D.
> FMRIB Centre, Dept. of Clinical Neurology
> University of Oxford
> Headley Way
> Oxford OX9 3DU
> Phone: +44-(0)1865-222-494
> Email:  [log in to unmask]