Key words therapy planning - oncology imaging - lesion detection - review - CT - spectral CT
Introduction
Thanks to increasing availability, spectral CT systems are more frequently being used
in daily clinical routine. Spectral CT, which allows specific material characterization,
can be used for a number of different applications in oncological imaging [1 ]. In addition to improved detection and characterization of malignant lesions, spectral
CT allows precise treatment planning and provides a novel imaging biomarker for tumor
vitality [2 ]
[3 ]
[4 ]
[5 ]. This article discusses specific spectral CT systems and spectral reconstructions
with a focus on oncological applications. In addition, spectral CT-specific pitfalls
that need to be known in order to avoid interpretation errors are discussed.
Basics of spectral CT imaging
Basics of spectral CT imaging
While conventional CT imaging is based on differences in physical density between
two neighboring structures, spectral CT imaging is based on differences in the basic
composition of structures [6 ]. By examining two different energy spectra, structures with a similar density but
a different basic composition can be differentiated from one another based on differences
in photon absorption. This material-specific imaging allows, for example, selective
visualization and quantification of intravenously applied iodine. However, only materials
with a strong photoelectric effect, e. g., calcium, iodine, barium, and xenon, can
be differentiated from other body tissues like fat with a weak photoelectric effect.
The two typically used energy spectra have a peak of 70–100 kilovolts (kVp) (low energy
spectrum) and 140–150 kVp (high energy spectrum) [1 ].
All currently available clinical CT systems that offer spectral imaging can be classified
into two groups: emission-based and detector-based systems [7 ]. Emission-based systems use X-ray beams with different energy spectra. This can
be achieved either by using two independent X-ray tubes with one tube generating a
low energy and the other having a high energy or by using a single X-ray tube that
quickly switches between low and high energies. A further possibility for creating
different energy spectra using only one X-ray source is to use a split beam with two
different filters, e. g. of tin and gold, during a 360° rotation, thus allowing filtration
into a low and a high energy spectrum [8 ]. Detector-based systems are based on the detectorʼs ability to separate energies
by separating the signals of the low-energy X-ray photons from the high-energy photons.
This separation can be achieved using a dual-layer energy-integrating detector with
different photon registration in every layer (e. g., using an yttrium-based scintillator)
or a photon-counting detector. The most common and most widely available clinical
spectral CT systems in Europe [2 ]
[3 ]
[4 ] are the dual-source CT system (e. g., SOMATOM Force or SOMATOM Definition Flash,
Siemens Healthineers), which is an emission-based system with two independent X-ray
tubes, the dual-layer CT system (e. g., IQon Spectral CT, Philips Healthcare), which
is a detector-based system with a dual-layer detector, and the ultrafast switching
CT system (e. g., Discovery HD, GE Healthcare), which is an emission-based system
with a single fast-switching X-ray tube. The first clinical photon-counting detector
CT system introduced last year (NAEOTOM Alpha; Siemens Healthineers), which allows
improved spectral separation, is not discussed here due to the small number of clinical
studies available.
Image reconstruction of spectral CT datasets
Image reconstruction of spectral CT datasets
Selective quantification of elements such as iodine can be achieved by a two- or three-material
decomposition algorithm in the projection domain (Ultrafast switching and dual-layer
CT systems) or in the image domain (dual-source CT systems) [7 ]. The following section describes the most common spectral CT image reconstructions
in oncological imaging.
Material density maps
Manufacturer-specific material decomposition algorithms can be used to create material
density maps that selectively display or remove materials. The most clinically relevant
application in oncology is the material density map that shows iodine or calcium and
to a lesser degree also the relative electron density and the effective atomic number. These
maps allow selective fading out of the soft-tissue background while highlighting specific
materials, the relative electron density, and the effective atomic number. Among other
things, these material density maps make it possible to characterize vascularized
and non-vascularized lesions (in the case of iodine maps), an important criterion
for the characterization of kidney or liver lesions [9 ].
On virtual non-contrast (VNC) CT images and virtual non-calcium (VNCa) CT images,
certain materials like iodine and calcium are selectively suppressed [7 ]. VNC images with selective suppression of iodine are comparable to true unenhanced
images, which are normally acquired prior to contrast administration in the case of
certain clinical questions. In oncological imaging, these images are often needed
for the characterization, for example, of incidental renal lesions during an initial
staging examination [10 ]
[11 ]. Other applications include the differentiation between therapy-induced tumor hemorrhages
and calcifications [12 ]. Due to the possibility of eliminating a true non-contrast CT examination, the radiation
dose can be reduced [13 ]. This is an important aspect of curative treatment approaches because the cumulative
radiation dose of imaging follow-up examinations can be significant [14 ].
