Nuclear Instruments and Methods in Physics Research B 318 (2014) 223–231

Contents lists available at ScienceDirect

Nuclear Instruments and Methods in Physics Research B

journal homepage: www.elsevier.com/locate/nimb

Concept of effective atomic number and effective mass density

in dual-energy X-ray computed tomography
Anne Bonnin a,b, Philippe Duvauchelle b,⇑, Valérie Kaftandjian b, Pascal Ponard c
a
ESRF, 6 Jules Horowitz, F-38073 Grenoble Cedex, France
b
LVA, Vibrations and Acoustic Laboratory, INSA-Lyon, Université de Lyon, F-69621 Villeurbanne Cedex, France
c
Thales Electron Devices SAS, 2 Rue Marcel Dassault, BP23 78141 Vélizy, Villacoublay Cedex, France

a r t i c l e i n f o a b s t r a c t

Article history: This paper focuses on dual-energy X-ray computed tomography and especially the decomposition of the
Received 29 May 2013 measured attenuation coefficient in a mass density and atomic number basis.
Received in revised form 5 September 2013 In particular, the concept of effective atomic number is discussed. Although the atomic number is well
Available online 5 November 2013
defined for chemical elements, the definition of an effective atomic number for any compound is not an
easy task. After reviewing different definitions available in literature, a definition related to the method of
Keywords: measurement and X-ray energy, is suggested. A new concept of effective mass density is then introduced
Effective atomic number
in order to characterize material from dual-energy computed tomography.
Effective mass density
Dual-energy
Finally, this new concept and definition are applied on a simulated case, focusing on explosives iden-
X-ray computed tomography tification in luggage.
Explosives identification Ó 2013 Elsevier B.V. All rights reserved.

1. Introduction
X-ray computed tomography (CT) is a well known 3D technique
which enables one to determine a local value of the linear attenu-
ation coefficient l of a material. This coefficient depends on the
X-ray energy and the characteristics of the material (its mass den-
sity q and its atomic number Z). Nevertheless, different materials
can have the same attenuation coefficient at a given energy. For this
reason, dual-energy techniques have been introduced. First applied
in the medical field, they have also been used in radioscopy mode
for industrial applications, amongst others in luggage inspection.
From the measurements acquired at two different energies, dif-
ferent choices have been proposed in the literature to differentiate
two or more materials. The first approach, described by Alvarez
et al. [1] and Macowski et al. [17], was to express l as a function
of mass density and atomic number. Thus using two energies, it
is possible to derive a (q, Z) couple from the two measured attenu-
ation coefficients. Other approaches showed that an unknown
material can be expressed as a linear combination of two compo-
nents forming a reference basis. These components can be, for
example, the two physical contributions resulting from photoelec-
tric absorption and Compton scattering or two reference materials
used during a calibration step (see [16]). For example, these two
materials can be bone and soft tissue [15], while in the industrial
⇑ Corresponding author. Tel.: +33 4 72 43 70 58.
E-mail addresses: annebonnin@free.fr (A. Bonnin), philippe.duvauchelle@
insa-lyon.fr (P. Duvauchelle).
field plexiglass and iron can be chosen for radioactive waste drums
inspection [23,22].
The discrimination of materials based on atomic numbers and
mass densities, as proposed originally by Alvarez, is mainly used
in luggage inspection and enables one to differentiate organic from
inorganic materials, or more precisely, light from heavy materials.1
A calibration process is also needed, from two to a set of different
materials [3]. In this case, materials are classified in categories (light
vs. heavy) but not really identified.
Our article is based on the decomposition of l in a (q, Z) basis,
whose principle is described in Section 2. A question arises then:
what is the meaning of the atomic number for a compound2 of
several chemical elements (which is the case of most common
materials)?
After reviewing several Zeff definitions in the third section, we
propose our own definition in Section 4, in which Zeff depends on
the energy. Therefore, in the context of dual-energy measurement,
we show in Section 5, that it is necessary to introduce also the
concept of effective mass density qeff. Hence any material is char-
acterized by a unique (q, Z) couple for the given couple of energies.
In Section 6, the influence of both energies is studied for water. In
the last part of this article, our method has been used on a phan-
tom in the context of explosives detection.
1
The frontier between light and heavy materials is considered to be situated at
q  2 g cm3 and Z  8.
2
In this publication, we will use the term ‘‘compound’’ to refer to any sample with
more than one chemical element such as chemical compounds, mixtures, alloys,
etc.. . .

