Hi,
in the full-tensor decomposition (TICA), you can decide which one you want to flip: if you change the sign of the timecourses (Tmodes) OR the spatial maps (Bmodes), you can also change the sign of the subject loadings (Smodes) AND keep the Bmodes OR the Tmodes, respectively.
Cheers-
Andreas
-----Ursprüngliche Nachricht-----
Von: FSL - FMRIB's Software Library im Auftrag von Michelle Voss
Gesendet: Do 14.02.2008 00:39
An: [log in to unmask]
Betreff: Re: [FSL] Melodic output
hello,
just to double-check my intuition, since subject loadings are factor
loadings, they do not need to be flipped here right? e.g., they represent
the same thing for both the original and "flipped" maps and timecourses,
which is how much each subject contributes to the component.
thanks
michelle
On Feb 13, 2008 4:54 PM, Christian F. Beckmann <[log in to unmask]>
wrote:
> Hi
>
> red-yellow shows increasing positive regression coefficient with the
> associated time course while dark blue-light blue shows increasing
> negative regression coefficient wrt the time course.
> If your time course is the inverse of your canonical design then blue
> values who still show negative regression against the time course)
> will in fact have positively regression coefficient wrt the original
> design. That is, you can freely 'flip' the time course but then also
> need to 'flip' the spatial map to make sure that once you multiply
> these two together you get the same result (basically 1*1 = -1 * -1)
> In TICA the associated time course basically is the avergae temporal
> response across all the voxels (weighted y the vaue in the spatial
> map) and also averaged across the population according to the subject/
> session mode vector. Yes, you can trial average this time course and
> get a representation of the overall avergae (across space and
> population) average HRF shape.
> best
> christian
>
>
>
> On 13 Feb 2008, at 19:57, Vishwadeep Ahluwalia wrote:
>
> > Hi,
> > I need help interpreting the IC maps. I'm confused about the color
> > coding
> > scheme.
> > assuming my stimulus paradigm is such: 0001100011000
> > Instance 1: i get a component with time series 0001100011000. How do i
> > interpret the red and blue voxels?
> > Instance 2: i get a component with time series 1110011100111. How do i
> > interpret the red and blue voxels?
> >
> > As a separate question;
> > Using TICA on 10 datasets, i got the task related component. Is it
> > true that
> > the time series for this component represents the time course of every
> > activated voxel in the respective spatial map? if so, can i trial
> > average
> > this time course and validly estimate the shape of the BOLD response
> > in
> > those voxels?
> > These ten datasets belong to one paradigm. Can i do a similar
> > analysis on
> > data from another paradigm(with different stimulus timings) and
> > compare the
> > effect sizes between the two paradigms? or would i have to run one
> > single
> > analysis for both?
> >
> > Thanks!
> >
> > -Vish
>
|