Print

Print


Hi


On 3 Apr 2015, at 09:02, Anderson M. Winkler <[log in to unmask]> wrote:
Hi Marie-Helene,
As far as I know, the derivatives are implemented as confound EVs and are orthogonalised wrt to the main EV,

right - I don't remember now if the derivatives are explicitly forced to be orthogonal to the "main" EVs - but in practice they will be either exactly orthogonal or v close.

and thus are computed, interpreted, and passed to higher levels in the same way as ordinary EVs for which the derivatives weren't included. Being derivatives, they have the interesting property that, as the fit happens, they act as if shifting (time) and scaling (dispersion) the respective main EV(s).

Almost - in FEAT only the first derivative is added on via the simple button-press option - but not also the second derivative (which would account for dispersion/scaling).

Cheers, Steve



The contrasts are t-tests on the main EV(s), unless you select "Real EVs" and enter them as F-tests.

The excerpt you found refers to an example in which a set of basis functions are used, which is a different thing.

All the best,

Anderson


On 2 April 2015 at 10:14, Marie-Helene Grosbras <[log in to unmask]> wrote:
Hi,
I am slightly confused about the way temporal derivatives are dealt with in FSL
(and consequently about other sets of basis functions).

When  temporal derivatives are added to EVs in the model specification and  the (default) option of working with original EVs is selected, then contrasts are specified as if only one regressor was entered, and the help
file states that this “will result in a single [1] contrast for each resulting real EV, and then an F-test across these”.
If this is the case, it makes sense but then I am wondering
-       what does the tstat map (corresponding to this contrast) represent (how is
this computed).
-       Which contrast is computed for a contrast specifying a difference between 2
original EVs (each having a temporal derivative). Is this the F test across the
2 contrasts [EV1_canonical]>[EV2_canonical] and [EV1_deriv]>[EV2_deriv]? or
across the 4 possible contrasts?
-       What is passed to the second level? According the the tutorials
I understand that it is the estimate for the canonical part of the EV? But then
I am not sure to which files this corresponds to, nor why....

Also I read in a recent paper (Pernet C Frontiers 2014) that the temporal
derivatives are orthogonalized with respect to all the other EVs, but could not
find any confirmation/explanation of that in the FSL help....

Any clarification would be welcome!
Many thanks



---------------------------------------------------------------------------
Stephen M. Smith, Professor of Biomedical Engineering
Associate Director,  Oxford University FMRIB Centre

FMRIB, JR Hospital, Headington, Oxford  OX3 9DU, UK
+44 (0) 1865 222726  (fax 222717)
[log in to unmask]    http://www.fmrib.ox.ac.uk/~steve
---------------------------------------------------------------------------

Stop the cultural destruction of Tibet