Introduction
Computed tomography (CT) was pioneered in the early 1970s by Sir Godfrey N. Hounsfield,
an English electrical engineer. The first CT scanner was a two-slice machine designed
to scan the brain with a pencil beam in a step and shoot mode. Following that, whole
body fan-beam CT scanners for general radiology were introduced. Helical CT acquisition
came into the arena in the early 1990s with the introduction of slip ring technology.
Volume date acquisition was possible due to continuous table translation in helical
CT.
In 1998 multislice CT systems were introduced by all major vendors with simultaneous
acquisition of four slices per rotation.[1] This breakthrough resulted in vast potential for increased isotropic voxel resolution,
leading to precise reformations and faster scanning times, resulting in accurate multiphase
imaging and angiography studies. Hence, there was a race for more slices in the consequent
years, which resulted in the advent of the eight-slice CT system in 2000, 16-slice,
32- slice, 64-slice henceforth. Furthermore, the sub-millimeter spatial resolution
of recent CT scanners and faster gantry rotation times up to 0.375 second have made
electrocardiogram-gated cardiac studies and angiography feasible with increased image
quality.[1]
However, the substances with different chemical compositions can show similar attenuation
(Hounsfield unit) in single-energy CT. Until recently, there was no tool for material
differentiation and tissue characterization. This problem was solved by introducing
dual-energy CT (DECT) in 2006 by Siemens (Somatom definition). DECT has its roots
as early as 1973 when Godfrey Hounsfield described that two pictures were taken of
the same slice at 100 kV and the other at 140 kV.[2] This was also further investigated by Alvarez and Macovski in 1976. They showed
that energy-dependent information could be obtained from the polychromatic X-ray spectrum
by separating the measured attenuation coefficients into their contributions from
photoelectric and Compton effects.[2]
[3]
[4] Due to the technological limitations at that time, the successful application of
these investigations was not feasible. The principle, techniques of acquisition with
their advantages and limitations, and an overview of image processing algorithms and
clinical applications will be presented in this review.
Principle
DECT, also known as spectral CT, uses two different energy spectra (low and high energy)
for imaging to make material decomposition and tissue characterization possible. Due
to two different energy spectra, data of tissue attenuation at these different energies
can be analyzed, leading to material differentiation. DECT is a subset of spectral
CT that uses two X-ray spectra for imaging which means that there is potential for
acquiring more than two energy levels by using energy resolving photon-counting detectors,
the subject of ongoing research, which can result in multi-energy CT.[2]
[5]
Energy and Material Dependence
Materials with different elemental compositions can have similar attenuation in CT
depending on their mass density. For their differentiation, it is necessary to use
a second X-ray spectrum for imaging. This provides an additional attenuation measurement,
and the difference in attenuation between the two energy spectra is because of the
energy and material dependence. Usually, in DECT, the two energy spectra used are
commonly in the range of 80 to 90 kV (low energy) and 140 to 150 kV (high energy).
Due to the maximum difference and better spectral separation, the standard peak energies
are 80 and 140 kV.[5]
[6]
For heavier patients requiring greater penetrance, 90 or 100 kV can be used for the
lower energy range. Energy lower than 80 kV is not advisable except for use in pediatric
patients because most of the photons in this energy range are absorbed by the body.
Due to suboptimal soft-tissue resolution and increased radiation dose, higher energies
greater than 150 kV could not be used. Filters can be used with the low- or high-energy
spectra to remove the overlap resulting in superior spectral separation.
X-ray attenuation by the material is accounted by the processes involved in the interaction
of photons with matter, out of which photoelectric and Compton effects play a significant
role. The Compton effect is minimally energy-dependent and depends mainly on the electron
density of the material. The photoelectric effect is strongly energy-dependent (probability
more at a low-energy range) and is critical for material differentiation. The photoelectric
effect also increases with the atomic number of the material. Most of the tissues
in the body are made predominantly of lower atomic number atoms like hydrogen, carbon,
and nitrogen and, hence, show a weak photoelectric effect except for calcium and magnesium,
which may show a more substantial effect. On the contrary, the contrast materials
used in imaging, i.e., iodine and barium, due to their higher atomic number and K-edge
effect, have a strong photoelectric effect resulting in strong spectral contrast in
DECT between the rest of the atoms of the body and the contrast material.[6] The difference in spectral properties of the materials/tissues, i.e., the difference
in atomic numbers, and the energy dependence, plays a pivotal role in optimal spectral
contrast in DECT images and thus in its clinical applications.
