Posted by Gary Turner on October 19, 2001 at 15:45:58:
I sent a message ealier re clustering and have run into one more snag.
I am running a standard boxcar design: 4 conditions vs. rest; 5 subjects; I have done individual subject fims (for each condition vs rest), talairached, smoothed and then used this as input into a 3danova2 - mixed design (conditions x subjects) with the outputs as factor 'a' (conditions)means.
All I need now is a cluster report. Thanks to Jim's earlier advice here is my command line for 3dclust:
3dclust -1dindex [data sub-brick] -1tindex [t-stat sub-brick] -noabs -1thr 4.45 [t-value; p < .0001] -1erode 50 -1dilate 1.9 [rmm] 65 [vol] [fbuc dataset] > [output text file]. NOTE: I'm presently doing this from memory so there may be slight errors in my recollection of the command line - but the command line appears to be working well.
Here is the problem: my cluster ouput does not match the afni display for the same dataset. I am keeping all of the parameters the same but clearly significant clusters of activation (large activation areas that are well above t-threshold) do not appear in the 3dclust report. Which output do I trust - the display or the 3dclust? Any advice as to how to reconcile them would be most appreciated.
Thanks. Gary