Nuclear Instruments and Methods in Physics Research B 241 (2005) 810–815

www.elsevier.com/locate/nimb

Understanding X-ray cargo imaging

Gongyin Chen *

L-3 Communications, Security and Detection Systems, 10 Commerce Way, Woburn, MA 01801, USA

Available online 19 August 2005

Abstract

Cargo imaging is the field of imaging large objects, usually cargo containers, trains, trucks or boats. Transmission
imaging with photons, especially X-rays of up to 9 MV is the dominant current technique, providing compelling details
of the contents of objects this large. This paper discusses the physics aspects of a good X-ray cargo imaging system. The
basic performance requirements, such as penetration, contrast and resolution and the components of a cargo imaging
system are introduced. The imaging process is divided in this paper into three stages: forming information (probing the
object), recording information and presenting information (image display). Their impact on performance is analyzed.
Ó 2005 Elsevier B.V. All rights reserved.

PACS: 07.85.Fv; 29.40.Mc; 52.70.La

Keywords: X-ray; Imaging; Cargo

1. Introduction photons, especially X-rays up to 9 MV is the

dominant current technique. A few well respected
Cargo imaging is the industry of imaging large companies in this field are leading the advancement
objects, such as cargo containers, trains, trucks of this technology [1–6].
or boats. It is mainly aimed at finding illegal The fundamental physics of X-ray imaging is
goods, such as explosives, drugs, smuggled goods the exponential law for a narrow beam:
and even human trafficking. Cargo imaging gained I ¼ I 0
el
t ; ð1Þ
popularity over the past decade and is now
rapidly expanding. Although many neutron based where I and I0 are the remaining and original
methods are being proposed to provide elemental X-ray intensity and l and t are the attenuation
specific information, transmission imaging with coefficients and the thickness of material in the
beam path, respectively. Since the attenuation
coefficient of a material is photon energy depen-
*
Tel.: +1 781 970 1757; fax: +1 781 939 3996. dent and X-rays have a continuous spectrum of
E-mail address: [email protected] different energy photons, the total effect is the

0168-583X/$ - see front matter Ó 2005 Elsevier B.V. All rights reserved.
doi:10.1016/j.nimb.2005.07.136
 G. Chen / Nucl. Instr. and Meth. in Phys. Res. B 241 (2005) 810–815 811