Technology in the Acquisition of DECT
The acquisition of two energy spectra can be done in varying ways and is broadly classified
as source-based and detector-based acquisitions ([Fig. 1]).
Fig. 1 Dual-energy computed tomography (DECT) techniques. (A) Dual-source DECT. Two X-ray tubes with corresponding detectors are mounted orthogonally
at 90°with different tube potentials. (B) Twin beam filtration. Single X-ray source with two different filters arranged along
the Z-axis. (C) Fast kilovolt switching. X-ray tube rapidly switches between the low and high voltages
in a short time (< 0.2 milliseconds) within a single gantry rotation. (D) Dual-layer detector. A single source and a layered or sandwich detector. (E) Photon counting devices.
Source-Based Acquisition
Dual-Source DECT (Ds DECT)
Two X-ray tubes with corresponding detectors are mounted orthogonally at 90° with
different tube potentials, usually 80 and 140 kV. Data are obtained by the simultaneous
acquisition of these two X-ray tubes operating at a different potential. The tube
voltage and current can be adjusted independently for both tubes resulting in optimal
spectral contrast. Prefiltration can be done by adding a filter for both tubes separately,
enabling better spectral separation and noise reduction.[5]
[7] More importantly, Ds DECT enables simultaneous imaging of a slice at two different
kilovolts without any temporal offset.[6] However, projection data are 90° out of phase because of the angular offset of the
X-ray tubes within the gantry, and so dual-energy processing algorithms are performed
in the image domain.[5]
[7] This is the technique developed by Siemens Healthineers, Germany.
Limitations
-
Small field of view (FOV): To accommodate two X-ray tubes with detectors within the
gantry, the second detector is restricted to a smaller FOV. The scan FOV for the Siemens
Ds DECT is as follows: first generation—26 cm, second generation—33 cm, and third
generation—35.6 cm.[7]
-
Cross-scattered radiation: Scattered radiation from one X-ray tube reaches the orthogonal
detector of the other X-ray tube, leading to spectral distortion. However, it can
be rectified by the appropriate algorithm.[2]
[6]
[7]
[8]
Rapid Kilovolt Switching
X-ray tube rapidly switches between the low and high voltages in a short time (< 0.2
milliseconds) within a single gantry rotation, enabling high temporal resolution and
minimal offset between the projections of two energy spectra. So, material decomposition
can be performed in the projection domain, thereby reducing the beam-hardening artifacts.[2]
[8]
[9] This technology was introduced by General Electric Healthcare, United States. Tube
current cannot be modulated for each tube potential in this configuration of rapid
voltage switching. Alternatively, asymmetric sampling can be implemented for optimizing
tube current by allowing a prolonged exposure and higher current for low kilovolts
(in the ratio of 80/140 = 60%/40%) or by acquiring two low-voltage projections for
every single high-voltage projection.[2]
[5]
[8] Spectral filtration is also not technically feasible because the same X-ray source
is used for both voltages.
Limitation
Twin Beam Filtration
In this source-based approach, a split filter is placed in front of the X-ray tube
in the collimator in a single source-detector configuration. It is composed of gold
and tin stacked adjacent to each other in a longitudinal direction to achieve spectral
separation into low- and high-energy beams, and the corresponding halves of the detector
detect the low- and high-energy spectra.[5]
[9] Tin and gold filter the polychromatic beam's low- and high-energy components, thereby
increasing and decreasing the mean energy of the beam, respectively, leading to spectral
separation.[7]
[8] This is the technique launched by Siemens Healthineers, Germany.
