The bias is modelled as Gaussian random noise, which is smoothed by a broad
Gaussian. This is then exponentiated and multipled by the image. The weight
of the regularisation relates to the inverse of the noise variance, and the
FWHM relates to the width of the Gaussian smoothing. If there is little or
no bias in your data, then you can enter a large FWHM. This also makes the
procedure faster, as fewer basis functions are used.
Best regards,
-John
On Friday 19 October 2007 12:32, Simone Reinders wrote:
> Dear list,
>
> For a VBM analysis I am working with data that is already intensity
> non-uniformity corrected using the N3 function
> (http://www.bic.mni.mcgill.ca/software/N3/). To avoid double bias
> correction or an interaction between the two bias corrections of N3
> and SPM5 I would like to optimize the setting for bias regularisation
> and bias FWHM within SPM5.
>
> Based on the emails in the SPM mailbase and the SPM-manual I
> understood that if data has little inhomogeneity the setting for the
> bias regularisation would be to allow for only very little
> flexibility, i.e. using the 'extremely heavy regularisation' setting
> in my case.
>
> However, I find it difficult to understand what to do with the bias
> FWHM. Do I choose a broader cutoff (i.e. a cutoff of 150 mm), or 'no
> correction' at all in combination with the 'extremely heavy
> regularisation'?
>
> Thank you in advance,
>
> Simone Reinders.
>
>
> http://www.jiscmail.ac.uk/cgi-bin/wa.exe?A2=ind03&L=SPM&P=R93815&I=-3
> http://www.jiscmail.ac.uk/cgi-bin/wa.exe?A2=ind05&L=SPM&P=R202337&I=-3
> http://www.jiscmail.ac.uk/cgi-bin/wa.exe?A2=ind06&L=SPM&P=R98508&I=-3
> http://www.jiscmail.ac.uk/cgi-bin/wa.exe?A2=ind0701&L=SPM&P=R40918&I=-3
> the spm-manual (section 5.3.6 and 5.3.7).
|