integration of Eq. (1) over the applicable energy the photodiode is properly protected from direct
range. The attenuated photons are most likely beam, such detectors have proven to work for
scattered through the Compton process. many years without noticeable performance degra-
A cargo imaging system consists of an X-ray dation. Detector size is balanced by the desire to
source, usually a linear accelerator (Linac) or a display smaller features and the need of a larger
DC X-ray tube, collimators and beam filters, area to capture more photons. The detector usually
X-ray detectors, data acquisition electronics, con- needs to be thick enough to detect most of the
trol system, computers and radiation shielding. X-rays reaching it. In low energy systems each pixel
In order to spot items of interest from the images, usually has two detectors, one selectively records
the system must have adequate penetration, con- lower energy photons and the other selectively re-
trast sensitivity and resolution. In some cases cords higher energy photons. This information is
material discrimination is also desired. The imag- used to identify the material atomic number.
ing process first involves the transmission of Data acquisition electronics include detector
X-rays through the large object. Some X-rays must amplifiers, sampling and holding and analog to
pass through and an item of interest should cause digital conversion. They usually have 16-bit or
adequate signal contrast. Detectors and electronics higher resolution and the electronic noise needs
should properly register the remaining X-rays with to be small enough to support the overall system
adequate spatial resolution, usually a few millime- dynamic range, up to 14 bits in high end systems.
ters. The acquired data are processed and pre- Computers equipped with an imaging algo-
sented to an inspector. rithm process the data from detectors. This
may include detector normalization, source varia-
tion correction, averaging, bad data rejection,
2. Components of a cargo imaging system material calculation and image manipulation and
analysis.
The imaging system must have a proper X-ray Shielding varies from a few millimeters of lead
source. For energy of up to 500 kV, this is usually in a 160 kV system to a massive steel and concrete
an X-ray tube – a DC potential type electron accel- installation in a 9 MV system. It is required that
erator plus a tungsten target. For higher energy, a external radiation levels meet local regulations.
Linac is used to accelerate electrons that generate For 9 MV systems, steel and/or concrete is neces-
X-rays at a tungsten target. A higher energy sys- sary because there is also a need to shield the neu-
tem has harder and more intense X-rays. The en- tron by product. The dose delivered to cargo is
ergy is generally limited to about 9 MV due to limited with beam collimation and filtering. An
safety concerns. Higher energy sources will have interlock system is always included to prevent acci-
significant neutron production and photonuclear dental radiation exposure.
reactions can also be a concern. A cargo imaging
system usually uses a linear array of detectors.
Since X-rays are generated in all directions, at least 3. Imaging requirements
one source collimator is needed to form a slice of
beam. A detector collimator may also be used to In order to detect an item under a background,
reduce scattered X-rays and therefore to improve the item must have adequate signal contrast so
system dynamic range. In many cases, especially that it is not covered up by the pixel noise. The
when there is no closed shielding, a beam filter noise source varies and the minimum signal
is desirable to remove these unwanted photons – contrast that is required at a given noise level is
the lower energy portion of the X-ray spectrum. human dependent. For an area object, the signal
Most cargo imaging systems use scintillator – contrast usually needs to be one-third of the stan-
photodiode detectors. It could be CsI or BGO for dard deviation of pixel noise. For line objects and
low energy and low dose rate systems, but CWO point objects, the contrast needs to be one and
is the usual choice in high energy systems. When four times pixel noise, respectively. Fig. 1 shows
812 G. Chen / Nucl. Instr. and Meth. in Phys. Res. B 241 (2005) 810–815

Fig. 1. Area, line and point items in a nosy background. Signal

contrast as a ratio of pixel noise is given. Fig. 2. Typical maximum steel penetration of cargo imaging
systems.

objects of different signal contrast with a noisy

background. is from X-ray quantum noise and artifacts. Low
Maximum penetration is the single most impor- energy systems with small detectors often have
tant performance parameter. In cargo imaging, 0.125 mm wire resolution and high energy systems
this is usually defined as the maximum steel thick- usually have around 1 mm wire resolution.
ness behind which a lead block can still be seen. Contrast sensitivity is defined as the ability to
Object thicker than that steel would be completely see a thin steel plate behind a thick steel plate, usu-
dark in the image. After penetrating that maxi- ally in the range of half of the maximum penetra-
mum thickness steel, the remaining signal must tion. For example, a typical 450 kV system has
not be washed out by pixel noise. Under such con- 100 mm maximum steel penetration and its con-
dition, the pixel noise sources are primarily elec- trast sensitivity is 1–2 mm behind 50 mm steel.
tronic noise, uneven scatter, and artifacts, such Contrast sensitivity measures the ability to unveil
as generated by inadequate normalization in either details in the real world. The signal contrast is
direction. X-ray quantum noise, or statistic noise, the attenuation by the thin plate and pixel noise
is negligible in such a situation. The maximum is predominantly X-ray quantum noise. Artifacts
penetration of cargo imaging systems ranges from and electronic noise are usually negligible in such
about 25 mm for 160 kV systems to more than condition. Similarly, wire contrast defines the abil-
400 mm for 9 MV systems. Fig. 2 shows the typical ity to detect a thin steel wire behind a thick steel
maximum penetration of different systems. plate. A typical 450 kV system can detect a 3 mm
Display resolution (pixel size) is determined by diameter steel wire behind 50 mm of steel.
detector size and imaging magnification. It defines Some systems offer material discrimination,
the ability to see small features under high signal usually by exploring the difference between photo-
contrast, low noise conditions. In cargo imaging, electric process and Compton process and are
display resolution is usually a few millimeters common for 160, 320 and 450 kV systems. As
and is tested with line pair grids made of 2 mm higher atomic number materials have higher
thick steel, with no background object (in air). photoelectric cross sections, the ratio of low energy
Wire resolution is the ability to detect a small wire photon attenuation to high energy photon atte-
in air. The signal contrast is the attenuation by the nuation tends to be larger. Materials are usually
wire, usually diluted because its shadow may not divided into three groups, namely organic, inor-
cover a whole detector element. The pixel noise ganic and metal. Fig. 3 shows a dual-energy
 G. Chen / Nucl. Instr. and Meth. in Phys. Res. B 241 (2005) 810–815 813

