A 33MHz 16 .. Bit Gradient Calculator for Real-Time Volume Imaging Martin Margala, Nelson G. Durdle. Scott Juskiw, 1 V. James Raso, and Doug L. Hill 2 L Electrical Engineering Department, The University of Alberta Edmonton, Alberta, Canada T6G 2G7 2. Rehabilitaion Engineering Department, Glenrose Rehabilitation Hospital Edmonton, Alberta, Canada T5G OB7 Abstract second image update rate. Jackel and Strasser [5) described the P ARCUM n system. which is used for handling solids This paper describes a gradient calculator which forms an reconstructed from CT scans. An emulation of this system important part of a shading processor being developed for a generated medical images in 38-110 seconds from data sets high resolution high performance real-time general purpose with 256 3 elements. Two projects have been proposed from volume imaging system. The proposed architecture overcomes Keio University in Japan. One is a prototype system SCOPE cunent image resolution and frame-rate limitations through the (solid and colored object projection environment) [6] and the use of custom high-speed processors. The gradient calculator other system-which is very similar to GODPA-is the 3Dp4 evaluates three arithmetic operations: a square and add architecture [7}. 3Dp4 bas been simulated in software, and a operation, square-root, and three division operations. Input- 256 3 hardware implementation is estimated to have a output delay time is 30 os with an accuracy of ±O.78%. The throughput of up to 10 frames per second. A commercial image algorithms and implementation in silicon are described in system, the Insight system of Phoenix Data Systems [8], detail. combines hardware and software. It can display complex objects at a "near real-time" rate of about one frame per second. 1. Introduction Kaufman and Bakalash [9] developed the CUBE architecture and constructed a prototype 16 3 system. All of these systems A Volume Imaging System (VIS) is a powerful tool in real- exhibit either low frame rates or low resolution. time manipulation of complex 3D data sets. Applications that Our objective is to design a system which overcomes these require such service are increasing. The data can be of medical. limitations with a high performance architecture which will geophysical, aerospace or other origins. Medical treatment render and display high resolution volume images from 256 3 planning sytems using high speed graphics workstations can and 512 3 data sets in real-time, that is at least 30 frames per provide powerful clinical tools for diagnostic evaluation of second. radiation therapy, for control in computer assisted surgery, for The objective of this paper is to present algorithms that laser surgery simulations, dental diagnostics or for facial yield an efficient, fast architecture for computing the reconstructions [l,2J. Geophysical imaging systems use normalized gradient magnitude and the normalized gradient rendering and displaying capabilities to visualize oil and gas direction. deposits. Imaging systems are used in the airline industry to localize possible leaks and defects in metal structures. Such 2. Gradient Calculation facilities would achieve maximum information extraction through high resolution reconstructions of examined This paper describes a gradient calculator which forms a structures. An interactive user interface requires real-time image subunit of the imaging system. The volume data received from update and continuous feedback to operator actions. This computed tomographs or magnetic resonance scans must be necessitates updates at the rate of 30 frames per second. The shaded and composited to produce a 3D image of the observed majority of existing 3D imaging systems employ general structure. The shading processor performs several arithmetic purpose workstations to manipulate complex data sets and logic operations. The flISt and most complex operation is compiled from computed tomography, magnetic resonance the calculation of a normalized gradient direction and images, or other scanned images. The processed data are three- normalized gradient magnitude. The gradient parameters defme dimensional density volumes of 256 3 or 512 3 generated as a the surface information of a single 3D volumetric element stack of slices. This amount of data creates major problems for called a voxeL software-oriented 3D reconstruction techniques. Software based In a volumetric data set, each voxel has 6 face-connected systems are too slow to provide real-time user feedback. Development of reconstruction methods using high speed neighbours. 12 edge-connected neighbours, and 8 corner- hardware has been actively pursued [3-9]. Traditional computer connected neighbours. Gradients can be calculated using any graphics software approaches, even those using hardware subset of these neighbours. The simplest case uses the face- assist. cannot convey the detailed information contained in connected neighbours. For example. the gradient of point (i, j. k) : . volume data sets in real-time. A voxel-based architecture GODPA (generalized object display process architecture), a in the X direction, is given by derivative of the Voxel Processor [3.4] has been implemented as a prototype system of 64 3 voxels. with a 16-frames-per- 80