On 12 Apr 2013, at 01:30, Takuya Hayashi <[log in to unmask]> wrote:

Dear FSLers,
I read the TFM paper with great interest. I would learn how to generate TFMs using our resting-state fMRI data but cannot figure out how it could be done at the last step. The fastICA may generate group-wise 21 TFM time series using 142-node time series data, which was calculated by multi-session temporal concatenation mode of melodic. How this group-wise TFM time courses can be fed into regression of original single session data?

We used dual-regression (see FSL wiki) to get all the subject timecourses corresponding to the group-ICA spatial maps, loaded them into Matlab (you can use FSLNets for this), removed the "bad" group-level components, demeaned all time series, concatenated them across subjects, and fed them into temporal ICA using FastICA (find on the web). Note that it is really hard to get reliable components from temporal ICA applied to FMRI time series - you need LOTS of timepoints (we only just had enough for the published paper) and very clean data.

Steve.

Can it be done directly using fsl_glm? and/or Should I use weighting matrix (At) to generate single session TFMs using single session 142 spatial nodes? Could anyone give me some hints to do this?

Thank you in advance.

Best regards,

Takuya

--

Takuya Hayashi, MD, PhD
Functional Architecture Imaging Unit
RIKEN Center for Life Science Technologies, Kobe, Japan

---------------------------------------------------------------------------
Stephen M. Smith, Professor of Biomedical Engineering
Associate Director, Oxford University FMRIB Centre

FMRIB, JR Hospital, Headington, Oxford OX3 9DU, UK
+44 (0) 1865 222726 (fax 222717)
[log in to unmask] http://www.fmrib.ox.ac.uk/~steve
---------------------------------------------------------------------------