Thank you John for such a rapid response! The script works now for batches in SPM8b - attached in case anyone wants it. I will look into BET or maybe using an explicit mask that excludes meninges when it comes time to analyze...but as you said, we do not expect any systematic bias across time (or between patient groups).
Another DARTEL question: For making the between-subject Template, would it be valid to run DARTEL on 4 batches of 25 subjects, and then DARTEL those 4 Templates to each other (and calculate the composition), rather than performing DARTEL on 100 subjects at once? If this is equivalent (or "equivalent enough"?), we could run on multiple processors and significantly cut the processing time.
Thanks again,
Dana
-----Original Message-----
From: John Ashburner [mailto:[log in to unmask]]
Sent: Wednesday, November 19, 2008 11:13 AM
To: Dana Perantie; [log in to unmask]
Subject: Re: [SPM] longitudinal DARTEL Qs
> I want to do longitudinal VBM and came across this very interesting and
> helpful line of discussion between Reinders & Ashburner:
> http://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=ind0804&L=SPM&P=R48484
>
> We have 100 subjects at 2 timepoints. I have begun to process the images
> in the way described and have some questions:
>
> 1) With cross-sectional VBM, we like to use the "thorough clean" option
> during Segment. Otherwise we see classification of meninges and other
> non-brain as gray matter. After importing *sn.mat into DARTEL, it does not
> look like the "clean" info came along. The inclusion of meninges can
> differ across time (attached = 1 subject after DARTEL import, baseline on
> top and follow-up on bottom,). Couldn't this subject's gray matter volume
> be misinterpreted as shrinking over time near those areas? Is it possible
> to clean before or after DARTEL import? For this subject, even after
> "clean" there are more meninges at baseline than follow-up. Any
> suggestions?
You are correct. GM volume could be interpreted as changing if some subjects'
data have meninges, and others don't. However, if this difference is not
systematic (e.g. early scans all have meninges in the GM, but late scans
don't), then it shouldn't show up in the results.
Some people have tried using e.g. BET, prior to using the SPM5 Segmentation.
This may (or may not) help. Also, there is a slightly experimental
Segmentation toolbox in SPM8b. It is essentially the same algorithm as
described in the Unified Segmentation paper, but extended slightly to make
use of additional tissue probability maps, along with a bit more flexibility
in the warping. It does not differ enough from the old implementation to
write anything up about it. The implementation could benefit from having
some better tissue probability maps, which I hope to get around to generating
eventually. It is possible (but not certain) that this implementation may
deal better with the meninges.
>
> 2) For inhomogeneity correction -- before the initial import into DARTEL,
> would it help to write out the bias-corrected images (during Segment) and
> then segment the bias-corrected images? Or is the bias correction
> information included in *sn.mat and subsequent DARTEL-imported images?
The bias correction is incorporated into the model when the data are imported.
The basis function coefficients that parameterise the bias field are saved in
the seg_sn.mat file.
>
> 3) The script posted by Reinders for making a batch for within-subject
> DARTELing is going to be very helpful. I am able to run it when I created
> the example job file "oneBLFUpair.mat" in spm5, but the structure and name
> of "jobs" has changed in spm8b ("matlabbatch"), and I'm having trouble with
> this line: data_start = jobs{1}.tools{1}.dartel{1}.warp.images
Sorry about the name change. Try changing the line to:
data_start = matlabbatch{1}.spm.tools.dartel.warp.images
Best regards,
-John
|