Hi,
this happens in cases where the mixture-model based inference step
detects that there is only a single Gaussian distribution present
(i.e. the histogram of the IC does not have fat tails). In this case
both thresholding under the alternative hyp and under the null hyp
becomes valid and melodic generates the latter at a p<0.05 level in
addition to the normal alternative hyp test. There are different
possible ways why this might happen, e.g. SNR is very low or the
mixture model simply failed to find a good estimate - you can check
what's going on by having a look at the mixture-model fit (click on
the thresholded map and you'll see the mm report page)
hth
Christian
On 4 Jan 2008, at 15:00, Angel Wong wrote:
> Dear FSL users I have performed a model-free group analysis using
> multi-subject tensor ICA. The components can be generated properly.
> But when I looked at the results, I found some components contain 2
> thresholded IC maps. One is entitled as 'thresholded IC map
> alternative hypothesis test at p>0.95', another is entitled as
> 'thresholded IC map (2) 2-sided null hypothesis test at 0.05'. I
> would like to ask: 1) Why there are two thresholded IC maps
> generated for some of the components? 2) What is the difference
> between the two thresholded IC maps and their significance? Many
> thanks Angel
|