Parallel implementation of octtree generation algorithm

Octtree data structure represents 3D objects as a disjoint union of cubes of varying sizes. The octtree encoding provides an efficient object representation of 3D objects in medical imaging also for storage of planner objects, both in image space and object space with a definite advantage of space saving. Although 2D display technique from a transactional CT scan is quite adequate for diagnosis, it does not optimally communicate the 3D nature of the involved anatomy or the full extent of pathology; 3D imaging presents complex anatomical findings easily, helps in surgical planning and is also used to display radiation beams and isodose surface for radiation therapy planning. Typically, a single CAT scan results in as many as 277 slices (spaced at 1 mm), each slice having 256×256 points (spaced at 1.2 mm×1.2 mm). The image space of (2ⁿ×2ⁿ×2ⁿ); n is an integer, is reasonably homogeneous and suggests itself the use of octtree encoding for storage. The process of constructing octtree from CAT scan slices has to start by looking at intensity distribution per slice and form possible cubes of larger size from voxel cubes generated for each plane. A processor managing P planes (P=2^p; p is an integer) can be “farmed” to different “worker” nodes to generate the tree up to the level p or octternary codes for level (0,1,...p) independently and communicate with their parent worker/farmer in the “tree” topology to generate the tree for higher levels. The i-o access time and communication harness is minimized in this transputer model as against an earlier report, since it communicates with each other via their communication links and memory shared between worker and parent worker/farmer. T9000 virtual channel processor would allow any number of logical channels to be multiplexed into four physical communication links. The paper addresses this issue for an efficient implementation on transputers