0168-583X/$ - see front matter Ó 2013 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.nimb.2013.09.012
224 A. Bonnin et al. / Nuclear Instruments and Methods in Physics Research B 318 (2014) 223–231

2. Theoretical background In 1952, Hine [13] introduced an expression of Zeff as a function of

each specific interaction process but only for the photoelectric
P P
In CT, materials are differentiated thanks to their linear X-ray effect, Z 3:1
eff 3a ¼
3:1
i xi Z i , and for the pair creation, Z eff 3b ¼ i xi Z i .
attenuation coefficient l. Following Alvarez et al. [1], for a given After comparison of this study, giving substantially the same results
energy E, l is a function of the mass density q and atomic number for different compounds for these two partial processes, Murty [19]
Z of the element as follows: assumed that the effective atomic number can be approximated by
a unique relation, defined in Eq. (6), valid also for the Compton
lE ðq; ZÞ ¼ aE qZ mE þ bE q: ð1Þ
effect at intermediate energies:
In order to determine q and Z at each point of the sample, the P xi Z i
i Ai
use of two energies (a low energy LE and a high energy HE) is Z eff 4 ¼ P xi : ð6Þ
needed to solve the equation system: i Ai

lLE ¼ aLE qZ mLE þ bLE q ð2Þ
lHE ¼ aHE qZ mHE þ bHE q Tsai and Cho [27] reviewed Spiers equation (Eq. (3)), by revalu-
ing the power factor from 2.94 to 3.4 on the basis of attenuation
The dual-energy CT method is realized in two steps.
measurements carried out on biological tissues. This time, the
The first step consists of a calibration with a sample of known
expression is valid only for energies lower than 150 keV:
materials (q, Z), in order to obtain the coefficients aLE, mLE, bLE
and aHE, mHE, bHE (see Eq. (2)) and then the two calibration sur- hX i1=3:4
Z eff 5 ¼ aei Z 3:4
i ð7Þ
faces: lLE = f(q, Z) and lHE = g(q, Z). i

The second step consists in a measurement of l at two energies

for an unknown material. The (q, Z) couple of this material can then To characterize organic matter (i.e. light elements with Z < 10)
be deduced by solving Eq. (2). For an element, there is no ambigu- at energies greater than 20 keV, Schätzler [24], applied the relation
ity on the determination of Z. But for a compound, the concept of proposed by Hine for pair creation, where the atomic number Zi of
effective atomic number is widely discussed in the literature and each element i, is weighted by the mass percentage xi of this
we cannot find a unique definition well accepted by the research element:
community. In the following section, we review some existing
definitions. X
Z eff 6 ¼ i
xi Z i ð8Þ

3. Review of existing methods to define Zeff

Later, on a limited range of Z, Manninen, and Koikkalainen [18]
Different calculation methods can be found in the literature to suggested a relation introducing the percentage of atoms aat
i :
define the effective atomic number Zeff for a compound. These "P #1=2
expressions are presented below. Some of them have already been aat Z 3 n
Z eff 7 ¼ P iat i i
with aat
i ¼
Pi : ð9Þ
i ai Z i
collected by [7]. i ni
They can be classified in two categories. The first one is based
on calculations involving the atomic number Zi of the ith atom con- Puumalainen [21] then simplified the relation as:
stituting the compound and for some of them, their validity field X
at
depends on the X-ray energy. The second category gathers the
Z eff 8 ¼ i i
aZi : ð10Þ
methods based on mass attenuation coefficients.
This list of calculation methods is not thorough. This illustrates
3.1. Zeff obtained from calculation methods that for each specific need, a particular method is applied, accord-
ing to the range of energy, the type of compounds, the field of
Zeff can be considered in a general way as a function of atomic application.
numbers Zi weighted by the percentages of each element i in the
compound. For an element i, the atomic number Zi represents the 3.2. Zeff obtained from mass attenuation coefficients database
number of protons or electrons contained in the corresponding
atom. Guru Prasad et al. [12] proposed a way to find the effective
Considering absorption in biological materials such as organs, atomic number for a compound at a given energy: for a series of
Spiers in 1946 [26], established an expression of the effective alloys, assuming that the photoelectric absorption is predominant
atomic number according to the percentage of electrons aei : between 10 and 150 keV, and knowing the tabulated mass attenu-
hX i1=2:94 ni Z i ation coefficients of elements lq ðZ; EÞ (see [14], [6] and the National
Z eff 1 ¼ aei Z 2:94
i with aei ¼ P ð3Þ
i ni Z i
i Institute of Standards and Technology (NIST) reference data web-
site, XCOM [2]), they calculate a so-called ‘‘cross-section per atom
where ni is the number of atoms.
on average in the alloy’’ ra, first presented by Conner et al. [4] and
One year later, in the radiological field, Glasser [10] established
defined for a compound by:
the following expression directly based on the mass percentage xi:
"P #1=3 ½ðl=qÞtabulated compound
xi Z 4i ½ra compound ¼ P : ð11Þ
Z eff 2 ¼ Pi ð4Þ NA i xAii
i xi Z i

