We consider the application of segmentation based on cluster classification
techniques to a series of images derived from the diffusion tensor. The ex
tension of a sophisticated cluster simulation tool used for optimizing data
acquisition for such segmentation methods to diffusion-based images is des
cribed. The characteristics of a variety of diffusion-based images includin
g fractional anisotropy images, diffusion tensor trace images, and isotropi
cally diffusion-weighted images are considered and their application to neu
rological image segmentation is investigated. The critical effect of the si
gnal-to-noise ratio on fractional anisotropy is analyzed and limitations of
current echo planar-based strategies are discussed. Segmentation is shown
to be possible using only images derived from the diffusion tensor, and suc
h images are shown to offer exciting new avenues for neurological segmentat
ion. (C) 1999 John Wiley & Sons, Inc.