This PDF file contains the front matter associated with SPIE Proceedings Volume 10632, including the Title Page, Copyright information, Table of Contents, and Conference Committee listing.

The Transportation Security Administration safeguards all United States air travel. To do so, they employ human inspectors to screen x-ray images of carry-on baggage for threats and other prohibited items, which can be challenging. On the other hand, recent research applying deep learning techniques to computer-aided security screening to assist operators has yielded encouraging results. Deep learning is a subfield of machine learning based on learning abstractions from data, as opposed to engineering features by hand. These techniques have proven to be quite effective in many domains, including computer vision, natural language processing, speech recognition, self-driving cars, and geographical mapping technology. In this paper, we present initial results of a collaboration between Smiths Detection and Duke University funded by the Transportation Security Administration. Using convolutional object detection algorithms trained on annotated x-ray images, we show real-time detection of prohibited items in carry-on luggage. Results of the work so far indicate that this approach can detect selected prohibited items with high accuracy and minimal impact on operational false alarm rates.

X-ray computed tomography is widely used in security applications. With growing interest in view-limited systems, which have increased throughput, there is a significant interest in constrained image reconstruction techniques that allows high fidelity reconstruction from limited data. These image reconstruction techniques are commonly characterized by their intense computational requirements making their deployment in real-time imaging applications challenging. Recent success of deep learning techniques in various signal and image processing applications has sparked an interest in using these techniques for image reconstruction problems. In this work, we explore the use of deep learning techniques for reconstruction of baggage CT data and compare these techniques to constrained reconstruction methods.

X-ray diffraction-based baggage screening provides the potential for the material sensitivity needed to realize high detection probabilities and low false alarm rates. However, the combination of noisy signals, variability in the XRD form factors based on slight material differences, and incomplete material libraries lead to decreased system performance. By using a machine learning classification approach to XRD-based explosives detection, we show that the probability of error can be reduced relative to traditional, correlation-based classifiers. This improved performance exists at a variety of noise levels and degrees of library completeness, and indicates a path toward increased XRD-based classifier robustness.

Conventional X-ray computed tomography (CT) only reconstructs the attenuation map. X-ray coherent scattering computed tomography (CSCT) probes the angular-dependent scattering profiles of the three-dimensional (3D) object, achieving a high structural specificity among materials with similar electron density. However, due to the low level of interaction of coherent scattering and the 3D scattering profiles of crystalline materials, the real-life anomaly detection using coherent scattering signature posed challenges in acquisition speed, algorithm efficiency, reconstruction quality, and detection accuracy. In this invited talk, we will discuss the efforts in accelerating the image acquisition speed while maintaining object-specific information from the Optical Imaging System Lab at the University of Central Florida. Specifically, the talk will be focus on maximize the system detection efficiency and improve the classification/reconstruction performance for low photon detection level to meet the need of practical detection.

The detection of prohibited items at airport checkpoints, especially energetic materials, by means of x-ray imaging technology, is one of the most important tasks in transportation security. Conventional checkpoint X-ray systems exploit the energy dependence of the material- specific attenuation coefficient to estimate an ‘effective’ atomic number (or Z_eff ) and, in some cases, the mass density (ρ) of a target material, which are then used to classify it. While this technology provides high quality imaging capabilities and satisfactory objects discrimination in many security applications, it also has known limitations. For example, differentiating objects with similar Z_eff and/or ρ, such as is often the case for many benign organic materials and explosives, can be a challenging task. X-ray Diffraction Tomography (XRDT), using a coded mask (down stream from the sample), provides structural information that can further enhance material discrimination from the unique chemical/molecular signatures. Here, we present experimental data obtained using our research prototype or ‘XRDT’ scanner, built with off-the shelf components. Using two different industrial solvents, one benign (H₂O or water) and one prohibited chemical precursor (2-butanone or methyl-ethyl-ketone (MEK)), we have evaluated the detection performance against material type, sample size, beam size, and investigated the effects of background. Within the scope of our study, we find that a satisfactory tomographic reconstruction and reliable bulk material identification can be achieved with the XRDT. These results will help guide the future development of coded aperture based screening technology at security checkpoint.

Concealed threat detection is a challenging task that requires a high degree of material specificity. X-ray diffraction tomography (XRDT) offers a solution to the problem, but may at times be too sensitive to the details and history of a given sample. One example of this effect arises in the dependence of the measured scatter signal to the orientation of the sample relative to the beam, which we refer to generally as texturing. To better understand texturing in real world materials and imaging scenarios, we develop two experimental systems for measuring scatter and create databases of the resulting scatter form factors over a range of energies and angles. We then use this data to develop a simulation tool to model XRDT systems in the presence of textured materials and analyze the results. While texturing introduces complications to accurate imaging of an object, we find that choices in the measurement strategy can mitigate these effects and improve performance.