where Ai is the atomic mass, xi the mass percentage of each ele- From Eq. (11) and for the different elements, he obtains ra = f(E)
ment i expressed as a function of ni and At the total atomic mass (as lq depends on E) and then for a given energy, ra = f(Z). A linear
of the compound: interpolation of all these values provides a theoretical curve of ra
n i Ai depending on the atomic number Z for a given energy: ra = f(Zeff).
xi ¼ ð5Þ The effective atomic number of a compound, Zeff9, is obtained
At
graphically thanks to the intersection of this curve and the
 A. Bonnin et al. / Nuclear Instruments and Methods in Physics Research B 318 (2014) 223–231 225

estimated value ra calculated with the mass attenuation coeffi-
cient tabulated at the given energy.
For 25 alloys, Guru Prasad presents the comparison between
this technique at different energies from 10 to 150 keV and the def-
inition of Hine and Murty in the photoelectric case Z 3:1
eff 7a [19]. He
shows that the effective atomic number depends on the energy,
and the obtained values can differ from that calculated theoreti-
cally. Therefore he cannot establish a general definition of Zeff,
but a particular value can be found at a given energy.
Still assuming that Zeff depends on the energy used, and that the
effective atomic number of a sample is ‘‘equivalent to an element
with atomic number Z’’, Gowda et al [11] define it as the ratio:
ra
Z eff 10 ¼ : ð12Þ
rel
ra is the ‘‘cross-section per atom on average in the alloy’’ previously
defined by Guru Prasad, rel is the electronic effective cross section
expressed as:
 
1 X Ai l ni
rel ¼ f i with f i ¼ P :
NA i Zi q i j nj
For thermoluminescent materials (compounds such as CaCO3 or C4-
H6BaO4), Gowda presents the Zeff values, obtained after a logarith-
 
mic regression analysis. The values obtained with lq are
  measured
very close to those obtained from q l
calculated from the
tabulated
XCOM database at different energies. However, this work was
carried out only in the monochromatic case at energies greater than
200 keV.
3.3. Comparison of different formula
Table 1 shows some examples of Zeff values for a selection of
compounds: Alcohol, Polyethylene, Silicone, Water, Nylon, Sugar,
Salt and Silica and some explosives TNT, C4, PETN and RDX.
Fig. 1 shows the Zeff values of the five first columns (for a selec-
tion of materials in Table 1) as a function of q.
When looking at the distribution of different materials in the
(q, Zeff) basis, one can observe a wide range of Zeff values for a same
compound, depending on the relation used. The observed gap
between the values is not systematic. For example, the Guru Prasad
values are lower than the Tsai ones for alcohol or water and greater
for salt or silica.
To better differentiate one material from another, we are
tempted to choose one method rather than another. For example,
the Puumalainen method might be more adapted than Glasser to
distinguish sugar and pentrite.
In our context of dual-energy CT, are we entitled to use any
method? What is hidden behind the concept of effective atomic
number? Clearly, the concept of atomic number, as a physical
property of an element, should be revisited for a compound.
4. Zeff Interpretation: proposed definition
Our concept of Zeff consists in relating the behavior of a com-
pound towards a particular X-rays measurement, denoted W. W re-
fers to any kind of experiment, like measurement of Compton
scattering, Rayleigh to Compton ratio [7] or transmission for
example.
We propose a general definition where any compound is seen as
an ‘‘average’’ element:
For a given experiment and experimental conditions laid down, any
compound giving a W signal is equivalent to a chemical element
ð13Þ whose atomic number is Z = Zeff and giving the same W signal.
According to this new definition, Zeff depends on:
– The type of experiment (scattering, transmission, . . .).
– The experimental conditions (energy, angles, geometry, . . .).
– The characteristics of the materials (chemical composition,
mass percentage of the constituents, mass density, . . .).
This means that the effective atomic number is no longer an
intrinsic characteristic of matter but it is rather a parameter
depending also on the experiment and acquisition conditions, such
as the attenuation coefficient.
It is then possible to express the physical quantity measured, W,
as a function of the newly defined Zeff:
W ¼ f ðZ eff ; E; q; . . .Þ
The determination of a unique Zeff is possible as long as the f
function is monotonic. It is worth noting that this unique Zeff value
is obtained at one given energy and thus this concept has to be
discussed in the case of dual energy.
4.1. Illustration for carbon and water
This definition is applied to carbon and water to show the
difference of behavior between an element and a compound.

