Affiliations: Medical Image Processing Group, Key Laboratory of
Complex Systems and Intelligence Science, Institute of Automation, Chinese
Academy of Sciences, Graduate School of the Chinese Academy of Sciences, P. O.
Box 2728, Beijing, 100190, China
Abstract: Three dimension Computed Tomography (CT) reconstruction is
computationally demanding. To accelerate the speed of reconstruction,
Application Specific Integrated Circuit (ASIC) or Field Programmable Gate Array
(FPGA) has been used, but they are expensive, inflexible and not easy to
upgrade. The modern Graphics Processing Unit (GPU) with its programmable
features improves this situation and becomes one of the powerful and flexible
tools for 3D CT reconstruction. In this paper, we implement
Feldkamp-Davis-Kress (FDK) algorithm on commodity GPU using an acceleration
scheme. In the scheme, two techniques are developed and combined. One is cyclic
render-to-texture (CRTT) which saves the copy time, and the other is the
combination of z-axis symmetry and multiple render targets (MRTs), which
reduces the computational cost on the geometry mapping between slices to be
reconstructed and projection views. Our algorithm performs reconstruction of a
512^{3} volume from 360 views of the size 512
× 512 about 5.2s on a single NVIDIA GeForce 8800GTX
card.
