| 摘要 |
<p>A parallel processing architecture and method, based on the 3D Algebra Reconstruction Technique (ART), for iteratively reconstructing from cone beam projection data a 3D voxel data set (V) representing an image of an object. The cone beam projection data is acquired by employing a cone beam x-ray source and a two-dimensional array detector to scan the object along a source scanning trajectory to obtain, at each of a plurality of discrete source positions (Υi) on the source scanning trajectory, a 2D measured cone beam pixel data set (Pi). The 3D voxel data set (V) is subdivided into or organized as a plurality (m) of independent voxel subcubes (V0) through (Vm-1) each containing a plurality of voxels. As a result of the subdivision of the 3D voxel data set (V) into voxel subcubes, the 2D measured cone beam pixel data set (P¿i?) (measured projection data array) is correspondingly subdivided for each source position (Υi) into projection data subsets, with overlapping regions between one or more adjacent projection data subsets corresponding to adjacent voxel subcubes. Each voxel subcube and its corresponding projection data strip or subset is processed in parallel with other pairs of voxel subcubes and corresponding projection data strips or subsets, without interference. A bi-level parallel-processor architecture is employed for iterative reproject and merge, and split and backproject operations.</p> |
