Hi Megan,
The SSE is the sum of squared errors you make relative to each data
point (the log transformed diffusion data!).
In DTIFIT, the model used is: Sj = S0 * exp(-bj gj^TDgj)
where gj refers to the j-th bvecs, and D to the tensor.
You can easily re-write this as: Y = M*x
Where Yj=log(Sj)
and x=[log(S0) D11 D12 ...]
M is a matrix formed using the bvecs and bvals.
If you want an estimate of the standard error on your parameters (e.g.
the tensor elements, or the intercept), then you can use the following
formula:
std of element i in x = sqrt( Xii * SSE / (N-7-1) )
Where Xii is the i-th diagonal element of inv(M^TM) and N is the
number of data points (volumes in your dti data).
If you are familiar with C++, you can easily add these outputs to the
dtifit code (which you will find in $FSLDIR/src/fdt/dtifit.cc), in
the function "tensorfit"
Cheers,
Saad.
On 14 May 2009, at 21:45, Megan Twiss wrote:
> Hello Saad,
> Is the sum of squares output the cumulative SSE for each of the matrix
> elements? I need to know specifically what this SSE output is
> calculated from. I am trying to come up with errors on my eigenvalue
> estimates etc. Is there any way that I can see the source code for
> this?
> I'm assuming that this sum of square errors is only relevant for those
> who use multiple b-values or have multiple samples with the same b-
> value
> (otherwise we would get zero as SSE). I just need to know more
> clearly
> how this parameter is calculated from the tensor elements and output
> as
> a single SSE, otherwise I'll need to code my own software, and I
> really
> don't want to re-invent the wheel, if you know what I mean!
>
> Thanks,
> Megan
>
Saad Jbabdi
Oxford University FMRIB Centre
JR Hospital, Headington, OX3 9DU, UK
+44 (0) 1865 222545 (fax 717)
www.fmrib.ox.ac.uk/~saad
|