eter wire, which could be detected in a 160 kV

system, would look almost transparent in the high
energy system. The source intensity, or beam cur-
rent, determines the initial amount of photons
and therefore the remaining number of photons
for a given source and specific contents of the ob-
jects. A good system can reach a dynamic range
of fourteen half value layers (14 HVLs) or more.
Besides collimators, a beam filter can be used for
scatter rejection. The best filter material for 9 MV
is polyethylene or similar content. For 3 MeV
X-rays, the best filter material is very high atomic
number material, such as lead or uranium.

Fig. 3. Dual-energy material discrimination map of a typical 5. Recording information

450 kV system. Horizontal axis is the logarithm of high-energy
detector signal and the vertical axis is the ratio of logarithm of The next step is to detect the X-rays that pass
two detector signals. through as a consequence of interactions with
the contents. Detector size, and therefore display
material discrimination map of a typical 450 kV resolution, sets the upper limit of spatial frequency
system. More recently, people have used a dual that the imaging system can show. In order to see
high energy X-ray source to identify very high small features in aviation security, the detector can
atomic number substances since they have a large be as small as 1 mm. On the other hand, a smaller
pair production cross section [7]. The effectiveness detector means fewer photons per pixel and larger
of material discrimination is usually very limited in statistical noise. In low X-ray attenuation situa-
cargo imaging mainly due to dilution along the tion, the pixel noise, or standard deviation of pixel
beam path – the items of interest are often under signal, is slightly larger than the square root of
a very complex background of other material. the average number of photons per pixel. Both
maximum penetration and contrast sensitivity will
suffer with the reduced photons per pixel. Wire res-
4. Probing the object olution, or the ability to detector a single wire
smaller than the detector size, is not affected
The first step of cargo imaging is letting X-rays by detector size to first order. When detector size
pass through a large object. In order to discover increases, signal contrast caused by the wire is
information about something inside, there must diluted but pixel noise is reduced and the two
be sufficient remaining photons after the beam effects compensate each other. Low energy systems
travels through the object and any item that we usually have detectors of high efficiency. For high
want to see in the image must cause a noticeable energy systems, the detector thickness and there-
different in X-ray intensity. In addition, scatter fore detection efficiency is usually limited by cost
must be controlled so that it does not add a strong consideration. Typical CWO detector thickness
background to cover weak signals. in cargo imaging is 20 or 30 mm. Gas detectors
For a lightly loaded air cargo or small parcels, a are theoretically more radiation resistant but their
160 kV system will provide adequate penetration low detection efficiency is a major disadvantage
and excellent contrast. At the other extreme, a fully and are rejected by most in the industry.
loaded Arab boat (called a Dhow) needs a 9 MV Crosstalk among detectors tends to blur images
system. Because 9 MV X-rays are more penetrat- and needs to be controlled within a few percent.
ing, a small piece of explosive or a 0.125 mm diam- Scintillation detectors naturally achieve this level
814 G. Chen / Nucl. Instr. and Meth. in Phys. Res. B 241 (2005) 810–815

