Increasing spatial and spectral image resolution, coupled with increased expectations both in visualization output and in the time-responsiveness of that output relative to certain clinical tasks have led t o a c omcomitant increase in the ro l e o f p arallel and distributed solutions to the visualization problem. In this paper we examine the distribution of classi cation and visualization tasks over a heterogenous distributed a r chitecture i n t h e c ontext of one speci c visualization problem viz. the directed visualization of small-scale structures from the Visible Human Dataset. Speci cally, we examine a Challenger-like architecture in which tasks are dynamically bid onto and transported a c r oss the fabric under the control of agents.
1: IntroductionThe clear trend in imaging physics is is towards the generation of highly spatially-resolved image datasets coupled with increasing spectral resolution. Added to this is an increased awareness of the value of sophisticated visualization techniques incorporating not only conventional surface/volume visualization technologies, but directed visualization (by which we mean visualization of in-image en tities which h a ve been identi ed by classi cation or segmentation algorithms 1,2]. The literature is lled with technical descriptions both of classi cation strategies 3,4] and visualization algorithms 5,6], and although they form the core problems addressed architectiurally in this paper a discussion of them in any but the most super cial terms is not appropriate here.A motivating example su ces: Figure 1 sho ws a visualized (surface) model of the urethra from the Visible Female Dataset 7]. The model is generated in a largely unsupervised process which i n volves the pre-segmentation of (about) 200 slices from the VF dataset using 3 unsupervised (pre)classi ers, followed by a (repeated, multi-seeded) Gibbs classi cation whose output is unioned, which generates a segmented version of the dataset. This segmented version of the dataset is aggregated into connected components, one of which i s t h e urethral lumen shown. This point-set geometric dataset is fused back i n to the original VF dataset which is stripped and smoothed using a volumetric Gaussian k ernel, and the output of this is in turn fed to an isosurface algorithm implemen ted in the Visualization Toolkit (vtk) 8], which is a commonly a vailable suite of imaging and visualization algorithms.The schema below illustrates the primary points of parallel activity the left bracket in-1 0-7695-0484-1/00 $10.00 ã 2000 IEEE