Selective imaging of bone marrow involvement using VNCa maps e. g., in patients with
multiple myeloma, can facilitate the detection of focal lesions and the differentiation
between osteoporotic changes and plasma cell infiltration [15 ]
[16 ].
Virtual monochromatic images (VMI)
The term polychromatic X-ray beam relates to an X-ray beam with a full energy spectrum
in which the kVp represents the upper limit of the energy spectrum. It should be noted
that low-energy photons of this polychromatic X-ray beam are responsible for a disproportionately
high proportion of background image noise and image artifacts (e. g. beam hardening
artifacts). Spectral imaging can be used to create virtual monochromatic images (VMI)
from material-specific images using a complex algorithm [7 ]. These are then comparable with image data that would be generated with a theoretical
monochromatic beam. The X-ray energy is measured in kiloelectron volt (keV) instead
of kVp and the VMI spectrum is 40–200 keV with manufacturer-specific differences (see
[Table 1 ]). VMI image data reconstructed between 75–77 keV are comparable with image data
from a polychromatic X-ray beam with 120 kVp. Low-energy VMI images (40–60 keV) result
in higher iodine absorption due to the closeness to the k-edge of iodine (33 keV),
leading to enhanced iodine contrast on the image [17 ]. This has a number of advantages for oncological imaging. Among other things, low-energy
VMI images make it possible to detect lesions, e. g. hypervascularized liver metastases,
with greater sensitivity [18 ]. In the case of high-energy VMC images (170–200 keV), artifacts caused by foreign
materials, e. g., dental prostheses or prosthetic joints, can be reduced, resulting
in better assessment of the size of lesions in the immediate vicinity of such foreign
objects [19 ].
Table 1
Image types and techniques of the 3 most common spectral CT systems and most important
applications in oncology.
Technique/Reconstruction
Siemens Healthineers
Philips Healthcare
GE Healthcare
Name of viewer
Syngo.via
Intellispace portal
Advantage workstation
Image reconstruction domain
Image
Projection
Projection
No. of material decomposition techniques
3
2
2
Virtual monochromatic images
Virtual monoenergetic or monoenergetic plus
Monoenergetic (MonoE)
VMC
Kiloelectronvolt (keV)
40–190
40–200
40–140
Oncological applications
Lesion detection and demarcation from surrounding tissue, enhanced vascular contrast
e. g., acute pulmonary artery embolism detection
Material decomposition images Iodine maps
Liver virtual non-contrast (liver only); virtual non-contrast (all other organs)
Iodine no water,
Iodine density
Iodine (water/fat)
Unit of measure
1 mg/mL
1 mg/mL
100 µg/cm3
Oncological applications
Characterization of lesion vascularization, therapy monitoring including detection
of tumor hemorrhages and calcifications
Virtual non-contrast images
Liver virtual non-contrast (only for liver); virtual non-contrast (for all other organs)
Virtual unenhanced
Virtual unenhanced
Unit of measure
Hounsfield unit (HU)
HU
HU
Oncological applications
Characterization of lesion vascularization, therapy monitoring with detection of tumor
hemorrhages and calcifications
Virtual non-calcium image
Bone removal, bone marrow
Calcium suppressed
Iodine (calcium)
Unit of measure
Hounsfield unit (HU)
HU
HU
Oncological applications
Detection of bone marrow malignancies
Oncological applications
Improved detection of oncological lesions
By using low-energy VMI images or iodine maps, hypervascularized as well as faint
hypodense lesions can be better delimited from parenchymatous background tissue ([Fig. 1 ]). VMI reconstructions between 50–55 keV provide the best contrast-to-noise ratio
for the majority of parenchymatous organs. However, it should be noted that other
keV reconstructions (e. g. 70 keV) are more suitable in highly vascularized tissue
(e. g. renal parenchyma) [20 ]
[21 ]
[22 ]. A number of studies were able to show the advantage of these two spectral CT reconstructions
for malignancies in the head-neck/neck region for primary and secondary hepatic masses
and pancreatic malignancies in which the lesion-to-tissue contrast is intrinsically
lower [23 ]. For example, in the case of hepatic steatosis, either medically (e. g. during chemotherapy)
or metabolically induced, it is often challenging to detect faint hypovascularized
lesions on conventional CT. By using low-energy VMI images or iodine maps, improved
lesion-to-tissue contrast can be achieved because the lesion has a greater density
while the fatty liver tissue appears hypodense. Thus, in particular, hepatocellular
carcinomas, cholangiocellular carcinomas, and hepatic metastases can be better detected.
