Search

2014 IEEE International Conference on Acoustic, Speech and Signal Processing (ICASSP) METAL ARTIFACT REDUCTION FOR CT-BASED LUGGAGE SCREENING Seemeen Karimi, Pamela Cosman Harry Martz University of California, San Diego Dept. of Electrical and Computer Engineering 9500 Gilman Dr., La Jolla, CA 92093-0407 Lawrence Livermore National Laboratories Center for Nondestructive Testing 7000 East Ave., Livermore, CA 94550 ABSTRACT In aviation security, checked luggage is screened by com- puted tomography (CT) scanning, followed by automatic tar- get recognition from the CT images. Metal objects in the bags cause image artifacts that degrade object representation, leading to increased false alarms. We develop a new method, which isolates and reduces artifacts in an intermediate image, based on a numerical optimization that de-emphasizes metal and has a novel constraint for beam hardening and scatter. Re- sults on test bags showed excellent artifact reduction, even for multiple metal objects. Index Terms— metal artifact reduction, computed to- mography, luggage screening, constrained optimization 1. INTRODUCTION In aviation security, checked luggage is scanned by explo- sives detection systems (EDS). Many EDS are based on x-ray computed tomography (CT). Automatic target recognition al- gorithms in these systems analyze the CT images for threats. Metal objects present in the luggage create image artifacts ap- pearing as shadows or streaks. These artifacts misrepresent the surrounding objects, and may lead to apparent splitting of single objects, or the merging of separate objects. Reducing the metal artifacts will likely lead to lower false alarms [1]. Metal artifacts are caused by beam hardening, photon scatter, partial volume effects, photon starvation, and data sampling errors [2, 3]. Beam hardening and scatter cause low-frequency artifacts [2], which are more difficult to re- move, while the other sources result in narrow streaks. Algorithms for metal artifact reduction (MAR) have been developed in medical CT imaging since the 1980s [4]. De- spite the advances, there are no widely accepted solutions, and MAR continues to be a challenging research problem. There are three main approaches - sinogram replacement [4– 15], energy decomposition with multiple scanning spectra, e.g., [16], and iterative reconstruction (IR) [17–20]. All these methods operate in Radon space (also called projections or sinograms). Sinogram replacement has been the most explored be- cause of its low complexity. In methods based on sinogram 978-1-4799-2893-4/14/$31.00 ©2014 IEEE replacement, a filtered-backprojection (FBP) image is recon- structed from scanner projections, and the metal objects are identified by image segmentation techniques [8,21]. The pro- jection data corresponding to rays that pass through metal (called traces) are identified by calculation, by forward pro- jection (reprojection) of the metals or even by segmentation in the sinogram. Metal trace data in the sinogram are replaced with an estimate of underlying data, and the corrected sino- gram is reconstructed by FBP. It is difficult to estimate the un- derlying data accurately. Interpolation across the metal traces removes edges from high-contrast structures and renders the projections inconsistent, leading to secondary artifacts. In re- cent years in medical MAR, image segmentation has been used to identify high-contrast structures, to develop an inter- mediate image that is called a prior-image [8, 12, 14, 22]. The prior-image is reprojected and thus used to guide data replace- ment in the scanner sinogram. In luggage, image segmenta- tion cannot separate artifacts from data, because assumptions cannot be made regarding the contents of the images, and be- cause more metal and more artifact interference are present in luggage images. IR algorithms, such as those based on expectation maxi- mization [23] or algebraic reconstruction technique [24], are used in x-ray CT to reconstruct data with poor signal-to-noise ratio, and where there is missing data [25]. A recent approach is to use IR or numerical optimization for MAR [13, 26, 27] in medical imaging. These algorithms discard all metal trace data. IR and numerical reconstructions usually result in a loss of texture or resolution, so in [13], the optimum solution is used as a prior-image, which is then reprojected, and metal traces from the original sinogram are replaced with the repro- jected traces. In luggage screening, a third or even half the sinogram may contain metal. If all these data are discarded, the recon- structions are poor, as we will demonstrate. Our approach is also hybrid in that it reconstructs a prior-image by optimiza- tion, followed by sinogram replacement and FBP. However, we retain the metal projection data with reduced weights, and add a new constraint. Our target prior-image is artifact-free and sparse. The final image has the texture and resolution of FBP reconstruction.

