Dear SPM developers and users:
My study is looking at the regional tissue growth during normal brain
development. I calculated the annualized tissue change map based on the
Jacobian map of the deformation field between the two time points. Then I
examined the mean at each voxel using t-test and see if it is away from
zero.
I use SnPM to correct for multiple comparisons across the whole brain
volume (7880599 voxels). So basically, I used the design scheme for
Multisubject: 1condition 1 scan per subject (13 subjects in total). The
results obtained are: Maxt = 14 & rank 1 out of 8192 completed
permutations; at FEW =0.05, voxel-wise statistic threshold = 6.74342.
Then I took this threshold to filter the t-map and found very few regions
reach the threshold. By doing the variance smoothing, the threshold went
down to 5.31 but it is still quite high. Is this the right way of doing
voxel level inference?
Since the study is normal brain growth, I expect to see more diffuse but
relatively slow growth over the entire brain especially in white matters.
So I am thinking maybe I will gain more power and sensitivity by using the
second level inference – cluster size. I searched over the SPM email
archives. It seems like I could use SnPM to perform second level
inference. But I could not find out how to do it.
I also tried to use the “voxel cluster result” option. After I pushed the
button, the only file popped up was the SnPM.m instead of SnPM_ST.m. I am
using Matlab 7 but I downloaded the bug fix for it (SnPM_pp.m3.60 and
SnPM_combo_pp.m3.22) . So I am not sure what else needs to be changed.
Thanks in advance for your time and help!
Best,
Xue
|