Advantages
-
Not much hardware modification is necessary, and hence, it is cost-effective. The
addition of a split filter to the tube collimator of an existing CT scanner is done
to perform dual-energy imaging.[5]
[7]
[9] Image acquisition is possible in the whole FOV, and automatic tube current modulation
optimizes radiation dose to patient.[5]
[7]
Limitations
-
Spectral separation resulting from a split filter is limited, compared with that achieved
in a dual-source dual-energy scanner.[8]
[9]
-
There is also potential for cross-scattering, with one side of the beam contaminating
the other side of the detector.
-
In addition, a higher power of the X-ray tube is necessary due to the prefiltration
of the beam, which is a constraint in imaging obese patients.
-
There is temporal skew between low- and high-energy data.[5]
Sequential Acquisition
This is one of the earliest approaches suggested for acquiring dual-energy images
without the need for dedicated hardware.[8] Imaging the entire volume is done sequentially in low and high kilovolts, but this
results in a long delay between the two datasets. Alternating the tube potential between
each successive gantry rotation can reduce this delay to some extent.[2]
[5]
[9]
Advantages
No significant hardware modification is needed—sequential scanning in any CT scanner
in two different voltages is acquired, and the dataset can be combined for spectral
analysis.[5]
Imaging in two different potentials at the exact location during each gantry rotation
provides the same view angle, leading to projection space alignment of the two datasets
and, hence, projection-based material decomposition.[7]
Automatic exposure control can be implemented, enabling tube current modulation for
different tube potentials providing optimal noise levels between the two potentials.[7]
Limitations
-
In sequential scanning, patient movement between the acquisitions can lead to spectral
distortion.
-
Sufficient temporal skew makes its implementation difficult in cardiac studies and
angiographic acquisitions, requiring high temporal resolution.[5] Thus, its use is limited to relatively static organs and non-contrast studies.[6]
[8]
Detector-Based Acquisition
Dual-Layer Detector
This design consists of a single source providing a high-energy beam (120 or 140 kV)
and a layered or sandwich detector. This detector consists of two layers with different
scintillator materials, and as a result, there is maximal sensitivity for different
photon energies. The inner or the top layer is made of zinc selenide or cesium iodide
crystals which absorb the low-energy photons, and the outer or bottom layer is made
of gadolinium oxysulfide, which is sensitive to high-energy photons.[2]
[5]
[7]
[8] This technique, thus, exploits the polychromatic nature of the X-ray beam.[5]
[8] The thickness of an individual layer is designed to enable comparable noise levels
across the two energy datasets.[2]
[9] This technology was established by Philips Healthcare, the Netherlands.
Advantages
Image acquisition can be performed at full FOV with no constraints in gantry rotation
time, dose optimization techniques.[7]
[8]
Since spectral separation is achieved at the level of the detector, there is a perfect
alignment of the low- and high-energy datasets with excellent temporal and spatial
registration, and hence, material decomposition can be performed in the projection
domain.[5]
[7]
The conventional dataset can be obtained by combining the low- and high-energy spectral
data and reconstructed by standard techniques of iterative reconstruction and filtered
back projection.[8]
Scan acquisition is always performed in a dual-energy mode allowing for retrospective
spectral analysis, especially useful in scenarios where there is no predetermined
indication for dual-energy scan.[5]
[7]
Limitations
-
Spectral analysis can be done only in scans performed with a higher energy beam (120
or 140 kV).[7]
-
There is suboptimal spectral resolution and material decomposition due to overlap
between the absorption properties of the scintillator material[5]
[6] (leading to unsharp distinction between low- and high-energy photons). This can
be overcome by introducing an interlayer filter between the two detector layers but
at the cost of dose efficiency.[5]
-
Photon interaction in one detector pixel can result in scattering and subsequent interaction
in another pixel, known as cross-talk between the two detector layers.[5]
[7]
Photon-Counting Detectors
This design of spectral CT is the subject of ongoing research and has the potential
for multi-energy imaging. A photon-counting detector (PCD) is an energy-resolving
detector that can count discrete photon interactions. These detectors are highly specialized
and are based on semiconductors such as cadmium telluride, cadmium zinc telluride
They convert the incident photons into electrical signals whose magnitude depends
on the energy of the incident X-ray photon. Pulse height discriminator analyses these
electrical pulses and place them in energy bins based on specific energy thresholds.[2]
[7]
[9] Electronic noise below the lowest energy threshold is wholly eliminated and, hence,
does not affect the count rates.[7]
Advantages
-
Improved spectral resolution and increased dose efficiency.