Fig. 1 66-year-old female patient with metastatic clear cell renal cell carcinoma to the
liver who had undergone previous atypical partial liver resection. At best, a faint
hypervascularized lesion can be detected at the margins of the atypical liver resection
in the 120 kVp equivalent images a . The iodine map images reconstructed from spectral CT b and virtual monoenergetic images at 50 keV c clearly show the hypervascularized portions of the lesion.
In patients with bone marrow involvement in a malignant disease, e. g., multiple myeloma,
MRI has been the imaging method of choice. Using spectral VNCa image data, both focal
bone marrow lesions (> 1 cm) and the pattern of involvement can be detected with a
similar accuracy to that of MRI with equally precise detection of osteolytic bone
lesions [15 ]
[16 ]. This is an advantage compared to conventional CT in which trabecular bone structures
currently cannot be sufficiently extracted making detection as well as the differentiation
between plasma cell infiltration and osteoporotic bone demineralization difficult
([Fig. 2 ]). However, it must be noted that the currently available clinical spectral CT systems
allow the generation of VNCa image data only from non-contrast spectral CT examinations.
Therefore, clinical application is currently limited. In the future, the use of the
aforementioned clinical photon-counting detector CT systems could be expanded to include
contrast-enhanced CT examinations [24 ].
Fig. 2 76-year-old patient with histologically confirmed multiple myeloma. a Conventional CT shows an inhomogeneous bone structure of the lumbar spine and femora,
but it is not possible to clearly differentiate between osteoporotic changes and multiple
myeloma involvement. In the virtual non-calcium images calculated from spectral CT
in coronary b and axial c reconstructions, bone marrow edema of the lumbar spine and distal metaphyseal femora
with density values of 12 HU compared to healthy fatty bone marrow of –97 HU is shown
as an indication of plasma cell infiltration related to multiple myeloma.
Patients with a malignancy have an increased risk of pulmonary embolism [25 ]. Depending on the phase, pulmonary embolisms are sometimes overlooked in staging
examinations, which are primarily performed in a portal venous phase [26 ]. Both low-energy VMI images and iodine maps provide better detection of pulmonary
embolisms due to the improved thrombus-to-vessel contrast or the evaluation of a filling
defect within the vessel ([Fig. 3 ]) [27 ].
Fig. 3 75-year-old female patient with metastatic malignant melanoma undergoing immunotherapy.
In the 120 kVp-equivalent images a , a faint filling defect of the lower lobe arteries can be seen. Virtual monoenergetic
images reconstructed from spectral CT at 45 keV b and iodine map images c show a clear filling defect suggestive of segmental pulmonary embolism.
Characterization of oncological lesions
Detection of vascularization of solid masses is an important clinical objective of
all imaging methods since it can help to characterize masses. In conventional staging
CT examinations performed in a single phase, incidentally detected lesions, particularly
adrenal and renal lesions often cannot be characterized in greater detail (Fig, 4,
5). As a result, it is recommended to perform supplementary imaging, follow-up examination,
or biopsy. Spectral CT imaging can provide important additional information both in
primary staging and in follow-up examinations with various image reconstructions and
allow early characterization [2 ]
[3 ]
[4 ]
[5 ]. This can reduce the patientʼs emotional burden, accelerate the introduction of
an appropriate treatment regimen, and reduce treatment costs.
VNC images or iodine maps reconstructed from a spectral CT dataset allow differentiation
between a vascularized and non-vascularized lesion which is particularly helpful for
the differentiation between a hemorrhagic/protein-rich cyst and a solid hepatic or
renal mass [9 ]
[28 ]. Based on the cutoff value of true non-contrast image data, a vascular lesion is
defined as an increase in the density (measured in Hounsfield units (HU)) between
the VNC and the contrast-enhanced image data of > 20 HU [29 ]. When using iodine maps, there are manufacturer-specific differences regarding the
definition of a contrast-enhancing lesion with iodine concentration cutoff values
ranging from 0.5 mg/mL to 2.0 mg/mL [30 ]
[31 ]
[32 ]. For the dual-source CT systems most commonly available in Germany, a cutoff value
of 0.5 mg/mL has been established [4 ]. Iodine maps can also be helpful for the differentiation between thromboses without
iodine uptake and a tumor thrombus with iodine uptake, e. g. in hepatocellular carcinoma
or renal cell carcinoma with venous infiltration.
