hi Ben
> I have a group of controls and patients that I have collected two
> anatomical sessions for. Is there an argument for averaging the T1
> scans prior to statistical analysis using VBM as opposed to using each
> scan as an individual subject, thus all subjects essentially are
> represented twice? Statistically speaking, would it make a large
> difference in the result to lean in favor of one method over the
> other?
I'm assuming that the sessions were collected nearby in time, and you
are thinking of including both to increase the signal-to-noise ratio?
Statistically speaking it shouldn't matter, because if you include 2
images per subject, you will need to account for subject effects in
your model, so you'll have the same degrees of freedom; so I think the
issue is just whether averaging 2 scans gives you a more accurate
segmentation than using a single scan.
I think most people get decent results using a single T1 image; I've
heard that for some scanners (e.g., especially 1.5T machines with a
single channel) this is not always the case. I don't have experience
using multiple scans per subject myself, so maybe others are better
able to comment. However, I suspect there's a tradeoff of potentially
getting better SNR, on the one hand, and dealing with imperfect
registration between the two scans, on the other, that might introduce
a tiny bit of blurring. Without knowing anything about your scans my
advice would be to try segmenting a single T1, and if those
segmentations seem reasonable, don't worry about the second
acquisition.
> One other question..Can anyone provide an argument for using the
> Jacobian tissue probability modulation in DARTEL? My preliminary
> result came out looking very odd and as I do not know much about this
> Jacobian transform I wonder what more useful information (or increase
> in SNR) can be drawn using this transform compared to standard tissue
> probability maps.
There's a succinct explanation of this in:
Mechelli A, Price CJ, Friston KJ, Ashburner J (2005) Voxel-based
morphometry of the human brain: Methods and applications. Current
Medical Imaging Reviews 1, 105-113.
http://dx.doi.org/10.2174/1573405054038726
Basically, adjusting for changes in size (i.e. modulation) gives you
information about the total volume, whereas not adjusting gives you
information about the local concentration. The modulated images are
therefore most comparable to other methods (e.g. volumetry done by
hand-tracing) and easier to interpret (at least for me!). What you
use depends, of course, on the specific question you are asking, but I
suspect you would want to use the modulated images.
Hope this helps,
Jonathan
|