Reza Pakdaman, M.D.
Postdoctoral Fellow
Dept. of Radiology- University of California , San Diego
Radiology Imaging Laboratory
3510 Dunhill St. MC 0852
San Diego, CA 92121
Phone:(858) 822-4404
Fax: (858) 534-6046
[log in to unmask]________________________________________
From: FSL - FMRIB's Software Library [
[log in to unmask]] On Behalf Of Jesper Andersson [
[log in to unmask]]
Sent: Tuesday, June 16, 2009 10:44 AM
To:
[log in to unmask]Subject: Re: [FSL] Hello!
Dear Reza,
We are processing DTI data in our LAB in University of california ,
San Diego with fsl software version 4.1.3 .Currently we are using
fslstats utility of your program. I have noticed a paradox in the
output of the program and hereby want to notify you about it :
I have used the following command on the FA map of the DTI of a
healthy subject processed by fsl :
fslstats 1_FAfsl.nii.gz -m -M -s -S -r -R
The FA values are between 0 and 1 . the results of the program are :
0.050226 0.212998 0.120878 0.165211 0.000000 0.487615
0.000000 1.195134
so in the output the program calculated the range for non-zero
voxels between zero and 1.195 which is not correct at all (0<FA
value <1)
FA values > 1 is a consequence of a negative eigenvalue of the tensor.
It is true that this is something that is "physically" impossible, but
in practice something that is frequently observed with noisy data.
Consider a voxel in an area with little/no signal. The true
eigenvalues should be [0 0 0] (i.e. we have no signal and no
diffusion). However, the estimated eigenvalues will be associated with
some uncertainty e, so you'll really observe [0ħe 0ħe 0ħe], and hence
some of those values will be negative leading to FA>1.
-in the second stage we used the internal capsule , and purely
internal capsule as the ROI and the results are as follows :
-fslstats IC_FA.nii.gz -m -M -s -S -r -R
0.000010 0.607764 0.002601 0.251947 0.565753 0.791101
0.000000 0.793482
So according to the program results the range of data f is 0.56 -
0.79 , the mean for non-zero is 0.607 , however the SD for non-zero
is 0.2519 .It looks completely unacceptable .
0.56 -- 0.79 is the "robust" range, i.e. ignoring the tails of the
observed distribution of values. The full range is 0 -- 0.79, the next
two values. And then an s.d. of 0.25 looks more plausible.
Good Luck Jesper