Dear Thomas
the spatial maps calculated in melodic are the projections of the data
(pre-processed within melodic, i.e. de-meaned in time and space and -
unless switched off - normalized by the voxel-wise standard deviation
[variance normalization]) onto the estimate of the unmixing matrix. This
is analogous to standard GLM, i.e. C=pinv(A)*Y where C ~ spatial maps, A
~ mixing matrix and Y ~ matrix of the individual voxels time courses.
Thus, the individual spatial maps are not simple correlation maps of the
voxels' time courses and the IC time course but the result of multiple
regression -- just like the parameter estimates in std GLM.
WRT the color coding: the color scheme is similar to MEDx color coding;
'hotter' colors (dark red to bright yellow) correspond to a voxel being
positively affected by the associated time course and 'cooler' colors
(dark blue to light blue) correspond to a voxel being negatively
affected by the associated time course. One of the fundamental
ambiguities of the decomposition is the sign, i.e. by inverting the time
courses and the corresponding spatial maps you obtain different color
coded maps.
You can threshold these maps but should be aware that thresholding is an
inferential step, which should only be done after considering the
distribution of the intensity values in the IC maps. You also need to
make sure that the clusters you are interested in are not coded as being
negatively affected by the time course and then vanish due to
thresholding.
The next version of melodic will allow to infer the 'optimal' number of
components to be extracted and will also perform IC map thresholding.
thanks
Christian
Thomas Stephan wrote:
>
> Dear List,
>
> I have some questions regarding the output of melodicreport:
> How do I have to interpret the colors in the output of melodicreport ?
> Can they be interpreted as a coding of correlation coefficients ?
> Can I threshold the output to see only the voxels above a certain degree of
> correlation ?
>
> Thank you for your help
> Thomas Stephan
>
> ======================================================================
> Thomas Stephan Email: [log in to unmask]
> Klinikum Grosshadern
> Neurologisches Forschungshaus
> Marchioninistr. 23 Fon: +49 089 / 7095-4819
> 81377 Muenchen Fax: +49 089 / 7095-4801
> ======================================================================
--
----------------------
Christian F. Beckmann
Address: Oxford University Centre for Functional
Magnetic Resonance Imaging of the Brain,
John Radcliffe Hospital, Headington, Oxford OX3 9DU, UK
Email: [log in to unmask] Phone: +44(0)1865 222704 Fax: 222717
|