Dear dr. Bencherif,
the following may shed some light on the ROI-SPM issue.
Both methods are intended to detect a spatial pattern (of a parameter, covariate,
contrast or whatever) under controlled risk of false positives. It helps if you realize
that they both consist of 2 steps.
1) A filter
2) A thresholding procedure
ROI:
1) Filter:
You have some expectations on location and size of the signal. Therefore you sample the
image with a template of ROIs. This is a filtering operation as you apply a mean filter
of a defined shape (the ROI shapes) in specific locations.
2) Thresholding:.
You can then derive your statistics and run the tests (say t-tests). Unless ROIs are
very small and adjacent, spatial correlation among them should be negligible and
therefore a simple Bonferroni correction will suffice for the multiple comparisons
issue.
SPM:
1) Filter:
You have no clue on location of your signal. You have some sense on its size though.
Therefore you select a smoother (say 12 mm ) and apply it to the entire volume.
In case you had no knowledge on the size either, you should use a range of smoothers.
2) Thresholding.
You are testing all the sites of the volume; adjacent sites are spatially correlated.
To control the risk of false positives you apply a threshold derived from gaussian
field theory.
These two methods are very different as they employ different a priori knowledge. A
"comparison" would not make much sense ( a sort of banana vs. apple problem).
They have the same specificity (Type I error) but different sensitivity depending on
the type of signal you have.
In short, if you have expectations on where your signal is, you are better off using
ROIs. If you don't, use a smoother or a range of smoothers.
Regards
Federico
-------------------------------------------------------------------
Federico E. Turkheimer Ph.D
PET Methodology Group
MRC Cyclotron Unit
Imperial College School of Medicine
Hammersmith Hospital
DuCane Road
London, W12 OHS
tel: 0208-383-3451
fax: 0208-383-2029
email: [log in to unmask]
WWW: http://www.cu.mrc.ac.uk/~fet/
-------------------------------------------------------------------
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|