Yongshun Xiao

A few-view reweighted sparsity hunting (FRESH) method for CT image reconstruction

In recent years, the total variation (TV) minimization method has been widely used for compressed sensing (CS) based CT image reconstruction. In this paper, we propose a few-view reweighted sparsity hunting (FRESH) method for CT image reconstruction, and demonstrate the superior performance of this method. Specifically, the key of the purposed method is that a reweighted total variation (RwTV) measure is used to characterize image sparsity in the cost function, outperforming the conventional TV counterpart. To solve the RwTV minimization problem efficiently, the Split-Bregman method and other state-of-the-art L1 optimization methods are compared. Inspired by the fast iterative shrinkage/thresholding algorithm (FISTA), a predication step is incorporated for fast computation in the Split-Bregman framework. Extensive numerical experiments have shown that our FRESH approach performs significantly better than competing algorithms in terms of image quality and convergence speed for few-view CT. High-quality images were reconstructed by our FRESH method after 250 iterations using only 15 few-view projections of the Forbild head phantom while other competitors needed more than 800 iterations. Remarkable improvements in details in the experimental evaluation using actual sheep thorax data further indicate the potential real-world application of the FRESH method.

Metal artifact reduction in CT images by sinogram TV inpainting

In this paper, a total variation (TV) inpainting method is proposed for metal artifact reduction in medical computed tomography (CT). Digital inpainting is an image processing method to fill in the lost image information in a consistent way. In our work, projection data with metal projection region (MPR) are treated as a damaged image, and TV inpainting is applied to “inpaint” the information missing region. Compared to conventional interpolation methods, the advantage of our algorithm lies in dealing with complicate cases such as an image with multiple metal objects. In numerical experiments, both TV inpainting and linear interpolation method are performed on noise-free and additive noisy projection of a modified Shepp-Logan phantom. Results show that the algorithm proposed fills metal projection gaps more smoothly and accurately than linear interpolation, hence produces images of superior quality after reconstruction. Relevant practical issues including the limitation of the algorithm and possible improvements for future work are discussed.

Metal artifact reduction in dual energy CT by sinogram segmentation based on active contour model and TV inpainting

In dual energy computerized tomography (DECT) which is widely used in industrial areas and security inspection, metal artifact reduction (MAR) is a troublesome problem. Pronounced streaks appear in the atomic number reconstruction and the value appears to be highly inaccurate when metal objects are present. In this article, a practical MAR method for DECT is proposed. Firstly, sinogram segmentation based on active contour model is implemented to obtain the metal projection region (MPR). Then, TV inpainting for sinogram was applied before reconstruction. Experiments demonstrate that, with our MAR method, the accuracy and image quality of the atomic number can be greatly improved.

X-ray spectrum estimation from transmission measurements using the expectation maximization method

X-ray spectrum estimation remains an active research area for its importance in medical imaging applications such as dose calculation and beam-hardening correction. In this paper, spectrum estimation is posed as an image reconstruction problem and the classical expectation maximization method (EM) for emission computed tomography is used to reconstruct the spectrum from transmission data. And we propose to compute the initial value for the EM iteration using Monte Carlo calculations. Numerical simulations were carried out to test the accuracy and robustness of the method. Experiments were conducted on an experimental CT system to evaluate the practicability of the algorithm. Both the simulations and experiments show that the proposed method can always produce a spectrum which fits the transmission data accurately and exhibits a rational shape.

Phase-contrast tomosynthetic experiment on biological samples with synchrotron radiation

Tomosynthesis is one of three-dimensional imaging techniques that can remove the effect of overlapping phenomena in radiography, except for computed tomography (CT). In general, CT needs at least hundreds of projections to reconstruct every cross-sectional slice of the samples accurately, while tomosynthesis just requires dozens of projections to reconstruct a series of tomosynthetic slices approximately, Conventional tomosynthesis based on attenuation contrast shows poor results when imaging weakly-absorption objects such as biological samples. In this paper, we present a new type of tomosynthesis, named phase contrast tomosynthesis, combining X-ray phase-contrast imaging mechanism and tomosynthesis, which can obtain higher resolution and image quality for the biological samples with fewer radiation doses and avoid overlapping phenomena. A phase contrast tomosynthetic experiment on a guinea pig cochlea at the Shanghai Synchrotron Radiation Facility (SSRF) was done to evaluate the performances of various reconstruction algorithms integrated with the in-line phase retrieval method.

