SPMers,
I have been conducting an spm analysis of a group of patients with
focal epilepsy against a group of 37 healthy normal controls. As I am
using interictal SPECT scans, I've been looking for regions of
hypoperfusion in the patients.
However, my analysis thus far has yielded mostly incorrect results
(the patients have had successful surgical outcomes, so we know where the
correct region of localization should be), and even those few patients for
whom we have obtained correct localization give us results at a set-level
of 0.998, which it seems to me implies a very poor statistical
significance.
Below is a brief outline of the procedure that I have used. I
would really appreciate it if any of you could give me some suggestions or
criticisms of the procedure that I have used. Thanks in advance.
Normalization:
I normalized the all of the healthy normals and patients
to the PET template (and using the same patient or control for both the
"derive from" and "apply to" groups). I also turned off the "affine
parameters," and made the number of "Z-basis functions = 1", and the
number of "nonlinear iterations = 0."
Smoothing:
All scans were smoothed to 10x10x12
Statistics:
I used the "Proportional Scaling" model with "Scaling of
Overall Grand Means." I used an "F-Threshold" of 0.001, 0.01, and 0.05,
to test the different levels of significance. And I used a contrast of -1
for the patients and a contrast of 1 for the controls (to test for
hypoperfusion in the patients). In the "select scans" area, I enterred
the smoothed & normalized scan for the patient of interest and each
control.
[It may also be noted that when I ran a statistical analysis of
the healthy normals against the healthhy normals (as a check), I obtained
results that were also at set-levels of approx. 0.998.]
Thanks,
Amar Krishnaswamy
Yale School of Medicine
New Haven, CT 06510
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|