-
Higher geometrical efficiency than energy integrating detectors.[5]
-
K-edge imaging—used in material differentiation wherein higher atomic number elements
can produce a k-edge effect, which enables their differentiation from other materials.
This can potentially be used to differentiate iodine from gadolinium contrast agents
and separation of iodine, gadolinium, and bismuth in multiphasic studies.[5]
[7]
-
Inherent spatial and temporal resolution enables multiphasic acquisition and imaging
of moving organs.[7]
Limitations
-
Pulse pile-up effect occurs due to high X-ray flux in CT in the order of 109 per second
per mm2. This happens when consecutive low-energy pulses at a short interval are incorrectly
registered, leading to the loss of photon counts and increased image noise.[7]
[9]
-
Charge sharing occurs due to pulse sharing across the multiple detector pixels, leading
to incorrect counts and decreased spectral resolution.[2]
[7]
-
K escape is the absorption of photon energies due to the k-edge effect and release
of characteristic X-ray resulting in energy loss.[2]
[7]
-
Count rate drift occurs due to crystal defects in the detector, causing the trapping
of electrons and holes, leading to changes in signal pulse characteristics in detector
elements, causing ring artifacts.[7]
-
Cross-talk effect occurs when photon interaction occurs near the border of detector
pixels.[9]
DECT Images
Mixed/Blended Set of Images
A single set of blended/mixed images is generated from the low- and high-energy datasets
([Fig. 2]). The recommended blending ratio is 0.3 (70% from the high-energy dataset and 30%
from the low-energy dataset). The images generated are equivalent to the images of
conventional 120 kVp single-energy data acquisition. The mixed images have the advantages
of high contrast from the low-energy dataset and less noise from the high-energy dataset.
These images can be generated easily without the requirement of complex post-processing
analysis.[10]
Fig. 2 Axial images of the contrast-enhanced CT of the abdomen at the same level in the
arterial phase showing a hyper-vascular lesion in the right lobe of liver. (A) Low-energy (100 kV dataset), (B) high energy (150 kV dataset), and (C) mixed set.
Image-Processing Algorithms
Post-processing in DECT is an essential technical aspect for acquiring data needed
for a wide range of clinical applications. The processing of data, especially for
material decomposition, can be performed either in the projection space (pre-reconstruction)
or in the image space (post-reconstruction).
The equivalent projection data obtained in the two energies are subtracted and reconstructed
by filtered back-projection to obtain the spectral information in the projection space
method. It is quantitatively more accurate, and beam hardening artifacts are eliminated
in this method, though, in practice, they may not be removed entirely. In the image
space approach, first, the reconstruction of the low- and high-energy datasets is
done, and then, the difference in corresponding voxels in the two datasets is processed
to obtain the spectral information. It is an easier to perform and more commonly used
approach. It is preferred when there is a spatial or temporal inconsistency between
the projection data, like in the case of dual-source scanners where there is a 90°
offset between the two datasets.[2]
[6]
[7]
Material Decomposition
DECT can decompose the material into its constituent elements based on energy and
material dependence of X-ray attenuation.[2] The basic mathematical algorithm to derive the material decomposition was first
proposed in 1976 by Alvarez and Macovski, after which it had undergone subsequent
alterations.[8]
The mass attenuation coefficient of a given material can be expressed as the linear
combination of photoelectric and Compton effects, ignoring the K- or L-edge effects
unless significant. It is calculated as the linear attenuation coefficient divided
by the density of the absorber. Alternatively, any material attenuation coefficient
can be given as a linear combination of attenuation coefficients of two basic materials
present at different mass densities. For example, iodine and water are commonly used
basic materials, and other materials can be expressed as the mixed or linear combination
of iodine and water. Series of mathematical equations and derivations depicting the
X-ray attenuation of the material in two energy spectra were devised, and as a result,
mass density, atomic number, and material-specific information of given material were
obtained.[2]
[7]
[8]
[9]
Material decomposition gives the amount of primary material in the image voxel required
to produce the observed X-ray attenuation rather than the actual composition of each
image voxel.[9]
Multimaterial Decomposition
It decomposes a material in a given voxel into three or more basic materials depending
on the need in specific clinical applications. This can be done if one of the materials
has a K-edge effect, i.e., K-edge in the given energy range. The accuracy of material
decomposition increases with more number of basic materials.[8] However, noise levels in each energy spectra and the difference in the atomic number
of basic materials can influence the accuracy of material decomposition.[2]
Material-Specific Images
In material-specific images, the selected basic materials can be detected, estimated,
or subtracted according to clinical needs and used in various clinical applications.
