Dear Experts, 

I have done an unpaired t-test TBSS analysis comparing a control group and another group with a genetic disorder (20 in each group).  I used TFCE to analyse the results.  I appreciate that TFCE can be very sensitive to diffuse changes, and indeed, my corrp results are very extensive consisting of one large contiguous cluster even at p<0.01.  Though it includes those areas we hypothesised would be affected, other areas are also affected.  This is useful but I want to hone in on those areas of biggest difference.  How should I improve my "p resolution" in order to break this cluster up and localise the biggest differences ?  My thoughts include :

I tried deriving TFCE t-stat image with -R option in randomise and then masked this with binarised, threhsolded (0.95) TFCE_corrp_image - having done this I have more "resolution" in the results and can nicely visualise the biggest differences  using FSLview with appropriate colour map, and have more leeway with thresholding to isolate the biggest differences - however even with this increase in "resolution" I still have 2 clusters 1000+ voxels big at stringent thresholds - how can I break down this cluster objectively in order to interpret it more easily ?

Which of the following is appropriate ?

1) I used cluster with --in as TFCE_corrp at 0.99 and --cope as masked TFCE t-stat image as - the number of local maxima derived is the same as without the --cope image - is it right therfore that the number of local maxima is not affected by the t-stat image ?

2) I tried clustrer with --in as masked TFCE t-stat image at various thresholds - is this valid ? - at most this seems to increase my main clusters by 1 and local maxima number is  the same

3) In view of the above fact is it reasonable to start changing cluster options in order to see if one break down the cluster at hand 

eg -connectivity option from 26 to 6 ?
eg change --peakdist  - although this does not appear to make a big difference 
eg. --num to increase local maxima  

OR is the point that even if I can break down the cluster into a greater number of main clusters or local maxima clusters it is not going to be any easier to interpret given the extent of the differences seen in which case is the following valid ...

4) Carry out the TFCE analysis to demonstrate widespread changes in patients and but then use a cluster extent/mass based approach with a threshold of 4 (chosen empirically)- when I do this the results (as expected) are more focal, and therefore easier to interpret.

So basically is it valid to include 2 methods of thresholding results in one analysis/write-up like this, the TFCE option showing the widespread nature of the changes, and the cluster option showing where the greatest focal changes are.

Thanks.

Mahinda