FS-FAST registration across sessions - Freesurfer

20 Jan 2006

      Hi
My question was asked last year by Tommi Raij (see email pasted below) but I 
couldn't find a reply to it in the archives. I am analysing retinotopy data 
that was collected over several sessions but I cannot see how to register 
scans from multiple sessions together, and to the subject's anatomical scan. 
Although Fs-Fast appears to combine the data from multiple sessions during the 
retinotopy analysis it only seems to be possible to use one registration file 
(register.dat) when overlaying the results onto the anatomical. I would have 
thought that registering scans from multiple sessions together would be a 
feature that many people would require. Is there anyway to do it? Tommi gives 
some possible methods in his email below.
When I do try and overlay my data onto a surface I see 'gaps' in the 
blue/yellow colouring of the fieldsign map overlaid onto the cortical surface. 
I wonder if this could be due to the inaccuracy of the registration given that 
I only used the register.dat from the registration of the first session.
hope that makes sense!
i'd be grateful for any help with this!
Jane
-- 
Dr Jane Aspell
Department of Experimental Psychology, University of Oxford,
South Parks Road, Oxford, OX1 3UD  
-------------------------------------------

[Freesurfer] FS-FAST Registration Question: registration across sessions 
within subjects
Tommi Raij
Thu, 20 Oct 2005 15:43:46 -0700

Hi Doug/All,

This is a question about registering several functional scans from the same 
subject in the same space. The functional scans are from different sessions, 
so even if the slice recording parameters are identical, the head will 
inavoidably be in different locations and with different rotation. The data 
have already been collected, so we cannot use any on-line corrections. 

Four different techniques come to mind that could be used to achieve this, but 
I would expect that the accuracy of the methods varies a lot. Optimally IMHO 
one should register the different functional runs with respect to each other 
directly, simply because the EPI images are very similar with respect to each 
other (or at least their T1 contrast is). The high-resolution structural image 
is very different compared to the EPI images so functional-to-structural 
registration is expected to be much more inaccurate. 

Here are the four different techniques (please tell me if you can think of 
more): 

1. One possibility would be to use mc-sess for all runs from the same subject 
(place all the runs from the different sessions in a single "bold" directory 
and just run mc-sess). Possible problem: If I have understood correctly, mc-
sess only corrects for motion within slices. Therefore the different head 
rotations would make this technique very inaccurate. Or does there exist an 
updated version of mc-sess that could 
handle even large displacements and rotations?

2. A second possibility would be to use one of the automatic functional-to-
structural registration programs and register each functional run to the same 
hires structural image. However, based on the discussion above, there are 
expected to be inaccuracies in the functional-to- 
structural registration that could be large enough to make comparisons
across the functional sessions meaningless (i.e., the differences would
mainly be driven by inaccuracies in the registration).

3. A third possibility would be to use the same technique as for registration 
across subjects, i.e., resampling the different runs from the same subject 
into a spherical space calculated from this specific subject. Again this is 
just as variety of registration from functional to structural, and would be 
expected to provide suboptimal results. 

4. A fourth possibility would be to find a method for registering all the 
functional images in a single space with respect to each other, e.g. by 
picking one of the functional sessions as the target space and resampling all 
other functional sessions from this subject into this space (without using the 
hires structurals at all). I guess this would be kind of motion correction but 
capable of handling rotations. 
If such a tool exists, it might be more accurate than any of the methods
above, simply because the functional images are more similar compared to
each other than compared to the subject's hires structural image.

Any ideas/suggestions which method would give the most accurate result? And 
what kind of tools currently exist? 

Thanks,

Tommi

---
Tommi Raij
MGH/MIT/HMS Athinoula A. Martinos Center for Biomedical Imaging
Bldg 149, 13th St
Charlestown, MA 02129
U.S.A.