Dear Haibo,
Just to get you started, under the assumption that the signal as a
function of TE is
y(TE) = C*exp(-TE/T2) + noise
, then a naive approach would be to take logarithms to obtain
ln(y(TE)) ~ ln(C) (-1/T2)*TE
which would likely be a good approximation if exp(-TE/T2) >> variance
of noise for all TE. If so, it would be reasonable to fit a linear
model via ordinary least squares with ln(y(TE)) as the dependent
variable and TE as the independent variable. The estimated slope of
this model would be (-1/T2); the intercept would be ln(C). Again, this
is not guaranteed to be reasonable because of the additive noise in
the original model (which would not behave well under logarithmic
transform if the inequality above is not true). I would look in Haacke
(Magnetic Resonance Imaging: Physical Principles and Sequence Design)
or some other good reference to understand more principled approaches.
Eric
Quoting "<Haibo> <Xu>" <[log in to unmask]>:
> Dear Spmlists,
>
> Sorry for bothering you again. I post the massage before and none answers
> it. We can get T2-map with two echos using ImgCalc, but do not know how to
> fit and get T2-map with more than 2 echos. Any one would like to give us a
> scrpit or suggestions with more than 2 echos will be greatly appreciated.
> Thanks in advance.
>
> Haibo
>
>
|