The general approach to spatial normalization using a deformation fiel
d is presented. Current high degree-of-freedom deformation methods are
extremely time-consuming (10-40 hr), and a k-tree method is proposed
to greatly reduce this time. A general k-tree method for analysis of s
ource and target images and synthesis of deformation fields is describ
ed. The k-tree method simplifies scale control and feature extraction
and matching, making it highly efficient. A two-dimensional (2-D), or
quadtree, application program was developed for preliminary testing. T
he k-tree method was evaluated with 2-D images to test rotating abilit
y, nonhomologous region matching, inner and outer brain-structure inde
pendence, and feasibility with human brain images. The results of thes
e tests indicate that a three-dimensional (3-D), or octree, method is
feasible. Preliminary work with an octree application program indicate
s that a processing time of under 10 min for 256(3) image arrays is at
tainable on a Sun Ultra30 workstation. (C) 1998 Wiley-Liss, Inc.