It is often useful to simulate radiographic images in order to optimize imaging trade-offs and to test tomographic techniques. HADES is a code that simulates radiography using ray tracing techniques. Although originally developed to simulate X-ray transmission radiography, HADES has grown to simulate neutron radiography over a wide range of energy, proton radiography in the 1 MeV to 100 GeV range, and recently phase contrast radiography using X-rays in the keV energy range. HADES can simulate parallel-ray or cone-beam radiography through a variety of mesh types, as well as through collections of geometric objects. HADES was originally developed for nondestructive evaluation (NDE) applications, but could be a useful tool for simulation of portal imaging, proton therapy imaging, and synchrotron studies of tissue. In this paper we describe HADES' current capabilities and discuss plans for a major revision of the code.

High Energy Density Physics (HEDP) Experiments play an important role in corroborating the improved physics codes that underlie LLNL's Stockpile Stewardship mission. Conducting these experiments, whether on the National Ignition Facility (NIF) or another national facility such as Omega, will require not only improvement in the diagnostics for measuring the experiment, but also detailed knowledge of the as-built target components and assemblies themselves. To assist in this effort, a defined set of well-known reference standards designed to represent a range of HEDP targets have been built and are being used to quantify the performance of different characterization techniques. Without the critical step of using reference standards for qualifying characterization tools there can be no verification of either commercial or internallydeveloped characterization techniques and thus an uncertainty in the as-built-model for the initial condition to the physics codes.

Our work focuses on the development of finite element models (FEMs) that describe the biomechanics of human joints. Finite element modeling is becoming a standard tool in industrial applications. In highly complex problems such as those found in biomechanics research, however, the full potential of FEMs is just beginning to be explored, due to the absence of precise, high resolution medical data and the difficulties encountered in converting these enormous datasets into a form that, is usable in FEMs. With increasing computing speed and memory available, it is now feasible to address these challenges. We address the first by acquiring data with a high resolution X-ray CT scanner and the latter by developing a semi-automated method for generating the volumetric meshes used in the FEM. Issues related to tomographic reconstruction, volume segmentation, the use of extracted surfaces to generate volumetric hexahedral meshes, and applications of the FEM are described.

With the advent of ever more powerful computing and finite element analysis (FEA) capabilities, the bone and joint geometry detail available from either commercial surface definitions or from medical CT scans is inadequate. For dynamic FEA modeling of joints, precise articular contours are necessary to get appropriate contact definition. In this project, a fresh cadaver extremity was suspended in paraffin in a lucite cylinder and then scanned with an industrial CT system to generate a high resolution data set for use in biomechanics modeling.

X- and gamma-ray imaging techniques in nondestructive evaluation (NDE) and assay (NDA) have been increasing use in an array of industrial, environmental, military, and medical applications. Much of this growth in recent years is attributed to the rapid development of computed tomography (CT) and the use of NDE throughout the life-cycle of a product. Two diverse examples of CT are discussed. (1) The computational approach to normal joint kinematics and prosthetic joint analysis offers an opportunity to evaluate and improve prosthetic human joint replacements before they are manufactured or surgically implanted. Computed tomography data from scanned joints are segmented, resulting in the identification of bone and other tissues of interest, with emphasis on the articular surfaces. (2) They are developing NDE and NDE techniques to analyze closed waste drums accurately and quantitatively. Active and passive computed tomography (A and PCT) is a comprehensive and accurate gamma-ray NDA method that can identify all detectable radioisotopes present in a container and measure their radioactivity.

Lawrence Livermore National Laboratory (LLNL) and Sandia National Laboratories, California (SNL) are collaborating on the development of new techniques to study damage evolution and growth in material specimens subjected to mechanical loading. These techniques include metallography, radiography, computed tomography (CT) and modeling. The material specimens being studied include cast magnesium and aluminum alloys, and forged stainless steel. We will concentrate on characterizing monotonically loaded magnesium alloy specimens using computed tomography. Several notched tensile specimens were uniaxially loaded to different percentages of the failure load. Specimens were initially characterized by radiography and computed tomography to determine the preloaded state. Subsequent CT scans were performed after the samples were loaded to different percentages of the load failure. The CT volumetric data are being used to measure void size, distribution and orientation in all three dimensions nondestructively to determine the effect of void growth on the mechanical behavior of the materials.