Table 1
Results for some Zeff values for a selection of materials (bold material names are those represented Fig. 1).
Vol.
Zeff 9 Zeff 9 Zeff 9
Sample mass Zeff 2 Zeff 5 Zeff 6 Zeff 8
(g.cm-3) (40keV) (80keV) (150 keV)

Alcohol (C2H6O) 0.79 7.03 5.65 6.04 2.88 3.42 2.96 2.89
Polyethylene (C2H4) 0.92 5.95 5.51 5.28 2.66 2.95 2.70 2.66
Silicone(C2H6OSi) 0.96 12.05 10.28 9.05 3.99 6.84 4.60 4.11
Water (H2O) 1,00 7.96 7.49 7.22 3.32 4.28 3.48 3.35
Nylon (C4H27O3N6) 1.14 7.10 5.02 6.21 2.91 3.49 3.00 2.93
Sugar (C12H22O11) 1.58 7.34 6.46 6.70 4.03 4.79 4.17 4.06
Trinitrotoluene/TNT
1.65 7.27 6.17 6.92 5.52 6.09 5.64 5.54
(C7H5O6N3)
C4 (C3H6O6N6) 1.67 7.34 6.05 6.97 5.00 5.71 5.15 5.03
Pentrite/PETN
1.77 7.58 6.86 7.26 5.57 6.28 5.74 5.61
(C5H8O12N4)
Hexogen /RDX
1.81 7.41 6.20 7.11 5.41 6.08 5.57 5.45
(C6H6O6N6)
Salt (NaCl) 2.16 15.68 8.36 14.64 13.99 14.88 14.65 14.26
Silica (SiO2) 2.32 12.30 6.65 10.81 10.00 11.07 10.53 10.13
226 A. Bonnin et al. / Nuclear Instruments and Methods in Physics Research B 318 (2014) 223–231
Fig. 1. Representation of some materials from Table 1 (written in bold) in the (q, Zeff) basis using the method of Glasser, Schatzler, Tsai, Puumalainen and Guru Prasad at
40 keV.
Fig. 2. Linear attenuation coefficient as a function of the effective atomic number. Examples are given for carbon (left, qC = 2.25 g cm3) and water (right qH2O = 1 g cm3) at
40 and 80 keV. The dotted lines correspond to measured values.
Let’s consider that we measure the linear attenuation coeffi-
cient l at energy E:
W ¼ l ¼ fE ðZ eff ; qÞ
 
From lq values and for a given element (the mass density q
tabulated
is a priori known), the discrete l = f(Z, q) function can be obtained.
By linear interpolation between consecutive points, the continuous
l = f(Zeff, q) function is derived.
Fig. 2 shows the curves l = f(Zeff) for carbon (mass density
qC = 2.25 g cm3, Fig. 2 left) and water (mass density
qH2 O = 1 g cm3, Fig. 2 right) at the two energies chosen (ELE = 40
keV and EHE = 80 keV). In addition, the dotted lines represent the
experimental measurements W = l (for both energies).
For the lower energy, the intersection between the dotted line
l = lLE and the curve lLE = f(Zeff) gives a first value of Zeff(ELE). For
the higher energy, a second value Zeff(EHE) is obtained.
In the case of an element such as carbon (Fig. 2 left), Zeff(ELE) =
Zeff(EHE) = 6. This Zeff is unique and corresponds to the atomic num-
ber of the element Z. It is worth noting that this conclusion is valid
for all elements.
However, in the case of a compound (e.g. water, Fig. 2 right), the
effective atomic number Zeff is not unique and depends on energy
(Zeff(ELE) = 8.4 and Zeff(EHE) = 11.4).
5. Concept of effective mass density
Let us recall that in the frame of dual-energy CT, the goal is to
assign any material a couple (q, Zeff) from the two physical quanti-
ties measured l.
First, a three-dimensional map of the theoretical surface
lthE ¼ fE ðq; Z eff Þ is built representing all possible theoretical values
of the linear attenuation coefficient, computed for all couples
(q, Z) in the range 3 6 Z 6 40 and 0 6 q 6 4 (see Fig. 3A at 40 keV
and Fig. 3B at 80 keV). A dot of this surface does not necessarily
correspond to an existing material as lth E has been computed from
 
the lq values and then multiplied by all possible q values.
tabulated
Let’s consider a material featured by a lth LE and lHE coefficients.
th
Each of these values can be plotted as a plane in the 3D map.
At a given energy, the intersection between this plane and the
3D map is a curve, representing the set of all (q, Z) couples which
could have this l value. The two curves have a unique intersection
when projected in the (q, Z) plane, as illustrated in Fig. 3C.
An important point to notice is that the obtained q value does not
correspond necessarily to the known mass density (Fig. 4). Thus, the
concept of effective mass density is introduced and denoted qeff. The
effective atomic number and effective mass density are no more
intrinsic properties of a material, but they rather represent physical
quantities that are accessible via dual-energy measurement of l.
 A. Bonnin et al. / Nuclear Instruments and Methods in Physics Research B 318 (2014) 223–231 227

μ
ρ ρ

Zeff

ρeff

Fig. 3. (A & B) 3D theoretical map of linear attenuation coefficients at 40 keV (A) and 80 keV (B). The horizontal plane represents the measured l value at the two energies.
The intersection of the measured l plane and the 3D map gives all the potential corresponding (qeff, Zeff) values. (C) The (qeff, Zeff) couple of the given material is given by the
intersection of the two projected curves.

