Aha - yes, there is a simple explanation.....you have found a flaw in the
concept of % change for the higher-level analyses - it won't give sensible
results, because the "% change" is calculated by dividing the effect size
(pe value, assumed to be multiplied by a waveform of height 1) by the
baseline signal:
At first level this baseline signal is sensible - it is the mean FMRI
intensity (which tends to be around 10000 because of grand mean scaling).
At second level this "baseline" is the mean pe (or cope) that was fed into
the higher-level analysis, the % isn't very useful. You would need to turn
off the % button and instead divide the resulting values by the original
baseline values.
Thanks for pointing this out - we'll add a warning about this in future.
Thanks, Steve.
On Tue, 12 Aug 2003, Goekoop, R. wrote:
> Okay, thanks. The first level design was a block design in this case, with
> minimal amplitude of -0.53 and max. amplitude of 0.57 for copeX
> (design.mat). This probably reflects values between 0-1 demeaned. At second
> level, EV values were either -1 or 1 for the contrast of interest (formed by
> comparing series of the first level copeX images taken under varying
> conditions). Thus, 2 when undemeaned.
>
> Without the option "Convert PE/COPE values to %" highlighted, Featquery
> produces the mean value 14.96 at a specified coordinate (min 0, max 14.96).
> With this option hghlighted, a contraintuitively high value of 49.36 is
> reported as percentual signal change. If I understand correctly, by
> calculating % signal changes, absolute values are made relative by dividing
> them by the global mean value of a time series. In this case, this mean
> value would be 0.30. This seems pretty strange to me. FSL view does not
> report a value of 14.96 at the same specified coordinate, thus I don't
> understand what is going on. Would there be a simple explanation for this?
>
> Thanks again, Rutger.
>
>
> -----Original Message-----
> From: Stephen Smith [mailto:[log in to unmask]]
> Sent: dinsdag 12 augustus 2003 12:40
> To: [log in to unmask]
> Subject: Re: [FSL] Rescaling Higher Level Featquery Results
>
>
> Hi - yes, that all makes sense. The Featquery manual page at
> http://www.fmrib.ox.ac.uk/fsl/feat5/featquery.html refers to this issue:
>
> "If you select Convert PE/COPE values to %, any PE or COPE parameter
> estimate or contrast values will be converted to percentage change values
> before reporting. This is achieved by dividing the PE/COPE values by the
> mean image from filtered_func_data. Warning: this % is based on the
> assumption that the "height" of the model waveform is 1, which in general it
> is for FEAT-created block designs, but in general is not for event-related
> designs or custom waveforms. In order to get a true % change value you must
> multiply the output by the height of the relevant model waveform (see the
> design.mat file). In the case of contrasts (COPEs), the interpretation of
> this % needs even more careful thought."
>
> Hope this answers the question - look at a first-level design.mat (ascii
> file) to find out what the bottom-peak height of the event-related waveform
> turned out as. I assume that the second-level design IS of height 1, in
> which case no further correction is needed.
>
> Thanks, Steve.
>
>
>
> On Tue, 12 Aug 2003, Goekoop, R. wrote:
>
> > Dear all,
> >
> > I have just run a Featquery of a higher-level cope1.feat directory.
> > I'm interested in the percentual signal change values for certain
> > stats images, so I highlighted the option "Convert PE/COPE images into
> > %". I then got a value of 45.6% signal change for the contrast of
> > interest, which of course is way too high. The problem of course is
> > that the higher-level model is event-related and not a block design,
> > so I need to rescale the data. Is there any reasonable way of doing
> > this (since the model looks very random and quite complicated)?
> >
> > Thanks,
> >
> > Rutger.
> >
> > Drs. R. Goekoop, MD.
> > Department of Neurology
> > Vrije Universiteit Medical Centre
> > P.O. Box 7057, 1007 MB
> > Amsterdam, the Netherlands
> > Phone: +31 20 444 0316
> > E-mail: <mailto:[log in to unmask]> [log in to unmask]
> >
> >
> >
>
> Stephen M. Smith MA DPhil CEng MIEE
> Associate Director, FMRIB and Analysis Research Coordinator
>
> Oxford University Centre for Functional MRI of the Brain
> John Radcliffe Hospital, Headington, Oxford OX3 9DU, UK
> +44 (0) 1865 222726 (fax 222717)
>
> [log in to unmask] http://www.fmrib.ox.ac.uk/~steve
>
Stephen M. Smith MA DPhil CEng MIEE
Associate Director, FMRIB and Analysis Research Coordinator
Oxford University Centre for Functional MRI of the Brain
John Radcliffe Hospital, Headington, Oxford OX3 9DU, UK
+44 (0) 1865 222726 (fax 222717)
[log in to unmask] http://www.fmrib.ox.ac.uk/~steve
|