VNC images and specific material density maps can also provide additional information
about the composition of a lesion, particularly about fatty, hemorrhagic, or faint
calcified areas [1 ]. For example, this is helpful in the characterization of adrenal masses, in which
fat-isodense lesion portions (density values in the VNC datasets < 10 HU) indicate
an adenoma [33 ]. The differentiation, for example, between a postoperative hematoma and a new metastasis
can be difficult in the case of conventional single-phase CT. However, VNC images
or iodine maps can be used for precise characterization without supplementary imaging.
Treatment planning and treatment monitoring
Precise knowledge of tumor location, extent, and relationship to surrounding tissue
and vascular structures is important for treatment planning. In particular, image-based
therapeutic methods like radiofrequency ablation, stereotactic radiotherapy, and intraarterial
therapies (e. g. selective internal radiotherapy or transarterial chemoembolization)
require imaging that is as exact as possible in order to facilitate planning.
VMI images in the low-energy range and iodine maps with a higher lesion-to-noise ratio
allow better differentiation of a tumor lesion from surrounding structures like vessels
and adjacent organs and exact determination of the number and size of lesions. Particularly
in liver or head/neck tumors, the determination of lesion margins compared to surrounding
tissue can be difficult on conventional CT imaging [34 ]
[35 ]
[36 ]. VMI images in the high-energy range can greatly reduce metal artifacts thus allowing
better delimitation of the area to be scanned, which can be advantageous, for example,
in the case of head/neck tumors with artifacts from dental fillings [37 ].
Spectral CT imaging makes it possible to estimate the relative electrode density (ρe)
and the effective atomic number (Zeff ). Determination of these tissue parameters is of particular interest for dose calculation
in radiation therapy treatment planning. They can be used either directly (e. g. the
relative electron density) or as a substitute for other parameters (e. g., the effective
atomic number as replacement for the average excitation energy) to calculate the stopping
power ratio [38 ].
The objective evaluation of treatment response to systemic or local treatment is important
to monitor the effect of oncological treatments so that treatment changes can be implemented
as early as possible. The established and currently used imaging criteria for evaluating
response, e. g. the “Response Evaluation Criteria in Solid Tumors” (RECIST1.1) and
the modified versions (iRECIST and irRECIST), are based on serial measurement of tumor
size. However, purely size-based response criteria overestimate and underestimate
the treatment success of modern therapies that are not necessarily cytotoxic but rather
cytostatic and thus do not necessarily result in a change in size. The treatment effect
can be better quantified by biomarkers that characterize tumor vitality (e. g. metabolic
or diffusion-weighted imaging) [39 ]. Spectral CT imaging provides advantages here compared to conventional CT since
it allows differentiation between iodine uptake in vital tumor tissue after i. v.
contrast injection and, for example, treatment-induced tumor hemorrhage ([Fig. 6 ]) [40 ]. Additional advantages include the elimination of errors that can occur in connection
with a spatial misregistration between non-contrast and contrast-enhanced datasets
(e. g., due to a different breathing positions between the examinations). In particular,
vitality quantification based on iodine maps is a promising approach that could have
an advantage over the above-mentioned purely size-based response criteria both in
systemic therapies (e. g. gastrointestinal stromal tumor under systemic tyrosine kinase
inhibitor therapy) [41 ] and in locally ablative methods (e. g., evaluation of the short-term and long-term
treatment success after hepatic and renal radiofrequency/microwave ablation) [42 ].
Fig. 6 51-year-old female patient with a metastatic gastrointestinal stromal tumor to the
liver on sunitinib therapy on baseline CT (a–b) and follow-up CT after 3 months (c–d).
Baseline spectral CT scan shows a vital tumor fraction with a difference between virtual
unenhanced images a and 120 kVp equivalent images b of 47 HU (25 HU to 73 HU). Follow-up spectral CT examination shows a size progression
of the lesion with diffuse density-enhanced portions within the lesion. The differences
between the virtual unenhanced images c and the 120 kVp equivalent images d is 6 HU (64 HU to 70 HU) and suggests a pseudoprogression with intra-tumoral hemorrhage
under sunitinib therapy. This is confirmed in the follow-up examinations, where the
lesion shows a stable size again without vital tumor portions.
Artifacts and pitfalls
Based on the special reconstructions algorithms, there are specific artifacts and
pitfalls with regard to spectral CT. As in the case of conventional polychromatic
image data, the density values of VMI image data depends on the energy. Therefore,
for example, the liver parenchyma can be 110H HU at 50 keV, 80 HU at 70 keV, and 65
HU at 140 keV [43 ]. Thus, the above-mentioned determination of the cutoff value for contrast enhancement
can only be applied to VMI images that are equivalent to a polychromatic image dataset
at 120 kVp (75–77 keV).
