High-speed reconstruction for oblique-view CT

Abstract

To construct an oblique-view CT, oblique projection data are captured by angling the phantom bed relative to the plane of the X-ray beam. The projected image can be calculated from the absorption signals detected by a 2D array detector. In general, the scanning axis of the X-ray source and detector is approximately parallel to the long axis of the phantom and perpendicular to the plane of the X-ray source. In this setup, the image projection is a series of axial cross-sections through the phantom. This axial projection image depends on the thickness of the phantom. Generally, the image has a relatively high resolution for a thick phantom (>5 mm thick) and poor resolution when the phantom is thin. Moreover, mechanical and other practical considerations may limit the tilt and swivel angles to no more than approximately 30° between the horizontal phantom bed and the X-ray beam plane. In such cases, an oblique approach is a more effective method. In this paper, we propose high-speed reconstruction using a graphic processor unit (GPU)-based acceleration technique. This oblique CT system obtains projected images from the rotating phantom using a fixed-mounted X-ray source and detector and then reconstructs 3D voxels using the filtered back-projection (FBP) technique. In the experiments, we acquired 400 rotational projection images for the oblique-view phantoms and then reconstructed the aligned 3D view of the phantom. The results showed that the GPU technique achieved 7 giga update per second (GUPS) processing performance for 512 <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">3</sup> volumetric reconstructions. The oblique-view CT was able to obtain the 3D inner shape of the phantom with the oblique rotational axis.