 |
 |
Hammers, Alexander wrote: > Hm - wasn't the reason to set the probability everywhere to > non-zero one or two SPM versions ago exactly that - if there really > was GM in unusual places (as happens regularly with epilepsy > patients' heterotopia, *particularly* around the ventricles), then > the clear message from the histogram could "override" the very low > but non-zero probability from the a priori maps? > > I for one would be very unhappy with a zero prior probability map > in these areas... and for subjects with non-malformed brains, I've > always found that the *tiny* rim of misclassified voxels around the > ventricles is a rather endearing feature that maps the outline > nicely. Excuse my ignorance, but in which application is this > really a problem? Hi Alexander, I think I agree with your first paragraph, but the rim of GM can be a problem if the ventricles are expanding (e.g. in AD) as it is seen as a false gain of GM. The spatial normalisation "bites" on the strong contrast between CSF and surrounding WM, and shrinks the ventricles of the AD subjects down to match the template. Because the spatial transformation model is fairly smooth (either DCT or even regularised HDW), it also ends up shrinking the rim of GM, which is subsequently "modulated" up in intensity, and e.g. appears as a gain in (misclassified) periventricular GM in Alzheimer's vs controls. If you had a group of subjects with typically larger ventricles and possible appearance of GM around the vents, and/or changes in the caudate, then it could get very confusing as to what was actually happening. Though, probably no less confusing than if the segmentation were forced to ignore GM near the vents, so again, on balance, I think I agree with your first point. Best, Ged.