Junjun Deng

A Parallel Implementation of the Katsevich Algorithm for 3-D CT Image Reconstruction

Yu and Wang [1, 2] implemented the first theoretically exact spiral cone-beam reconstruction algorithm developed by Katsevich [3, 4]. This algorithm requires a high computational cost when the data amount becomes large. Here we study a parallel computing scheme for the Katsevich algorithm to facilitate the image reconstruction. Based on the proposed parallel algorithm, several numerical tests are conducted on a high performance computing (HPC) cluster with thirty two 64-bit AMD-based Opteron processors. The standard phantom data [5] is used to establish the performance benchmarks. The results show that our parallel algorithm significantly reduces the reconstruction time, achieving high speedup and efficiency.

Automated Least-Squares Calibration of the Coregistration Parameters for a Micro PET-CT System

PET-CT coregistration parameters can be derived from PET and CT images of a four-point-source calibration phantom for a micro PET-CT scanner. An automated segmentation method has been developed, based on thresholding and application of constraints on the sizes of point sources in the images. After point sources are identified on PET and CT images, coregistration is performed using an analytic rigid-body registration algorithm which is based on singular value decomposition and minimization of the coregistration error. The coregistration parameters thus derived can then be applied to coregister other PET and CT images from the same system. Twenty PET-CT images of the calibration phantom at various locations and/or orientations were obtained on a Siemens Inveon® Multi-Modality scanner. We tested the use of from 1 to 10 data sets to derive the coregistration parameters, and found that the coregistration accuracy improves with increasing number of data sets until it stabilizes. Coregistration of PET-CT images with an accuracy of 0.33±0.11 mm has been achieved by this method on the Inveon Multi-Modality scanner.

Parallelism of iterative CT reconstruction based on local reconstruction algorithm

An iterative algorithm is suited to reconstruct CT images from noisy or truncated projection data. However, as a disadvantage, the algorithm requires significant computational time. Although a parallel technique can be used to reduce the computational time, a large amount of communication overhead becomes an obstacle to its performance (Li et al. in J. X-Ray Sci. Technol. 13:1–10, 2005). To overcome this problem, we proposed an innovative parallel method based on the local iterative CT reconstruction algorithm (Wang et al. in Scanning 18:582–588, 1996 and IEEE Trans. Med. Imaging 15(5):657–664, 1996). The object to be reconstructed is partitioned into a number of subregions and assigned to different processing elements (PEs). Within each PE, local iterative reconstruction is performed to recover the subregion. Several numerical experiments were conducted on a high performance computing cluster. And the FORBILD head phantom (Lauritsch and Bruder http://www.imp.uni-erlangen.de/phantoms/head/head.html) was used as benchmark to measure the parallel performance. The experimental results showed that the proposed parallel algorithm significantly reduces the reconstruction time, hence achieving a high speedup and efficiency.

Analysis of Performance Evaluation of Parallel Katsevich Algorithm for 3-D CT Image Reconstruction

The first theoretically exact spiral cone-beam CT reconstruction algorithm developed was by Katsevich. Recently, Yu et al. implemented the algorithm numerically. Although the method is very promising, the computation is very intensive. It requires huge amount of computer time. Recently, people began to parallelize the algorithm for achieving high performance computation. This paper presents an analysis of data decomposition and data communication in the parallel Katsevich algorithm and develops an analysis expression to evaluate the performance of the algorithm parallelism. The results based on the analytical model and numerical benchmarks compared in a fare agreement. The analytical model provides a great tool to evaluate high performance computing benchmarks in the parallel Katsevich algorithms

Deployment of One-Sided Communication Technique for Parallel Computing in Katsevich CT Image Reconstruction

This paper focuses on the implementation of parallel approach of Katsevich CT image reconstruction algorithm using one-sided communication technique. One-sided communication is a new feature provided by MPI-2 standard. It natively supports new network techniques such as InfiniBand and Myrinet. In this project, we implement three kinds of one-sided approaches for Katsevich algorithm. The results are compared the other P2P models on the performance metrics speedup, efficiency, and cost

Beam hardening correction using a conical water-equivalent phantom for preclinical micro-CT