The next generation, multi-detector active and passive computed tomography (A&PCT) scanner will be optimized for speed and accuracy. At the Lawrence Livermore National Lab (LLNL) we have demonstrated the trade-offs between different A&PCT design parameters that affect the speed and quality of the assay results. These fundamental parameters govern the optimum system design. Although the multi-detector scanner design has priority put on speed to increase waste drum throughput, higher speed should not compromise assay accuracy. One way to increase the speed of the A&PCT technology is to use multiple detectors. This yields a linear speedup by a factor approximately equal to the number of detectors used without a compromise in system accuracy. There are many different design scenarios that can be developed using multiple detectors. Here we describe four different scenarios and discuss the trade-offs between them. Also, some considerations are given in this design description for the implementation of a multiple detector technology in a field- deployable mobile trailer system.

Objective is to develop new approaches to electronically capture digital radiography and computed tomography images at high x-ray energies to satisfy spatial and contrast requirements for inspection of high-density weapons components.

Gamma-ray-based computed tomography (CT) requires that two different measurements be made on a closed waste container. [MAR92 and ROB94] When the results from these two measurements are combined, it becomes possible to identify and quantify all detectable gamma-ray emitting radioisotopes within a container. All measurements are made in a tomographic manner, i.e., the container is moved sequentially through well- known and accurately reproducible translation, rotation, and elevation positions in order to obtain gamma-ray data that is reconstructed by computer into images that represent waste contents. [ROB94] The two measurements modes are called active (A) and passive (P) CT. In the ACT mode, a collimated gamma-ray source external to the waste container emits multiple, mono-energetic gamma rays that pass through the container and are detected on the opposite side. The attenuated gamma-rays transmitted are measured as a function of both energy and position of the container. Thus, container contents are `mapped` via the measured amount of attenuation suffered at each gamma-ray energy. In effect, a three dimensional (3D) image of gamma- ray attenuation versus waste content is obtained. In the PCT measurement mode, the external radioactive source is shuttered turned- off, and the waste container, is moved through similar positions usedmore » for the ACT measurements. However, this time the radiation detectors record any gamma-rays emitted by radioactive sources on the inside of the waste container. Thus, internal radioactive content is mapped or 3D-imaged in the same tomographic manner as the attenuating matrix materials were in the ACT measurement mode.« less

The development of advanced materials and their improvement is strictly connected to the understanding of their properties and behavior as a function of both their macro- and micro-structures. The application of X-ray computed tomography (CT) to these materials allows for a better understanding of the materials properties and behavior on either macro- or micro-structure scales. We studied five Carbon-Polymer composite samples, static and fatigue tensile-tensile loading tested, characterized by different matrix composition (Epoxy and Peek), different fibers orientation and layer sequences. We considered both unnotched and notched (by a circular hole) samples too. We scanned these material with the Lawrence Livermore National Laboratory PCAT area array system, in order to investigate the feasibility of CT to study the defects and other features for such materials.

This paper describes the progress of a multi-modality (three detector type) system for x-rays and gamma-rays developed for the waste inspection tomography (WIT) program. WIT provides mobile semi-trailer mounted nondestructive examination (NDE) and assay (NDA) for nuclear waste drum characterization. WIT uses various computed tomography (CT) methods for both NDE and NDA of nuclear waste drums. Without opening waste drums, WIT inspects and characterizes radioactive waste, including low level (LLW), transuranic (TRU), and mixed waste. With externally transmitted x-ray NDE techniques, WIT has the ability to identify high density waste materials like heavy metals, define drum contents in 2D and 3D space, quantify free liquid volumes through density and x-ray attenuation coefficient discrimination, and measure drum wall thickness. With waste emitting gamma-ray NDA techniques, WIT can locate gamma emitting radioactive sources in 2D and 3D space, identify gamma emitting isotopic species, identify gamma emitting isotopic species, identify the extermal activity approximations, and provide the data needed for waste classification as LLW or TRU.

The technology being proposed by LLNL is an Active and Passive Computed Tomography (A P CT) Drum Scanner for contact-handled (CH) wastes. It combines the advantages offered by two well-developed nondestructive assay technologies: gamma-ray spectrometry and computed tomography (CT). Coupled together, these two technologies offer to nondestructively and quantitatively characterize mixed- wastes forms. Gamma-ray spectroscopy uses one or more external radiation detectors to passively and nondestructively measure the energy spectrum emitted from a closed container. From the resulting spectrum one can identify most radioactivities detected, be they transuranic isotopes, mixed-fission products, activation products or environmental radioactivities. Spectral libraries exist at LLNL for all four. Active (A) or transmission CT is a well-developed, nondestructive medical and industrial technique that uses an external-radiation beam to map regions of varying attenuation within a container. Passive (P) or emission CT is a technique mainly developed for medical application, e.g., single-photon emission CT. Nondestructive industrial uses of PCT are under development and just coming into use. This report discuses work on the A P CT Drum Scanner at LLNL.

