Dear John and SPM experts,
I thank John for fixing my_fseek errors during estimation by modifying spm_append.m script.
I have other questions about setting contrast and the result output of statistical inference.
I have 13 subjects with two conditions, test and retest. I actually have total 26 normalized images.
I used SPM99 to analysis these images using PET/SPECT models as:
Population main effect: 2 cond's, 1scan/cond (paired t-test)
#subjects=13
subject 1: select images: condition 1-2; I select test image for condition 1 and retest image for condition 2 on subject 1. I did the same work until subject 13.
GloNorm:Select global normalisation=no
GMsca: grand mean scaling =no
Threshold masking =absolute
analysis threshold=0
Global calculation =omit
2 conditions, 15 total, and 12 degrees of freedom from 26 images
estimate?=now: No error message
Statistical inference: Results
Contrast: 1 -1
corrected p value (p=0.05)
Because the condition 1 and 2 are almost the same, therefore, I expect not significant difference between two conditions, which can expect a reliability of perfusion measurement.
My questions are:
a) Is the contrast (1 -1) right to compare test images and retest images of 13 subjects?
b) From the "RESULTS in SPM99", I clicked "volume", however, the message is " no suprathreshold clusters" with T=7.26(p=0.000) and k=0 voxels (p=0.119). Does this mean no significant difference between test and retest images?
c) Based imaging artifacts of some of subjects, I may expect significant difference between test and retest images at some regions. How can I obtain "significant difference cluster map"? Do I just increase p value such as p=0.001 rather than p=0.05?
I appreciate any your comments and suggestions.
Geon-Ho Jahng, Ph.D.
UCSF/VAMC
