External Email - Use Caution
Hi experts,
Upon reading your 2018 paper (cited below) and following threads on the archives, I am aware that I should use permutation testing for lGI analyses rather than monte carlo at p<0.05. However, I have a few questions below that I would really appreciate your input on:
1. For permutation testing on lGI data, what are the CFT and cwp levels you recommend? Would it be appropriate for me to use CFT=1.3 and cwp=0.05?
2. Will you confirm my interpretation for permutation command line below? mri_glmfit-sim --glmdir lh.lgi.TD_PTSD_age_SA_IQ_FWHM0.glmdir --perm 1000 2 abs --perm-resid --cwp 0.05 --2spaces --bg 1
Above command indicates that I used a cluster-forming threshold of p<0.01 and a cluster-wise significance threshold of p<0.05 - meaning that clusters are initially obtained using p<0.01 (CFT) but I report them in my paper if they meet additional cluster-wise probability of p<0.05 (cwp). Both are two-tailed (since I am using abs). Is this interpretation correct?
3. Would it be okay to correct with monte carlo using CFT=3 and cwp= 0.05 (below command line) instead of permutation testing for lGI data? mri_glmfit-sim --glmdir lh.ASDvsADHD.glmdir --cache 3 abs --cwp 0.05 --2spaces
3. I am not smoothing my lGI data (as lGI is a relatively smooth measure and to keep this consistent with the literature I have reviewed), however, after reading your 2018 paper, I wanted to ask if this is appropriate? Your paper recommends CFT<0.001 and FWHM>10mm for surface area analyses. Do I have to smooth lGI when doing a) permutation testing or b) monte carlo correction (steps 1 and 3 above)? If yes, what level of smoothness do you recommend?
4. After correcting with monte-carlo at p<0.05 - When I use the below command line to see my corrected results in freeview, I notice the significant cluster is gray rather than blue (I know it should be blue because I've ran this same analysis using QDEC). In order to see this cluster's true colour (and thus be able to interpret my results) do I have to change the threshold by clicking ‘show outline overlay’ then clicking ‘configure’ and setting the min value to 1.3? Is this the correct way of visualizing my corrected results?
vglrun freeview -f $SUBJECTS_DIR/fsaverage/surf/lh.pial:overlay=lh.lgi.TD_PTSD_age_SA_IQ_FWHM0.glmdir/group.diff/cache.th13.abs.sig.cluster.mgh:overlay_threshold=1.3,5:annot=lh.lgi.TD_PTSD_age_SA_IQ_FWHM0.glmdir/group.diff/cache.th13.abs.sig.ocn.annot -viewport 3d
reference Greve DN & Fischl B. 2018. False positive rates in surface-based anatomical analysis. NeuroImage.
Thank you so much,
Avi