Conventional computed tomography (CT) only reconstructs the attenuation map within a sample. X-ray coherent scattering computed tomography (CSCT), which probes the angular-dependent scattering profiles of a 3D object, achieves high-contrast and specificity among materials or tissues with similar attenuation cross-section. Due to the weak coherent scattering cross-section, CSCT using a pencil-beam either requires a brilliant source, such as synchrotron, or tens of hours in image acquisition using a traditional X-ray tube. Fan-beam CSCT using table-top source has been proposed to parallelize the acquisition of each projection, but the use of collimators on the detector plane limits the collection efficiency. Moreover, conventional CSCT systems cannot further parallelize the image acquisition among layers by switching to cone-beam illumination, because the scattering signal from different layers will overlap on the detector. Here we propose a fast, high-efficiency multiplexing scheme using structured cone-beam illumination to image a 3D sample. Our system improves the source utilization compared to pencil-beam CSCT, yet does not require detector collimator to localize and resolve the scattering profile of each point. We have reconstructed the coherent scattering profile within a volumetric object, and demonstrated the material classification of the 3D sample. Compared to previous systems, our method reduces the imaging time by one order of magnitude. We believe our multiplexed CSCT scheme could become the next generation X-ray coherent scattering tomography system.

Detecting the presence of hazardous materials in luggage is an important problem in aviation security. The current generation of inspection systems is based on X-ray computed tomography, followed by recognition systems to identify potential prohibited materials. As such, the image formation algorithms are designed independently of the recognition algorithms. In this paper, we present a new class of algorithms for processing the X-ray data by simultaneously forming images from the collected X-ray observations and identifying the underlying materials in the images. These algorithms exploit information about the possible materials in the image to modify the image reconstruction techniques, as well as material identification. We evaluate our joint algorithm on simulated phantoms using multi-spectral computed tomography, and compare our reconstruction and classification results with alternative state of the art approaches. Our experiments indicate that there are significant improvements in recognition performance possible through our joint approach.

An adaptive automatic threat recognition system (AATR) developed at the Lawrence Livermore National Laboratory (LLNL) is described for x-ray CT images of baggage. The AATR automatically adapts to the input object requirement specification (ORS), which can change or evolve over time. These specifications characterize materials of interest (MOIs), basic physical features of interest (FOIs) (such a mass and thickness) and performance goals (detection and false alarm probability) for objects of interest (OOIs). The need and technical requirements for an AATR were developed in collaboration with DHS’s Explosives Division and Northeastern University’s Awareness and Localization of Explosives-Related Threats (ALERT) Center, a DHS Center of Excellence (http://www.northeastern.edu/alert/). Independent of the input ORS, LLNL’s AATR always uses the same algorithm and codes to process CT images. The algorithm adapts in real-time to changes in the input ORS. LLNL’s AATR is thus suitable for dynamic scenarios in which the nature of the OOIs can change rapidly. The AATR uses a spatial consensus relaxation method to determine the most likely material composition for each CT image voxel. The resulting image of most likely material compositions is segmented. An OOI classification statistic (OOI score) is computed for each voxel and each extracted image volume. OOI recognition performance is reported using various metrics on a test set of ~180 plastic bins supplied by the ALERT Center of Excellence. A method is then proposed for automatic decision threshold estimation that can adapt to the detection performance goal, the most likely material composition, and the contents of the baggage.

Differentiating material anomalies requires a measurement system that can reliably inform the user/classifier of pertinent material characteristics. In past work, we have developed a simulation framework capable of making simulated x-ray transmission and scatter measurements of virtual baggage. Using this simulated data, we have demonstrated how an information-theoretic approach to x-ray system design and analysis provides insight into system performance. Moreover, we have shown how performance limits relate to architectural variations in source fluence, view number, spectral resolution, spatial resolution, etc. However, our previous investigations did not include material variability in the description of the materials which make up the virtual baggage. One would expect the material variability to dramatically affect the results of the information-theoretic metric, and thus we now include it in our analysis. Previously, material information was captured as energy-dependent mean attenuation values. Because of this, material differentiation can always become easier with an improvement in SNR. When there is no variation to obscure class differences, improvements in SNR will indefinitely improve performance. Therefore, we saw a monotonic increase of the metric with source fluence. However there is inherent variability in materials from chemical impurities, texturing, or macroscopic variation. When this variability is accounted for, we better understand system performance limits at higher SNR as well as better represent the distributions of material characteristics. We will report on the analysis of real world system geometries and the fundamental limits of performance limits after incorporating these material variability improvements.

The radiation dosage delivered to the sample is a constant challenge facing X-ray imaging systems. In conventional transmission-based computed tomography (CT), a beam penetrating through thick, high-attenuation region in the sample results in low signal on the detector, and therefore a higher power (e.g., tube current modulation) or longer integration time is often required to maintain signal quality. The issue of radiation dose becomes more sever in coherent scattering X-ray tomography, in which the scattering signal is typically orders of magnitude weaker than the transmitted beam. With X-ray photon-counting detectors, transmitted (or scattered) X-ray photons can be acquired at extremely low photon flux, which enables us to greatly reduce the imaging time and dose administrated to the sample. Instead of counting the average the number of photons within a fixed time interval, the arrival times of only a few photons detected in sequence contain sufficient information to estimate the attenuation (or scattering) property, which allows object reconstruction based on our measurement geometry and noise model. We will also discuss compressive or adaptive data acquisition schemes to implement material identification utilizing the energy sensitivity of X-ray photon-counting detector. Our method can be further parallelized with a photon-counting detector array to achieve fast, low-dose X-ray tomographic imaging based on either attenuation or scattering signals, which could find broad applications in medical diagnosis and security screening.