as long as detector units are optically isolated. No root of pixel data and to map them the 256 gray
X-ray separator between units or in-plane collima- levels. Square root display is an equal-noise dis-
tion is required. In tomography, where further play method. That is, whether the signal is at 250
reconstruction is needed, X-ray separators are or 25 on the gray scale, the X-ray quantum noise
essential. corresponds to the same number of gray levels.
Electronic noise, including that from photodi- Noise reduction is usually built into these non-
odes, amplifiers and other circuits, plays a major linear transfer processes. That is, the logarithm
role in determining the maximum penetration. It is or square root lookup table can be modified to
the primary contributor to pixel noise in very low suppress noise.
signal conditions, as the case for testing maximum The single most helpful process is AHE, or
penetration. Everything else being equal, doubling adaptive histogram equalization. After taking
electronic noise means twice as much remaining sig- some non-linear transfer, the image is divided into
nal is needed to detect an item, as Fig. 1 shows. The smaller sections. In each section, pixels are as-
loss in penetration would be one HVL, or about signed a new value so that the number of pixels
30 mm of steel for a 9 MV system. that fall into each gray level is equal. Bilinear
Each detector channel has different amplifica-
tion and must be normalized. For a pulsed X-ray
source, the intensity of each pulse of X-rays should
be compensated with reference detectors. The
remaining error of the normalization and pulse
variation compensation causes horizontal and ver-
tical artifacts in the image, respectively. Artifacts
can also be introduced by discontinuity in detector
layout (such as between detector modules) or other
structure. Artifacts contribute to the overall pixel
noise and should be reduced whenever possible.

6. Presenting information

The last major step is to present the image to an

operator. The image data usually have 16 bits, or
64K levels. Gray scale display uses only 8 bits, or
256 levels and it cannot present all image informa-
tion at a time. Pseudo-color uses 8 bits for each of
the three basic colors, or 16 M levels. In this situ-
ation, the effectiveness is limited by the ability of
human eyes to tell subtle color differences.
X-ray signal decreases exponentially. With lin-
ear display (normalized pixel data are linearly
mapped to the 256 gray levels), most part of the
image would appear dark and most object detail
is lost. A better way is to map the logarithm of
normalized pixel data to the 256 gray levels. With
Fig. 4. Image of a fuel truck at Logan Airport. Top: detector
logarithm display, the brightness is proportional
data are linear mapped to 256 gray levels; middle: logarithm of
to the object thickness. Compared with linear detector signal is mapped to 256 gray levels; bottom: detector
mapping, logarithm display does not show thin signal is mapped to 256 gray levels after adaptive histogram
items very well. A similar way is to take the square equalization.
 G. Chen / Nucl. Instr. and Meth. in Phys. Res. B 241 (2005) 810–815 815

interpolations smooth the transition between sec- penetration, resolution, contrast sensitivity, mate-
tions. AHE usually provide the best overall look rial discrimination and radiation safety are the
of the object but the true object thickness informa- most important features and they are addressed
tion is lost. Different areas could have very differ- in the whole imaging process, including X-ray
ent gray level mapping. Fig. 4 demonstrates the interactions, image recording and image presenta-
effect of linear, logarithm and AHE displays. tion. Depending on the object, parameters are
For best result at a specific region, imaging soft- balanced to achieve performance and safety
ware usually allows the operator to adjust display goals.
contrast. Due to the complexity of the object and
the usually high pixel noise, edge enhancement
technique has very limited effectiveness in cargo References
imaging. For the same reason, automatic image
analysis does not work very well either. [1] http://www.L-3com.com.
[2] http://www.heimannsystems.com.
[3] http://www.nuctech.com.
[4] http://www.rapiscan.com.
7. Summary
[5] http://www.saic.com.
[6] http://www.as-e.com.
Understanding the physics is essential for [7] S. Ogorodnikov, V. Petrunin, Phys. Rev. (Special topics) 5
designing a good cargo imaging system. Maximum (2002) 104701.