Table 2
Coefficients a, b and m obtained from the fit of the tabulated values of the linear For an easier implementation, the analytical expression (Eq. (2))
attenuation coefficient at 40 and 80 keV (see Eq. (2)). has been used to fit the tabulated values of the linear attenuation
Energy (keV) a (105) b m (102) coefficient. The coefficients a, b and m are given in Table 2 for 40
and 80 keV. Using the fitted surfaces at ELE = 40 keV and EHE = 80
40 47.775 2.729 9.036
80 3.079 2.936 14.660 keV, the proposed method has been applied for water as a matter
of illustration (Fig. 4).

6. Energy influence
It is worth noting that this couple (qeff, Zeff) is only valid for this
particular measurement, i.e. dual energy CT at those two energies.
Fig. 5 represents the Zeff = fE(qeff) functions for water at different
Another energy choice would lead to another couple of values (See
energies (from 30 to 150 keV, per 10 keV step). The intersection
part 6). However, the very important point is that this couple is
point between two of these curves represents the (qeff, Zeff) couple
unique for given measurement and material.
for these particular energies considered.

ρ
μ

μ
μ

Fig. 4. Determination of (qeff, Zeff) couple at 40 and 80 keV for water.

228 A. Bonnin et al. / Nuclear Instruments and Methods in Physics Research B 318 (2014) 223–231

Fig. 5. Evolution of (qeff, Zeff) as a function of the energy couple (LE, HE). On the left, curves represent the function Zeff = fE(qeff) from E = 30–150 keV. On the right, this is a
zoomed area of the intersection zone where each dot corresponds to a particular couple of energies, i.e. values of (qeff, Zeff).

Table 3
th
Results of ðqth
eff ; Z eff Þ for a selection of materials for 40 and 80 keV (explosives are in grey cells).

Sample Formula
g.cm-3
(%)
Alcohol C2H6O 0.79 0.192 0.144 0.91 7.59 15.2
Polyethylene C2H4 0.92 0.209 0.168 1.07 7.23 16.3
Silicone C2H6OSi 0.96 0.400 0.190 1.08 10.34 12.5
Water H2O 1,00 0.268 0.184 1.14 8.12 14.0
Nylon C4H27N6O3 1.14 0.281 0.209 1.32 7.64 15.8
Sugar C12H22O11 1.58 0.384 0.275 1.72 7.86 8.9
Trinitrotoluène (TNT) C7H5N3O6 1.65 0.391 0.277 1.73 7.91 4.8
C4 C3H6N6O6 1.67 0.401 0.284 1.77 7.93 6.0
Pentrite (PETN) C5H8N4O12 1.77 0.435 0.299 1.86 8.08 5.1
Hexogen (RDX) C6H6N6O6 1.81 0.436 0.305 1.90 7.99 5.0
Salt NaCl 2.16 1.806 0.507 2.23 14.60 3.2
Silica SiO2 2.32 1.081 0.449 2.44 11.25 5.2

In the case of water the variation of qeff is small, from 1.13 to

1.15, i.e. 2% of variation, but it can be noticed that the value is
far from the real value of qwater (>13%). For Zeff, the higher the ener-
gies, the lower the Zeff. The variation is about 0.44 (from 8.17 to
7.73), i.e. 5% of variation. This variation of Zeff shows that the
dependence of Z with energy must be taken into account.

7. Application

7.1. Calibration: (qeff, Zeff) values at 40 and 80 keV for a representative

set of materials
Fig. 6. Scheme of a polyethylen cylinder with holes of different diameters
m
To determine the ðqm eff ; Z eff Þ values of an unknown material, a containing one material under study.
calibration in three steps is necessary.
th
The theoretical surface lth th
LE ¼ fLE ðqeff ;  Z eff Þ and lth
HE ¼
th th
fHE ðqeff ;  Z eff Þ are obtained after interpolating and fitting the
tabulated mass attenuation coefficient (see Section 5).
Knowing the theoretical values of lth LE and lHE of a representa-
th

tive set of materials for two energies chosen, we determine their

th
ðqth
eff ;  Z eff Þ couples with the method explained Section 5: their
lLE and lthHE values are reported on their respective 3D theoretical
th

map.
The attenuation coefficients of each calibration material are
measured giving the experimental calibration surfaces for this set
of materials.
Thus, for each material and for the energies chosen, four values
th
are obtained: qth eff ; Z eff ; lLE ; lHE . Calibration surfaces can be
m m

obtained by representing: Fig. 7. 3D simulated scene.