Accelerated CT Reconstruction Using GPU SIMD Parallel Computing with Bilinear Warping Method

For high resolution CT image reconstruction, algebraic reconstruction methods are widely used to get relatively good image quality. However, the slow speed has prohibited its routine use in clinical and industrial applications. Many GPU-based accelerated algebraic reconstruction methods have been presented for three-dimensional (3D) reconstruction in cone beam CT. But for the high resolution two-dimensional (2D) reconstruction in linear detector fan beam CT, such as industrial CT (ICT), the projection and back-projection are not convenient accelerated by texture hardware and 3D geometric transformation. In this paper, we focus on using the GPU SIMD parallel computing to accelerate the 2D algebraic reconstruction in linear detector fan beam CT. We will present the Bilinear Warping method for implementing the resample texture coordinates calculation, and give the GPU-based simultaneous algebraic reconstruction algorithm with OpenGLs evaluator of Bilinear Bezier surface. The implementation algorithm on HP workstation indicates a speedup of about 15, and the speedup increases with the reconstruction image size, that means this method will be suitable for high resolution CT image reconstruction in practice.

Improve spatial resolution by Modeling Finite Focal Spot (MFFS) for industrial CT reconstruction

The finite focal spot is one of the major limitations of the high spatial resolution CT, especially to the high-energy industrial CT system with a macro-focus x-ray source. In this paper, we propose an efficient reconstruction framework through finite focal spot size based projection modeling to improve the spatial resolution of current industrial CT system, and demonstrate the superior performance of this method. First of all, the blurred projection produced by a finite size source is modeled as the integral ideal projection of a given point source over the finite focal spot support. Under the model discretization, the approximate linear equivalence relation between the actual finite focus model and the ideal point source model is established. Then a projection recovery method with this relationship is presented to recover the projection of the finer focal spot from the blurred projection. Finally, a high-spatial resolution image can be reconstructed from the recovered projections using the standard Filtered Back-Projection (FBP) algorithm. Furthermore the noise in the reconstructed image with different model parameters is studied and a difference image based fusion method is presented for the further suppression of the noise caused by the projection analysis processing. Both numerical simulations and real experiments have shown that the proposed reconstruction framework with the outstanding performance and efficiency characteristics can significantly enhance the spatial resolutions of current high-energy industrial CT systems.

An interaction based CT reconstruction algorithm for blocked projection data in a dynamic ICT system

In order to detect deformations of parts during the operating test, a novel dynamic industry computed tomography (ICT) system taking advantage of the rotation of specimens itself was purposed by us. However the stationary parts such as the shell around the turbine tips, which are hardly removed due to some industrial reasons, contaminate the projection data, so the blocks are not easily corrected from the projections as what we did in the traditional detector correction procedure. In this work, an interaction based CT reconstruction algorithm is purposed to deal the problem. First of all, we directly reconstruct the image with the contaminated projection data and an interactive match between the reconstructed image and the prior image is performed according to some obvious features. Then a forward-projection of the matched prior image with the practical geometric parameters is made. The block components in the projection data are estimated by calculating the average difference between the forward projections and the real projections of certain detectors. Finally, a new image can be reconstructed using the corrected data. Furthermore, the efficiency of the purposed algorithm is proved by both numerical simulation and practical experiments.

A reweighted total variation minimization method for few view CT reconstruction in the instant CT

In recent years, total variation (TV) minimization method has been extensively studied as one famous way of compressed sensing (CS) based CT reconstruction algorithms. Its great success makes it possible to reduce the X-ray dose because it needs much less data comparing to conventional reconstruction method. In this work, a reweighted total variation (RwTV) instead of TV is adopted as a better proxy of L0 minimization regularization. To solve the RwTV minimization constrain reconstruction problem, we treat the raw data fidelity and the sparseness constraint separately in an alternating manner as it is often used in the TV-based reconstruction problems. The key of our method is the choice of the RwTVs weighting parameters which influence the balance between data fidelity and RwTV minimization during the convergence process. Moreover, the RwTV stopping criteria is introduced based on the SNR of reconstructed image to guarantee an appropriate iteration number for the RwTV minimization process. Furthermore the FISTA method is incorporated to achieve a faster convergence rate. Finally numerical experiments show the advantage in image quality of our approach compared to the TV minimization method while the projection data of only 10 views are used.

SNM detection based on X-ray scattering

We present in this paper the study of special nuclear materials (SNM) detection method based on Pair production and Compton scattering Atomic Number Identification (PCANI) technique. PCANI, a technique that uses the information of annihilation and Compton scattered photons from target is applied at first to achieve the atomic number(Z) identification. Because of the Z2 and Z1 relationship of pair production cross section and Compton scattering cross section with Z, the ratio of 511keV γ number versus scattered γ number is proportional to Z, which gives us a chance to know what the atom is. PCANI is sensitive to high Z materials so it is not difficult to separate heavy metal from common goods that are mixed together in cargoes. The results of Monte Carlo calculations show that PCANI method can identify elements effectively, especially for high Z materials such as SNM. An experimental facility based on 7 MeV LINAC has been setup for feasibility study. Annihilation and backscattered photons have been observed with LaBr3(Ce) detector. Preliminary results show that high Z materials can be discriminated successfully.