Numerous studies were able to demonstrate an excellent correlation between the density
values of VNC images and true non-contrast image data with density differences < 5
HU [11 ]
[44 ]
[45 ]
[46 ]
[47 ]
[48 ]
[49 ]
[50 ]
[51 ]
[52 ]. However, it should be noted that the density values for VNC images are subject
to various influencing factors, e. g. patient habitus or the contrast phase, with
some differences between VNC and true non-contrast image data being significant, which
can result in an incorrect classification of lesions [11 ]
[52 ]. A common cause of these discrepancies is incomplete subtraction of iodine from
VNC images, which can be observed, in particular, in image areas with a very high
iodine concentration. For example, contrast pooling in the pelvicalyceal system can
result in incomplete subtraction, resulting in incorrectly high density values in
the VNC images, leading to a risk of misinterpretation of blood clots or calcifications
within the pelvicalyceal system. It should be noted that the application of iodine-containing
materials, e. g., lipiodol in transarterial chemoembolization, results in a subtraction
from the VNC images, thus complicating post-interventional characterization of remaining
vital portions.
Moreover, calcium is selectively characterized using a cutoff value on iodine maps
as well as VNC images. This results in small (< 2 mm) or faint calcifications (< 380
HU) being incorrectly extracted from the VNC image [53 ]
[54 ]. This effect is enhanced particularly in the case of images with high background
noise, for example, in obese patients. In these cases, a true non-contrast examination
can sometimes be indicated.
Clinically accepted areas of application
Clinically accepted areas of application
In spite of the numerous potential applications of spectral CT mentioned above, only
VMI and VNC image data has achieved clinical acceptance so far. For lesion detection
and determination of the local size of hypervascularized lesions (e. g., hepatocellular
carcinoma, clear cell renal cell carcinoma metastases, and neuroendocrine tumors)
and head-neck tumors, VMI image data in the low-energy range (50–55 keV) should be
used as the primary image data due to the improved lesion-to-tissue contrast and resulting
increased diagnostic accuracy [21 ]
[34 ].
VNC image data have become clinically established particularly in the evaluation of
the macroscopic fat components of incidentalomas and renal masses due to their high
diagnostic accuracy [11 ]
[33 ]
[34 ]. Precisely characterizing these lesions in single-phase CT examinations avoids additional
radiation exposure with multiphase CT examinations, subsequent costs incurred by additional
examinations, and patient uncertainty. In the case of explicit characterization of
renal lesions or adrenal lesions, using VNC datasets and consequently dispensing with
a true non-contrast examination reduces the radiation by one third (e. g. characterization
of renal lesions: spectral nephrographic phase including VNC image data and washout
phase instead of a non-contrast, nephrographic, and washout phase). In patients undergoing
treatment with tyrosine kinase inhibitors (e. g., imatinib) or monoclonal antibodies
(e. g., bevacizumab or nivolumab) with single-phase CT staging examinations, VNC image
data should always also be considered to differentiate rare pseudoprogression caused
by a tumor hemorrhage from true progression [4 ]
[11 ].
Conclusion
There are a number of possible applications for spectral CT in oncological imaging.
In particular, VMI image data in the low-energy range (e. g. 50–55 keV), iodine maps,
and VNC images provide advantages regarding the detection of, e. g., hypervascularized
tumors and the differentiation between vascularized and non-vascularized liver or
kidney lesions. Iodine quantification with respect to a tumor provides a potential
biomarker for the assessment of treatment response.
Fig. 4 53-year-old patient with a T1c bronchial carcinoma and incidentaloma of the left
adrenal gland. In the 120 kVp equivalent images a , the mass has density values averaging 27 HU. The VNC and iodine map images calculated
from spectral CT b give density values of < 10 HU (VNC) and an iodine uptake of 1 mg /mL. Thus, the
mass represents an adrenal adenoma, which is confirmed in the subsequent guideline-oriented
FDG-PET-CT examination c , in which the adrenal mass shows no increased metabolic activity.
Fig. 5 68-year-old patient with detection of an exophytic hyperdense lesion ventrally in
the middle bulb of the right kidney. The mass has density values of 99 HU on the 120
kVp equivalent images a . Given this image, it is not possible to differentiate between a hemorrhagic renal
cyst and renal cell carcinoma with certainty. The VNC b and iodine map images c calculated from spectral CT already show a density enhancement (91 HU) in the VNC
images and no relevant iodine uptake (0.3 mg/mL) in the iodine map images. Thus, the
lesion corresponds to a hemorrhagic renal cyst (Bosniak II) and was constant in size
over 3 years.