 A. Bonnin et al. / Nuclear Instruments and Methods in Physics Research B 318 (2014) 223–231 229

(
Z th
eff ¼ gðlLE ; lHE Þ
m m number of points describing the surfaces: the higher the number
: ð14Þ of materials, the better sampled the surface.
qtheff ¼ hðlmLE ; lmHE Þ
In Table 3, the calibration has been done for the chosen energies
(40 and 80 keV) and for the set of materials already used in Table 1.
Although in practice, lm
LE and lHE are obtained from experi-
m
For this particular set of materials, the range of Z th eff values
ments, in this study we have simulated them. The number of mate- stretch from 7.23 to 14.60. The real mass density q values and
rials chosen for the calibration is important. It determines the the determined qth th
eff values are close, although qeff values are

Fig. 8. (a) Phantom used to simulate dual energy CT. (b, c) Simulated tomography slices reconstructed using a FBP algorithm. Result of the dual energy CT analysis: (d)
effective mass density image (e) effective atomic number image.
230 A. Bonnin et al. / Nuclear Instruments and Methods in Physics Research B 318 (2014) 223–231

Table 4
Results of the dual energy CT analysis at 40 and 80 keV: average values of the attenuation coefficients for the raw images (lm LE and lHE )
m
m
compared to the theoretical values of the attenuation coefficients at these two energies (lth
LE and lHE ). Average qeff and Z eff are also reported
th m
th
from the processed images with their corresponding r errors. They can be compared to the calculation of qth eff and Z eff .

Sample ±σ ±σ
(cm-1) (cm-1) (cm-1) (cm-1) (g.cm-3) (g.cm ) -3

Alcohol 0.193 0.192 0.145 0.144 0.91±0.02 0.91 7.59±0.27 7.59

Polyethylene 0.210 0.209 0.168 0.168 1.07±0.02 1.07 7.22±0.25 7.23
Silicone 0.401 0.400 0.191 0.190 1.08±0.02 1.08
10.33±0.1
10.34
0
Water 0.269 0.268 0.184 0.184 1.14±0.02 1.14 8.12±0.14 8.12
Nylon 0.280 0.281 0.208 0.209 1.31±0.03 1.32 7.64±0.18 7.64
Sugar 0.385 0.384 0.275 0.275 1.72±0.02 1.72 7.86±0.11 7.86
TNT 0.388 0.391 0.276 0.277 1.72±0.02 1.73 7.91±0.09 7.91
C4 0.401 0.401 0.283 0.284 1.77±0.02 1.77 7.93±0.12 7.93
Pentrite/PET
0.437 0.435 0.301 0.299 1.87±0.03 1.86 8.08±0.05 8.08
N
Hexogen/RD
0.437 0.436 0.306 0.305 1.91±0.02 1.90 7.99±0.07 7.99
X
Salt 1.805 1.806 0.507 0.507 2.23±0.03 2.23
14.60±0.0
14.60
5
Silica 1.078 1.081 0.448 0.449 2.43±0.03 2.44
11.24±0.0
11.25
4

always greater than q, the highest difference being for Polyethyl-

ene (16.3%) and the lowest for salt (3.2%).
0.55

0.5 Salt
7.2. Luggage inspection
0.45 Silica
In this section, our method is illustrated on a 3D dual energy CT
0.4
for luggage inspection and explosives detection.
A representative 3-D phantom is modeled, including several 0.35
μmHE(cm-1)

materials of interest (explosives and common materials). Images RDX

0.3 PETN
presented below have been obtained by sinogram simulation using C4
Sugar TNT
the Virtual X-ray Imaging code [8,9]. This simulation code devel- 0.25
oped at INSA-Lyon, is based on ray-tracing techniques and on the Nylon
0.2 Water Silicone
X-ray attenuation law. Projections can then be reconstructed using
0.15 Polyethylene
any adapted algorithm (in our case, we used the FBP algorithm).
Alcohol
0.1
0 0.5 1 1.5 2
7.2.1. Simulation and samples
μmLE (cm-1)
Polyethylene cylinders (see Fig. 6), with holes of different
diameters, are modeled. The holes in each cylinder are filled with
Fig. 9. Distribution of materials in the (lm
LE , lmHE ) basis.
a different compound: seven ‘‘common’’ materials (silicone, sugar,
nylon, salt, alcohol, water and sand) and four explosives (C4, PETN,
RDX, TNT).
For our simulation, the eleven cylinders are placed in a cone
beam CT system (Fig. 7). The device consists of a monochromatic 15 Salt
X-ray source and a matrix detector. Samples are placed between
14
the source and the detector, and rotate over 360°, with 650
projections. 13

12 Silica
7.2.2. Results 11 Silicone
Zeff

Fig. 8(a) represents the disposition used for our simulation.