In our prior work, we had employed a fixed photo-absorption, coherent, and incoherent cross-section material model to derive a shot-noise limited description of the X-ray measurements in check-point or a checked baggage threat-detection systems. Using this measurement model, we developed an information-theoretic metric, which provides an upper-bound on the performance of a threat-detection system. However, the fixed cross-section material model does not incorporate material variability arising from inherent variations in its composition and density. In this work, we develop a multi-energy model of material variability based on composition and density variations and combine it with the shot-noise photon detection process to derive a new X-ray measurement model. We derive a computationally scalable analytic approximation of an information-theoretic metric, i.e. Cauchy-Schwarz mutual information, based on this material variability model to quantify the upper-bound on the performance of the threat-detection task. We demonstrate the effect of material variations on the performance bounds of X-ray transmission-based threat detection systems as a function of detector energy resolution and source fluence.

X-ray phase-contrast imaging methods exploit variations in an object’s 3D refractive index distribution to form projection or volumetric images of weakly absorbing objects. Such techniques can resolve subtle tissue structures by employing coherent imaging principles, but retain the ability of traditional (incoherent) X-ray methods to image deep into tissue. In this talk, we describe recent advancements in image formation methods for benchtop applications of X-ray phase-contrast imaging and tomography and present applications of pre-clinical in vivo imaging.

X-ray phase contrast imaging (XPCI) reveals structure and detail of low density materials with a sensitivity not accessible to conventional absorption based x-ray imaging or other non-destructive inspection techniques. The wide use of low density materials in defense and security applications has driven development outside the medical domain. In the laboratory environment (instantiations that do not employ a synchrotron), XPCI has moved beyond nascent demonstrations. Advances have been made in grating fabrication, source development, and specialized detectors. As the application space grows, new algorithms for acquisition, reconstruction, and corrections are being developed. I will review the state of the art in laboratory grating-based XPCI with emphasis on the growing interest in materials science applications. Hurdles remain for XPCI to move beyond laboratory demonstrations and become a widely used non-destructive inspection technique. The most common three-grating system has limitations defined by grating fabrication limits, which determine attainable energy levels, and relevant samples. The system geometry, signal levels, and speed of acquisition must be realistic for real world applications. This talk will provide a perspective on the global state of XPCI and development trends that seek to expand the operational space.

Conventionally, the contrast of X-ray images is due to the attenuation of intensity of x-ray beams after penetrating materials, which is proportional to the imaginary part of the complex refractive index. Subtle density variations within soft tissue yields poor contrast. One method to improve the contrast of x-ray images is to utilize phase information, which could provide a signature 1000 times larger than attenuation. However, phase imaging relies critically on the spatial coherence of the x-ray beam which traditionally requires synchrotron sources, small-spot, low power laboratory sources, or precisely aligned gratings and multiple exposures. An additional source of tissue-typing information, which is simply discarded in a conventional mammogram, is coherent scatter. Coherent scatter imaging relies on diffraction within the tissue and hence produces a signature that depends on the molecular structure, but as conventionally collected requires raster-scanning of the beam and multiple exposures. None of these methods is compatible with conventional screening mammography. We will discuss two methods to achieve phase imaging with large-spot sources practical for clinical use. The first uses polycapillary optics to focus x-rays from a large-spot source and achieve the necessary coherence for propagation-based phase imaging. The second uses structured illumination implemented with a coarse wire mesh to enhance phase signatures and relax the coherence requirement. We will present recent results from both methods, including computational algorithms for phase contrast, phase retrieval and resolution enhancement. We will also present a slot-scanning coherent scatter system which utilizes a slot to shape the beam and shielding placed at specific angles to capture specific coherent scatter signatures in a geometry that is compatible with slot-scan mammograpy.

Foams and encapsulants serve important roles in the protection of the components they surround. These low density materials may be used to provide shock protection, to protect against high voltage breakdown, or to minimize thermal fluctuations. Voids and gaps in the material, delaminations from a mating material, or non-uniformities in the encapsulating materials can lead to critical failures in the encapsulated component. Despite the important role these low density materials serve, traditional non-destructive inspection tools are limited in their ability to study this material set, especially in the presence of high density materials such as wires. The default approach has been destructive post-mortums where components are deconstructed after a failure and cause and effect are difficult to distinguish. X-ray phase contrast imaging has a longer history at synchrotrons, but this is not a realistic solution for non-destructive inspection. We have demonstrated grating-based x-ray phase contrast 3-D tomography in a laboratory environment with a conventional x-ray tube. Our large format grating fabrication capability enables imaging with large fields of view (10 cm2) at 28 keV for the successful non-destructive inspection of these low-density materials. We demonstrate that the complementary image modalities available with XPCI provide unique information and higher contrast for the inspection of defects in low density materials than conventional x-ray alone.