Hi Dani,
There are many different reasons why more than 6 directions are needed to
perform diffusion MRI analysis. First of all, 6 comes from the DTI tensor
model and is the min number of directions needed to estimate the 6 unknown
elements of the tensor. However, if you run bedpostx a different model is
fitted. This has 5 parameters (including b=0) for modelling 1 fibre, 8 for 2
fibres, 11 for 3 fibres and so on. So, you would need at least 5,8,11
directions respectively. Now we usually acquire much more than that, some of
the reasons are:
1) The more directions you have the more reproducible and robust your
results will be. Imagine that you want to fit a line through some
datapoints. You can do it with 2 points, however if you repeat the
experiment and measure two new points, the new fitted line will be very
sensitive to any perturbations of these points. If instead you use 10
points, the fitted line will be much less sensitive to small perturbations
of each point, i.e. your results are much more robust.
2) Quantitative metrics, like the FA and MD, need at least >20-25 directions
to be rotationally invariant (Jones, MRM, 2004), if you do DTI analysis
(i.e. dtifit). This means that with 6 directions for example, the estimated
FA and MD can depend on the way the subject's head is placed in the scanner,
relative to the reference coordinate system.
3) You need many directions and high angular resolution in your acquisition
to resolve fibre crossings (see last figures in Behrens, NeuroImage, 2007)
Hope this helps,
Stam
----- Original Message -----
From: "Daniel Simmonds" <[log in to unmask]>
To: <[log in to unmask]>
Sent: Friday, January 14, 2011 8:35 PM
Subject: [FSL] tractography with low (<25) directions
Hello,
I am currently working with a dataset that has 6 directions (+B0) and am
interested in tractography. I have run some of the subjects through bedpostx
and probtrackx. The analysis ran with no problems, and the results were
reasonable looking and quite robust. I have searched for some explanation
for this recommendation, and have found very little. Can anyone give me a
rundown of the rationale for this recommendation, and whether/why my results
would be invalid/unreliable (and does this depend on whether the data is
primarily being used for individual analyses or group analyses)? Thank you.
Dani
|