Parallel implementation of the filtered back projection algorithm for tomographic imaging
Computer Tomography (CT) is used in several applications — medicine, non-destructive testing/evaluation, astronomy and others to look inside objects and analyze internal structures. However, the problem in general is computationally very intensive. The large computational requirement has led to large times for CT reconstruction. This has hindered the use of CT in many applications where CT image data is required in real time. Parallel processing is one of the techniques available which can reduce the reconstruction time very significantly. In this project, we describe a parallel implementation of the filtered back projection algorithm for finding 2D cross sectional images. The results show that parallel processing is very viable for CT reconstruction and results in significant run time savings. An implementation on two different machines is described - the Intel Paragon and the Connection Machine. For comparison, the filtered back projection algorithm was first programmed as a serial algorithm using Matlab software on a Sun Sparc 10. It has been found that while the serial program in Matlab takes approximately 12 minutes for a reconstruction, the same data can be parallelized on the Intel Paragon in about 2.2 secs and on the CM5 in about 3 secs.