 |
 |
Hi, If you feed all_FA (non-skeletonised) into randomise you will get a VBM-style analysis (there's no good reason for using GRF to do the thresholding, as opposed to randomise; there's various pitfalls with GRF in that context). If you do that then there is the danger of the alignment problems and interpretability that are discussed in the original TBSS paper. Smoothing before randomise in that scenario helps with some of those issues, and makes others worse. In this case yes you can use TFCE, with the -T option not --T2 as this is now a "3D" analysis. Your histogram may well be ok - the peaks probably represent separate 'clusters' in the p image. Cheers. On 15 Feb 2009, at 10:50, Yuzheng HU wrote: > Dear Prof. Smith, > Thanks for your reply. I think I did not express my idea clearly > about raw FA > statistic in the last email. > > ...... >>> # Section 3, statistic for raw FA in TBSS/stas?directory >>> # threshold FA data by 0.2 >>> fslmaths mean_FA.nii.gz thr 0.2 mean_FA_mask_2.nii.gz fslmaths >>> # make a binary mask >>> mean_FA_mask_2.nii.gz bin mean_FA_mask_2_bin.nii.gz >>> #make inference, is this appropriate as follow? >>> randomise -i all_FA -o FA2 -m mean_FA_mask_2_bin -d design.mat >>> -t >>> design.con -c 1.6839 -V >>> # statistic for FA end >> >> You mean you want to compare the mean-across-space values in the >> groups? Not quite, not, I would apply the mask (that's already done >> for you hence doesn't appear below) and then summarise over space, >> e.g. > > No, My purpos is to compare non-skeletonised FA for all voxels in > the brain > across all subjects between two groups. Though TBSS can deal with > alignment > issue, there have been a great may papers used GRF theory to make > inference for non-skeletonised FA ,e.g. FA maps got from DTIFit, > between > groups. To compare with results from GRF, I wonder if I can fed a 4D > non- > skeletonised FA map without spatial smooth (the fourth demension is > subject > ID) to randomise to obtain inference for FA? If this makes sense, my > procedure > above is appropriate? shall I still use TFCE option? > > I have another question. I have used --T2 option in TBSS. i.e. > randomise -i all_FA_skeletonised -o tbss -m mean_FA_skeleton_mask -d > design.mat -t design.con --T2 –V > It produced (1-p) maps with multiple comarsion corrected, pleaes see > the > attachment tbss_tfce_corrp_tstat2.gif. The histogram of > tbss_tfce_corrp_tstat2.nii has many sharp peaks. Is this distribution > reasonable? > > Thanks a lot. > > Yuzheng > <tbss_tfce_corrp_tstat2.gif> --------------------------------------------------------------------------- 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 ---------------------------------------------------------------------------