Parallel CT image reconstruction based on GPUs

Abstract

H I G H L I G H T SWe developed GPU-based iterative algorithm to reconstruct images. Iterative algorithms are capable to reconstruct images from under sampled set of projections. The computer cost of the implementation of the developed algorithm is low. The efficiency of the algorithm increases for the large scale problems. Keywords:CT image reconstruction GPU-based algorithm CUDA C a b s t r a c t In X-ray computed tomography (CT) iterative methods are more suitable for the reconstruction of images with high contrast and precision in noisy conditions from a small number of projections. However, in practice, these methods are not widely used due to the high computational cost of their implementation. Nowadays technology provides the possibility to reduce effectively this drawback. It is the goal of this work to develop a fast GPU-based algorithm to reconstruct high quality images from under sampled and noisy projection data.

Evaluation of image filters for their integration with LSQR computerized tomography reconstruction method

“…That is why the algebraic methods of approximation started to be applied to reconstruct CT images, such as the LSQR (Least Squares QR) algorithm. They are capable of working with fewer views, as we have shown in our previous works [2][3][4][5]. But working with less projections also means more noise in the images.…”

“…N denotes the resolution of the image (128×128 pixels, 256×256 pixels, etc). This phase has been analyzed in previous works [2][3][4][5]9], using several different methods to solve the equations system. Nevertheless, in those works, the approach did not include an image filter combined with the iterative process of resolution, which we will study in this paper.…”

Computed tomography medical image reconstruction on affordable equipment by using Out-Of-Core techniques

“…A common approach to reducing the radiation dose is the use of iterative methods, which do not require a complete set of projections, nor are they restricted in terms of projection angles [9], [10], [11], [12], [13], [14]. These type of methods require fewer projections to reconstruct an image.…”

A rapid parallelization of cone-beam projection and back-projection operator based on texture fetching interpolation

“…The CUDA (Compute Unified Device Architecture) technology provides a software platform to easily design parallelized tasks with C-style code, and has become the most popular choices for developers. In the past years, various methods have been well exploited to accelerate the projection and back-projection in CT. Flores et.al parallelized the multiplication of the pre-stored system matrix by CUDA in 2-D iterative CT reconstruction [5]. In [6], Gao accelerated the projection and back-projection for iterative reconstructions by parallelizing the interpolation in Siddon's ray-driven algorithm.…”