The extremum stack, as proposed by Koenderink, is a multiresolution im
age description and segmentation scheme which examines intensity extre
ma (minima and maxima) as they move and merge through a series of prog
ressively isotropically diffused images known as scale space. Such a d
ata-driven approach is attractive because it is claimed to be a genera
lly applicable and natural method of image segmentation, The performan
ce of the extremum stack is evaluated here using the case of neurologi
cal magnetic resonance imaging data as a specific example, and means o
f improving its performance proposed, It is confirmed experimentally t
hat the extremum stack has the desirable property of being shift-, sca
le-, and rotation-invariant, and produces natural results for many com
pact regions of anatomy, It handles elongated objects poorly, however,
and subsections of regions may merge prematurely before each region i
s represented as a single node. It is shown that this premature mergin
g can often be avoided by the application of either a variable conduct
ance-diffusing preprocessing step, or more effectively, the use of an
adaptive variable conductance diffusion method within the extremum sta
ck itself in place of the isotropic Gaussian diffusion proposed by Koe
nderink.