Fig. 8(b) and (c) present raw CT images obtained after simulation 10
and reconstruction at 40 and 80 keV. The grey level of each pixel 9
Water C4 PETN
of the image represents the linear attenuation coefficient at the Alcohol Nylon Sugar
8
corresponding energy: ðlm LE Þpixel and ðlHE Þpixel .
m
RDX
TNT
From the raw images, the next step is to calculate for each pixel 7
m
Polyethylene
(i.e. for each l value), the corresponding values of ðqm eff ; Z eff Þ. To 6
do that, we use the calibration surfaces obtained at the calibration 0.5 1 1.5 2 2.5
stage. Two new images are thus built, one corresponding to the ρeff (g.cm-3)
effective mass density (Fig. 8(d)), the other corresponding to the
m
effective atomic number (Fig. 8(e)). Fig. 10. Distribution of materials in the (qm
eff ; Z eff ) basis.
 A. Bonnin et al. / Nuclear Instruments and Methods in Physics Research B 318 (2014) 223–231
Areas have been selected in order to calculate average values
and standard deviation r. Table 4 presents the average values ob-
tained for both raw images (at 40 and 80 keV) and processed
m computerized tomography, Phys. Med. Biol. 21 (5) (1976) 733–744.
images (qm eff and Z eff ).
The mean values of the simulated linear attenuation coefficients
are close to theoretical values. The standard deviation (±0.01 for
lmLE and even less for lmHE ) is only due to reconstruction noise as
we did not introduce any noise in the simulation process. The
results obtained after computing (qeff, Zeff) are close to the theoret-
ical values recalled in Table 3. However, the standard deviation is
bigger due to the (qeff, Zeff) processing. Figs. 9 and 10 show the
m
distribution of materials in the (lm LE , lHE ) and (qeff , Z eff ) basis,
m m
respectively.
Globally, the (qeff, Zeff) distribution improves material discrimi-
nation. For example, silicone is much more separated from nylon
and water thanks to its Zeff. Explosives are close together in the
graph, well separated from most common materials, except sugar.
8. Conclusions
Dual energy CT consists of measuring two linear attenuation
coefficients of a material allowing derivation of its mass density
and atomic number. The relation between l, q and Z is well known
for chemical elements but not for compounds, for which even the
notion of Zeff is not well established. After a short review of the
state of the art for determining an effective atomic number, this
work presents a general definition of Zeff based on the physical
quantity measured and thus, on the experimental conditions.
For the particular case of dual-energy CT, this definition leads to
introduce the concept of effective mass density, qeff. Both qeff and
Zeff are no longer intrinsic properties of materials but they charac-
terize their behavior towards a specific experiment, such as cross
sections.
Thanks to this new approach, we may obtain a unique couple
(qeff, Zeff), featuring the compound, for a given couple of energies.
This is an effective method to exploit dual-energy CT for a mass
density and atomic number decomposition.
In this paper, we have illustrated the method in the context of
explosives detection. However, we did not investigate the influ-
ence of experimental conditions, our goal being to develop a rigor-
ous method. Despite the fact that experimental conditions were
not optimized, the distribution of materials in the (qeff, Zeff) basis
improves their discrimination. However, some cases remain
critical, such as sugar compared to explosives. Others complemen-
tary techniques may be used to try to solve the problem. In the
framework of another project, X-ray diffraction has been investi-
gated and results lead to promising results [20,25,5].
Note we have developed this method with two monochromatic
energies; for practical use, however, the method should be general-
ized to polychromatic beams, which will be our next step.
Acknowledgments
The authors wish to thank G. Peix and J.M. Dinten for fruitful
discussion and initiation of the project. We would like also to thank
CEA-Leti and Thales Components & Subsystems which are our
partners in this project, and more particularly the ‘‘Ministère de
l’Intérieur de l’outre-mer et des collectivités territoriales’’ for
funding.
231
References
[1] R.E. Alvarez, A. Macovski, Energy-selective reconstructions in X-ray
[2] M.J. Berger, J.H. Hubbell, S.M. Seltzer, J. Chang, J.S. Coursey, R. Sukumar, D.S.
Zucker, Xcom: photon cross sections database, National Institute of Standards