The distribution of the given material's mass density can be mapped. Material-specific
images of the commonly used basic materials, iodine and water, are called iodine and
water maps, respectively. Materials like iodine and calcium can be removed from the
voxels to generate virtual non-contrast (VNC) and virtual non-calcium images, respectively.[7]
[8] However, the signal is also contributed by materials present other than the primary
materials in a given voxel during material decomposition. For example, in iodine-water
decomposition, there may be other materials like fat or bone, represented both in
iodine-removed (VNC) and in water-removed images. Hence, there is a difference in
the Hounsfield unit value of fat in the VNC image compared with a conventional non-contrast
image.[7] Increased noise, scatter, and beam hardening artifacts can cause error and decrease
the accuracy of material separation.[7]
Virtual Non-Contrast Images
This is one of the potential applications of material-specific images in DECT, where
the iodine component can be subtracted from the voxels in a contrast-enhanced scan
to yield a VNC image ([Fig. 3]). VNC images are being evaluated as a substitute for a proper precontrast scan in
a multiphasic acquisition and can exclude initial non-contrast scan resulting in radiation
dose reduction. It can also be used as a problem-solving tool when only a contrast
scan is acquired. A subtly increased attenuation is found: VNC can differentiate between
subtle contrast enhancement and increased attenuation in soft tissue.[2]
Fig. 3 Axial images of the contrast-enhanced CT of the abdomen at the same level in the
arterial phase showing a hyper-vascular lesion in the right lobe of liver. (A) Virtual non-contrast image, with no iodine density in the region of lesion in the
right lobe of liver (blue arrow). (B) Iodine overlay image. (C) Iodine map image.
Virtual Monoenergetic Images
Virtual monoenergetic images (VMI), also known as virtual monochromatic images, can
be generated from dual-energy acquisition, and these portray the appearance of images
when a monochromatic beam of X-rays is used. VMI has its roots in as early as the
1970s when Alvarez and Macovski suggested that CT images can be synthesized at any
energy within diagnostic range and that the images have less beam hardening artifacts
and more accurate attenuation measurements.[3]
[10]
This is mainly based on the material decomposition of basic materials. As mentioned
earlier, the linear attenuation coefficient can be given by the linear combination
of the mass attenuation coefficient of two basic materials, and the mass attenuation
coefficient can be expressed as a linear combination of photoelectric and Compton
effects. A mass density map of each primary material at the two energies is obtained
by solving the relevant equations. Monoenergetic images can then be synthesized from
mass density and mass attenuation coefficients of basic materials at each energy.[7]
[10] The usual energy range for VMI is 40 to 200 keV ([Fig. 4]), and the optimal contrast-to-noise ratio for iodine is obtained at the range of
40 to 70 keV.[2]
[7]
Fig. 4 (A–H) Virtual monoenergetic images. Axial images of the contrast-enhanced CT of the abdomen
at the same level in the arterial phase showing a hyper-vascular lesion in the right
lobe of liver in varying energy levels from 40 to 180 kV.
In low-kiloelectron-volt images, contrast enhancement is more pronounced so that even
subtle enhancement can be assessed, and there is scope for reducing contrast dose.
However, there is increased noise at lower energies compared with the higher energy
range. High-kiloelectron-volt images reduce beam hardening artifacts, photon starvation,
and metallic artifacts.
