Hi All,
I don’t know if this question is useful, but I thought I would try. I’m running 3dpc on 315 volume dataset with ROIs that are about 10-30 voxels in subject space. The basic 3dpc command I ran is this:
3dpc -prefix “test1” -pcsave 1 -vmean -nscal -mask mask.nii.gz 4dfunc.nii.gz
The output is similar to below no matter the roi I use:
#Num. --Eigenvalue-- -Var.Fraction- -Cumul.Fract.-
1 1820.601 0.1344821 0.1344821
2 1515.083 0.1119144 0.2463965
3 1256.588 0.09282023 0.3392167
4 1192.299 0.08807138 0.4272881
5 1018.198 0.07521106 0.5024992
6 829.2309 0.06125268 0.5637518
7 757.5972 0.05596133 0.6197131
8 682.5212 0.0504157 0.6701288
9 657.3946 0.04855967 0.7186885
10 582.4139 0.04302109 0.7617096
11 576.2439 0.04256533 0.8042749
12 534.1663 0.03945719 0.8437321
13 501.874 0.03707186 0.880804
14 463.1735 0.03421317 0.9150172
15 436.9332 0.03227489 0.9472921
16 377.7954 0.02790656 0.9751986
17 335.7578 0.02480138 1
18 1.331998e-05 9.839054e-10 1
19 1.048822e-05 7.747322e-10 1
20 1.001387e-05 7.396931e-10 1
I don’t really see a sharp drop-off in variance explained per component, is this a typical or atypical result? When would I be justified in using the 1st eignenvariate to represent the timecourse of the seed (e.g. are there rules of thumb for how much variance the first eigenvariate has to explain in order to represent the timecourse?)
Thanks!
James