pengxu wei wrote:
> the reason why low variance areas outside the brain can have falsely
> high t-values?
T-values are ratios of signal to noise, where "noise" is quantified by
the standard error of the signal of interest. Far from the brain,
where both the signal and the noise are very low, this ratio becomes
rather unreliable -- a random tiny signal can result in a large
t-value if the variance is even smaller.
To check if this is happening, you could look at the con images, as
well as the spmT images. The con image is just the "signal" instead of
the SNR, if it is very small outside the brain then you can assume
that any high t-values are a result of the above numerical problem
with the ratio. If there is large signal, then you might have another
problem that should be investigated.
Best,
Ged.
|