and Technology (NIST), page NIST Standard Reference Database 8 (XGAM),
2010. http://physics.nist.gov/cgi-bin/Xcom/xcom3_1.
[3] J.G.C. Coenen, J.G. Maas, Material classification by dual-energy computerized
X-ray tomography, Int. Symp. Comput. Tomography Ind. Appl. (1994) 120–
125.
[4] A.L. Conner, H.F. Atwater, E.H. Plassmann, J.H. McCrary, Gamma-ray
attenuation-coefficient measurements, Phys. Rev. A1 (1970) 539.
[5] C. Crespy, P. Duvauchelle, V. Kaftandjian, F. Soulez, P. Ponard, Energy dispersive
X-ray diffraction to identify explosive substances: spectra analysis procedure
optimization, Nucl. Instrum. Methods Phys. Res. A 623 (3) (2010) 1050–1060.
[6] D.E. Cullen, J.H. Hubbell, L. Kissel, Epdl97: the evaluated photon data library’,
97 version, UCRL-50400, 6 (5) (1997).
[7] Ph. Duvauchelle, G. Peix, D. Babot, Effective atomic number in the Rayleigh to
Compton scattering ratio, Nucl. Instrum. Methods Phys. Res. B 155 (1999) 221–
228.
[8] Ph. Duvauchelle, N. Freud, V. Kaftandjian, D. Babot, A computer code to
simulate X-ray imaging techniques, Nucl. Instrum. Methods Phys. Res. B 170
(2000) 245–258.
[9] N. Freud, P. Duvauchelle, J.M. Letang, et al., Fast and robust ray casting
algorithms for virtual X-ray imaging, Nucl. Instrum. Methods Phys. Res. B 248
(2006) 175–180.
[10] O. Glasser, Physical Foundations of Radiology, Harper, 1947.
[11] S. Gowda, S. Krishnaveni, T. Yashoda, T.K. Umesh, T. Gowda, Photon mass
attenuation coeffcients, effective atomic numbers and electron densities of
some thermoluminescent dosimetric compounds, Pramana (Journal of
Physics) 63 (3) (2004) 529–541.
[12] S. Guru Prasad, K. Parthasaradhi, W.D. Bloomer, Effective atomic numbers for
photoabsorption in alloys in the energy region of absorption edges, Radiat.
Phys. Chem. 53 (1998) 449–453.
[13] G.E. Hine, Secondary electrons emission and effective atomic number,
Nucleonics 10 (1952) 9–15.
[14] J.H. Hubbell, S.M. Seltzer, Tables of X-ray Mass Attenuation Coefficients and
Mass-Energy Absorption Coefficients from 1keV to 20MeV for Elements Z=1 to
92 and 48 Additional Substances of Dosimetric Interest, Report NISTIR 5632,
1995.
[15] T.R.C. Johnson, B. Krauss, M. Sedlmair, M. Grasruck, H. Bruder, D. Morhard,
et al., Material differentiation by dual energy CT: initial experience, Eur. Radiol.
17 (2007) 1510–1517.
[16] L.A. Lehmann, R.E. Alvarez, A. Macovski, W.R. Brody, Generalized image
combinations in dual KVP digital radiography, Med. Phys. 8 (5) (1981) 659–
667.
[17] A. Macovski, R.E. Alvarez, J.L. Chan, J.P. Stonestrom, L.M. Zatz, Energy
dependent reconstruction in X-ray computerized tomography, Comput. Biol.
Med. 6 (4) (1976) 325–336.
[18] S. Manninen, S. Koikkalainen, Determination of the effective atomic number
using elastic and inelastic scattering of gamma-rays, Int. J. Appl. Radiat. Isot. 35
(10) (1984) 965–968.
[19] R.C. Murty, Effective atomic numbers of heterogeneous materials, Nature 207
(1965) 398.
[20] A. Peterzol, P. Duvauchelle, V. Kaftandjian, P. Ponard, Modeling-based
optimization study for an EDXRD system in a portable configuration, Nucl.
Instrum. Methods Phys. Res. A 654 (1) (2011) 450–463.
[21] P. Puumalainen, H. Olkkonen, P. Sikanen, Assesment of fat content of liver by
photon scattering technique, Int. J. Radiat. Appl. Instrum. 28 (1977) 785–787.
[22] V. Rebuffel, J.-M. Dinten, Dual-energy X-ray imaging: benefits and limits,
ECNDT 2006 – Th.1.3.1, 2006.
[23] C. Robert-Coutant, V. Moulin, R. Sauze, P. Rizo, J.M. Casagrande, Estimation of
the matrix attenuation in heterogeneous radioactive waste drums using dual-
energy computed tomography, Nucl. Instrum. Methods Phys. Res. A 422
(1999) 949–956.
[24] H.P. Schatzler, Basic aspects on the use of elastic and inelastic scattered
gamma-radiation for the determination of binary-systems with effective
atomic numbers of less than 10, Int. J. Appl. Radiat. Isot. 2 (30) (1978) 115.
[25] F. Soulez, C. Crespy, V. Kaftandjian, P. Duvauchelle, Diffraction peaks
restoration and extraction in energy dispersive X-ray diffraction, Nucl.
Instrum. Methods Phys. Res. A 654 (1) (2011) 441–449.
[26] R.C. Spiers, Effective atomic numbers of heterogeneous materials, Nature 207
(1946) 398.
[27] C.M. Tsai, Z.H. Cho, Physics of contrast mechanism and averaging effect of
linear attenuation coefficients in a computerized transverse axial tomography
(ctat) transmission scanner, Phys. Med. Biol. 21 (4) (1976) 544.