Accelerating simultaneous algebraic reconstruction technique with motion compensation using CUDA-enabled GPU

Int J Comput Assist Radiol Surg. 2011 Mar;6(2):187-99. doi: 10.1007/s11548-010-0499-3. Epub 2010 Jun 30.

Abstract

Purpose: To accelerate the simultaneous algebraic reconstruction technique (SART) with motion compensation for speedy and quality computed tomography reconstruction by exploiting CUDA-enabled GPU.

Methods: Two core techniques are proposed to fit SART into the CUDA architecture: (1) a ray-driven projection along with hardware trilinear interpolation, and (2) a voxel-driven back-projection that can avoid redundant computation by combining CUDA shared memory. We utilize the independence of each ray and voxel on both techniques to design CUDA kernel to represent a ray in the projection and a voxel in the back-projection respectively. Thus, significant parallelization and performance boost can be achieved. For motion compensation, we rectify each ray's direction during the projection and back-projection stages based on a known motion vector field.

Results: Extensive experiments demonstrate the proposed techniques can provide faster reconstruction without compromising image quality. The process rate is nearly 100 projections s (-1), and it is about 150 times faster than a CPU-based SART. The reconstructed image is compared against ground truth visually and quantitatively by peak signal-to-noise ratio (PSNR) and line profiles. We further evaluate the reconstruction quality using quantitative metrics such as signal-to-noise ratio (SNR) and mean-square-error (MSE). All these reveal that satisfactory results are achieved. The effects of major parameters such as ray sampling interval and relaxation parameter are also investigated by a series of experiments. A simulated dataset is used for testing the effectiveness of our motion compensation technique. The results demonstrate our reconstructed volume can eliminate undesirable artifacts like blurring.

Conclusion: Our proposed method has potential to realize instantaneous presentation of 3D CT volume to physicians once the projection data are acquired.

Publication types

  • Research Support, Non-U.S. Gov't

MeSH terms

  • Algorithms
  • Artifacts
  • Computer Graphics*
  • Humans
  • Image Enhancement / methods
  • Imaging, Three-Dimensional
  • Mathematics
  • Motion
  • Radiographic Image Interpretation, Computer-Assisted / methods*
  • Software
  • Time Factors
  • Tomography, X-Ray Computed*