Brett (and any statistician who would like to add their thoughts):
I wonder if you could comment about the use of uncorrected thresholds in
combination with extent thresholds.
My lab routinely screens (and often publishes) results using an uncorrected
intensity ("voxel-wise") threshold of p<0.001 in combination with an extent
threshold of 10 voxels (applied to images re-sliced at 3x3x3 mm voxels and
smoothed with either a 7 mm or 10 mm Gaussian kernel). We find that this
pretty well eliminates activation outside of any area that we could expect
to find activation on theoretical grounds. Another lab on campus has run
monte carlo simulations on various intensity / extent combinations (using
AFNI), and found a false discovery rate < 0.05 whenever the uncorrected
Z-statistic is greater than about 4.00 -- which almost exactly corresponds
to the Z-statistic values we find with our standard intensity/extent threshold.
If uncorrected p < 0.001 is valid when applied to a single voxel (without
examining other brain voxels), the chance of reaching this threshold for 10
contiguous voxels (before correcting for multiple comparisons) should be
roughly (0.001)^10 -- which would survive the FWE correction. This is not
really statistically accurate, though, in part because the "activation" of
contiguous voxels is influenced by the smoothing process. Even after
accounting for smoothing effects, though, the chance of 10 contiguous
voxels reaching some statistical threshold is clearly far less than the
likelihood of a single voxel reaching that threshold.
Since SPM does not include a toolbox for monte carlo simulations that can
examine the likelihood of various intensity / extent combinations, I wonder
whether there is any statistical argument which can be used to justify a
combined intensity/extent threshold -- particularly one that would satisfy
reviewers.
Doug Burman
At 02:26 PM 2/24/2007 +0000, Matthew Brett wrote:
>Hi,
>
>I've summarized the arguments here:
>
>http://imaging.mrc-cbu.cam.ac.uk/imaging/UncorrectedThreshold
>
>Best,
>
>Matthew
Dept. of Communication Sciences & Disorders
Northwestern University
2240 Campus Drive
Frances Searle Building, Room 2-356
Evanston, IL 60208
phone 847-467-1549
fax 847-491-4975
email: [log in to unmask]
|