Mark,

> Hi,
> I have a followup to the response below concerning scaling the PEs to
> account for the amplitude of the columns in the design.mat.
>
> Namely, if FILM has run correction for auto-correlations, then isn't
> there technically a different design matrix for every voxel?

Yes, that is true - for the purpose of estimation there is a different
design matrix at each voxel.

> So wouldn't
> the PE corrections would need to be voxel-wise? But there is only one
> design.mat written out, implying global auto-correlation correction.

Local autocorrelation correction is carried out in FILM. A global design
matrix is written out which represents the GLM to be fit. The
autocorrelation estimates at each voxel are written out in a file
threshac1, these can be used in combination with the design matrix to get
the local prewhitened design matrix used at each voxel for estimation
(these are not written out so as to save disk space).

> What is  the right way to re-scale the PEs?

Note that the prewhitening design matrix (and indeed the prewhitened data)
are products of the _estimation_ being used, but the model being inferred
upon is still the original non-prewhitened design matrix. Hence, the PEs
obtained are estimates on the original non-prewhitened design matrix.
Hence, you should base any PE rescaling on the original global design
matrix.

Best regards, Mark Woolrich.

> On Thu, 9 Dec 2004 01:26:31 +0000, Stephen Smith
> <[log in to unmask]> wrote:
>
>>Hi - in order to interpret the absolute size of estimated PE values,
>> you need to first multiply them by the peak-to-peak height in the model
> (i.e.,
>>look inside design.mat) - after doing this, regardless of the
> convolution
>>type, you know the fitted peak-to-peak height. In none of this, at this
>> point, are things normalised into percentages (etc.).
>>
>>Cheers, Steve
>>
>>
>>On Wed, 8 Dec 2004, Ramapriyan Pratiwadi wrote:
>>
>>&gt; Hi,
>>&gt;
>>&gt; We recently ran an analysis using the FIR basis set, rather than
> the the
>>&gt; double-gamma HRF.  To our surpise, the range of values for the
> pe images
>>&gt; were much much lower than expected; in fact, they look like
> percent
>>&gt; signal change.  And when the betas for the basis set were
> graphed, they
>>&gt; indeed looked like a typical HRF.
>>&gt;
>>&gt; Do the pe images need to be scaled in some way, or are they
> already in
>>&gt; percent deviation from baseline.
>>&gt;
>>&gt; Thanks,
>>&gt;
>>&gt; Ram
>>&gt;
>>&gt;
>>&gt;
>>&gt; ------------------------------------
>>&gt; Ram Pratiwadi
>>&gt; Research Assistant
>>&gt; University of Pennsylvania, Brain Behavior Laboratory
>>&gt;
>>&gt; --
>>&gt; This message has been scanned for viruses and
>>&gt; dangerous content by MailScanner, and is
>>&gt; believed to be clean.
>>&gt; MailScanner thanks transtec Computers for their support.
>>&gt;
>>
>>--
>> Stephen M. Smith  DPhil
>> 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