Beam hardening artifacts are a common occurrence in x-ray CT images. X-ray sources typically produce polychromatic photons and their relative absorption is a strong function of their energy. Despite this fact, most reconstruction algorithms assume the attenuation coefficient of the subject being scanned is invariant with the energy of the incident photons, and the quality of the reconstructed images is reduced. A practical method of correction for those artifacts is to calibrate the acquired data to the expected projections of a known geometry. Generally this method requires scans of multiple phantoms varying in size to calculate the parameters of the calibration function. The selection of the phantom data for determining the parameters can also affect the performance of the method. This work proposes a beam hardening correction (BHC) scheme using a specially designed, conical phantom for preclinical micro-CT. The conical shape simplifies both the data acquisition and the calculation of the calibration function for different object sizes. A 3rd degree polynomial is chosen as the calibration function used to correct the CT projection data. Experiments conducted with the Siemens Inveon™ micro CT showed that the artifacts were greatly suppressed.

Speedup performance analysis of parallel Katsevich algorithm for 3D CT image reconstruction

The first exact spiral cone-beam CT reconstruction algorithm was developed by Katsevich (2002, 2004). Recently, Yu and Wang (2004a, 2004b) implemented the algorithm numerically. Although the method is very promising, the computation is very intensive. It requires huge amounts of computational time. Recently, people (Deng et al., 2006; Yang et al., 2006) began to parallelise the algorithm for achieving high performance computing (HPC). This paper presents a performance analysis of the parallel Katsevich algorithm (Deng et al., 2006) by developing an analytical expression to evaluate the performance of the algorithm parallelism. The results from the analytical model and numerical experiments are compared in a fair agreement. The analytical model provides a useful tool to estimate HPC benchmarks in the parallel Katsevich algorithm.

Parallelism of iterative CT algorithm based on local reconstruction

An iterative algorithm is suited to reconstruct CT images from noisy or truncated projection data. However, as a disadvantage, the algorithm requires significant computational time. Although a parallel technique can be used to reduce the computational time, a large amount of communication overhead becomes an obstacle to its performance. To overcome this problem, we proposed an innovative parallel method based on the local iterative CT reconstruction algorithm. The object to be reconstructed is partitioned into a number of sub-regions and assigned to different processing elements (PEs). Within each PE, local iterative reconstruction is performed to recover the sub-region. Several numerical experiments were conducted on a high performance computing cluster. And the FORBILD head phantom was used as benchmark to measure the parallel performance. The experimental results showed that the proposed parallel algorithm significantly reduces the reconstruction time, hence achieving a high speedup and efficiency.

Beam hardening correction using an attenuation coefficient decomposition approach

In X-ray CT system, the X-ray source generates x-rays with a broad range of spectrum. When the polychromatic photons travel through a subject, the lower energy photons are attenuated more effectively than the higher energy ones. This causes different attenuation coefficient (AC) of the material with respect to different X-ray photon energy. The reconstruction algorithms assuming invariant A C will introduce beam hardening artifacts in the reconstructed images. One effective method to address this is to model the X-ray attenuation process during the reconstruction. Generally an objective function is derived and an iterative method is used to minimize the objective function. This can be very time-consuming, and meanwhile it is relatively more difficult to compute such iterative algorithms in parallel. This work introduces an approach that uses an intermediate step to reduce the beam hardening artifacts. Due to the independence between the computation for each projection, it has the advantage of being able to compute in parallel.

3D cone-beam rebinning and reconstruction for animal PET transmission tomography

In PET scanners, one or multiple collimated point sources are used to acquire transmission measurements to generate attenuation maps for emission tomography [1]. The transmission acquisition is, intrinsically, 3D cone beam geometry. The acquired list-mode data are transformed into sinograms in the histogram process. Thereafter, transmission images are reconstructed from the sinograms, which are then re-projected to generate an attenuation map. Conventionally, a 2D rebinning method is used to transform the list-mode data into 2D parallel beam sinograms [2], and, accordingly, 2D reconstruction algorithms are employed to generate transmission images. Due to the inaccuracy of the 2D rebinning method, only limited oblique Lines of Response (LOR) can be used, causing limited axial coverage. If more oblique LORs were to be accepted in the rebinning process, artifacts would be introduced in the transmission images that may result in inaccurate attenuation correction factors. To address this issue, a 3D cone-beam rebinning process is proposed to faithfully transform the list-mode data, and the associated reconstruction algorithms for cone-beam geometry have been adopted to generate the transmission images. The experimental results showed the new method produced better images, especially in the axial direction.