Dear Community,
I want to ask a little about the general opinion about EPI ghosting in
relation to the specificity (type I error) of our SPMs.
I had a little look at an experiment with a few epochs of reversing
checkerboard versus fixation. I wanted to look for false positives in
the frontal lobe where I expected the ghost of the visual cortex to
appear. In four sessions I found on the average 5 false positives
surviving the corrected 0.05 threshold. Although this particular
experiment (with a very strong activation in an area whose ghost
overlaps with the brain) is a worst case, I still found the result quite
disturbing. It should be mentioned that although the ghosts were clearly
visible, I wouldn't consider them "extreme".
What are the opinions regarding what constitutes "acceptable" levels of
ghosting. Clearly too much ghosting makes a mockery of all our attempts
to control type I errors, effectively rendering all tests "omnibus
tests" with no localising power. But what is too much?
I would also very much like to hear of the experiences of anyone who has
worked on reducing ghosting on a GE Signa scanner. It is my (layman)
understanding that the pre-scan as implemented on the standard EPI
sequence estimates constant (predominantly timing error?) and linear
(predominantly eddy currents constant over time?) phase shifts, and that
these are incorporated into the reconstruction. Is there an opinion on
the source of the residual ghosts. Are they caused by inadequacies in
the existing correction of the constant and linear phase shifts, or are
they caused by non-linear (time varying eddycurrents and/or deviations
between the requested and actual shape of the read-out gradient)?
I guess it all boils down to; Should I worry? If yes, does anyone have a
good solution?
Jesper Anderson
|