This document establishes the software design, nomenclature, and conventions for industrial Computed Tomography (CT) used in the Nondestructive Evaluation Section at Lawrence Livermore National Laboratory. It is mainly a users guide to the technical use of the CT computer codes, but also presents a proposed standard for describing CT experiments and reconstructions. Each part of this document specifies different aspects of the CT software organization. A set of tables at the end describes the CT parameters of interest in our project. 4 refs., 6 figs., 1 tab.

X-ray computed axial tomography (CAT or CT) is an advanced imaging technique which provides three-dimensional nondestructive inspection and characterization of materials, components and assemblies. The Lawrence Livermore National Laboratory (LLNL) and the Mason Hanger-Silas Mason Co., Pantex Plant are cooperating to examine the use of CT Technology to inspect high-explosive (HE) pressings. The goals of this joint project are to study the HE pressing process using computed tomography scanners, to provide a foundation for the further development of industrial CT scanners, and to work with the private sector on the design and implementation of an automated CT inspection scanner to be used in the manufacturing process of high explosives at the Pantex Plant. The early detection of defective HE pressings will reduce the cost and improve the safety of the high-explosives production cycle. At present, all pressings are bulk density tested. In addition, samples are randomly selected from each pressing lot for destructive core sampling to obtain an estimate of internal density gradients. CT is a viable technology for a high-explosives inspection system. CT measures density variations, detects foreign inclusions and voids with a probability of detection as good as or better than present inspection techniques. In addition, CT canmore » be fully automated to allow for a 100% sample inspection. LLNL has the advantage of several different in-house built CT scanners to bring to bear on the inspection of high explosives. A preliminary study was performed to evaluate the LLNL CT scanners and prove the characterization principles associated with CT on small PBX9502 high-explosive samples. This paper summarizes the results obtained, and introduces the capabilities of LLNL's Nondestructive Evaluation Section CT scanners. 36 refs., 50 figs.« less

Radioscopy and transmission x-ray computerized tomography (TCT) were used to analyze as simulated waste canister. This study was performed to demonstrate the qualitative and quantitative possibilities obtained by TCT scanners over radioscopy and how TCT can be applied toward the solution of materials and waste management problems. This report provides a brief summary of the scanner (CATALYST) used, a description of the waste canister and its contents, and an analysis of three radiographic views and the two scans obtained at two different slice planes on the canister. The usefulness of TCT is demonstrated by how well it reveals the interior details and spatial positions of items within the simulated waste canister. This paper ends with a summary of this study and what further development is required for this technology to be better used in materials and waste management. 9 refs., 13 figs.

In x-ray and ion-beam computerized tomography, there are a number of reconstruction effects, manifested as artifacts, that can be attributed to the geometry of the experimental setup and of the object being scanned. In this work, we will examine four geometric effects that are common to first-and third-generation (parallel beam, 180 degree) computerized tomography (CT) scanners and suggest solutions for each problem. The geometric effects focused on in this paper are: X-pattern'' artifacts (believed to be caused by several errors), edge-generated ringing artifacts (due to improper choice of the reconstruction filter and cutoff frequency), circular-ring artifacts (caused by employing uncalibrated detectors), and tuning-fork artifacts (generated by an incorrectly specified center-of-rotation). Examples of four effects are presented. The X-pattern and edge-generated ringing artifacts are presented with actual experimental data introducing the artifact. given the source of the artifact, we present simulated data designed to replicate the artifact. Finally, we suggest ways to reduce or completely remove these artifacts. The circular-ring and tuning-fork artifacts are introduced with actual experimental data as well, while digital signal processing solutions are employed to remove the artifacts from the data. 15 refs., 12 figs.

There is a veritable renaissance occurring in x-ray imaging. X-ray imaging by radiography has been a highly developed technology in medicine and industry for many years. However, high resolution imaging has not generally been practical because sources have been relatively dim and diffuse, optical elements have been nonexistant for most applications, and detectors have been slow and of low resolution. Materials analysis needs have therefore gone unmet. Rapid progress is now taking place because we are able to exploit developments in microelectronics and related material fabrication techniques, and because of the availability of intense x-ray sources.