Article Text
PDF
Abstract
Background Dual-energy CT (DECT) has been shown to be a useful modality in neuroradiology.
Objective To assess failure modes and limitations of DECT in different neuroimaging applications.
Patients and methods Dual-source DECT scans were performed in 72 patients over 30 months to differentiate contrast agent staining or extravasation from intracranial hemorrhage (ICH) (n=40); to differentiate calcium from ICH (n=2); for metal-artifact reduction (n=5); and for angiographic assessment (n=25). A three-material decomposition algorithm was used to obtain virtual non-contrast (VNC) and iodine (or calcium) overlay images. Images were analyzed in consensus by two board-certified radiologists to determine the success of the algorithm and to assess confounding factors. Furthermore, a dilution experiment using cylinders containing defined heparinized swine blood, normal saline, and selected iodine concentrations was conducted to assess other possible confounding factors.
Results Dual-energy analysis was successful in 65 (90.2%) patients. However, the algorithm failed when images were affected by beam hardening (n=3, 4.2%), the presence of a fourth material (parenchymal calcification) (n=3, 4.2%), or motion (n=1, 1.4%). In the dilution experiment, a saturation effect was seen at high iodine concentrations (≥37 mg/ml). VNC and iodine overlay images were not reliable above this concentration, and beam-hardening artifacts were noted.
Conclusions DECT material decomposition is usually successful in neuroradiology. However, it can only distinguish up to three preselected materials. A fourth material such as parenchymal calcium may confound the analysis. Artifacts such as beam hardening, metallic streak, or saturation effect can also impair material decomposition.
- CT
- Hemorrhage
- Complication
Statistics from Altmetric.com
Introduction
Clinical applications of CT have continuously broadened and its diagnostic value has increased thanks to technological advances such as helical and multi-detector acquisition. The recent introduction of dual-energy CT systems (DECT) promises further clinical advances.1
DECT depends on differences in X-ray attenuation of specific materials. For each material, the energy dependence of the absorption is specific to the atomic number, thus enabling material analysis in addition to the spatial distribution of the attenuation provided by conventional CT.2
Because different tissues are composed of different elemental populations, they can, in theory, be characterized by scrutiny of their absorption spectra. In practice, these differences are generally small, although medically relevant differences can be seen for soft tissue (approximating water), iodine, and calcium/bone.2–6 Several clinical DECT applications in neuroradiology have been introduced based on these differences, including automated bone and plaque removal for CT angiography and virtual unenhanced CT.7 ,8 Using a three-material decomposition algorithm, DECT has proved to be very helpful in the differentiation of contrast-related hyperdensity from acute intracranial hemorrhage (ICH).7 ,8
Although DECT appears to be a promising option, its limitations in neuroimaging have not been systematically studied. Knowledge of these limitations is essential before DECT can be considered as a standard modality in neuroradiology. In this study, a retrospective analysis was performed to analyze failure modes and limitations of DECT in neuroradiology. To further illustrate potential limitations of DECT, clinical analysis was supplemented with an in vitro dilution experiment using cylinders containing predetermined concentrations of heparinized swine blood, normal saline, and iodine.
Patients and methods
Patients
This study was approved by the local institutional review board (IRB protocol #2008P002351) of our hospital.
Data were retrieved from 72 consecutive patients who had been scheduled for neurointerventional treatment and who received head CT scans using dual-source DECT in our neuroradiology department between October 2009 and March 2012. These patients had been referred for assessment of treatment-related complications after catheter angiography for intra-arterial therapy (n=40), for differentiating possible ICH from hyperattenuation foci such as calcification (n=2), for metal-artifact reduction (n=5), and for assessment of supra-aortic vessel stenosis (n=25). The mean age of the patients was 64.9 years (range 28–94). There were 45 men (mean age 63.1 years; range 28–94) and 27 women (mean age 68.7 years; range 43–84).
Dual-energy CT scanning
DECT scanning was performed on a Somatom Definition (Siemens Medical Solutions, Forchheim, Germany). Non-contrast CT examinations were performed using the following protocol for identification of ICH: tube A, 80 kVp, 499 mA; tube B, 140 kVp, 118 mA (effective mA 714 and 168, respectively) and a collimation of 32×1.2 mm. For head and neck CT angiography or metal-artifact reduction, the protocol was slightly altered by raising the tube voltage on tube A and using a tin (Sn) filter on tube B: tube A, 100 kVp, 120 mA; tube B, 140 kVp (Sn), 118 mA (effective mA 404 and 168, respectively) and a collimation of 32×1.2 mm. For CT angiography, 80 cc of Isovue-370 (iopamidol, Bracco Diagnostics Inc, Princeton, New Jersey, USA) was administered via the intravenous route through an 18 gauge IV line at an injection rate of 5 cc/s. The timing of the injection was governed by bolus tracking over the ascending aorta.
The average CT dose index-volume (CTDIvol) for a dual-energy head scan was 66 mGy, which is similar to that for a single-energy conventional head CT and within the American College of Radiology guidelines (CTDIvol up to 80 mGy).
Image reconstruction
Projection data acquired at low and high energy were reconstructed separately, generating three sets of images: low-energy, high-energy, and mixed images. Images were reconstructed at 1.5 mm section thickness, with 0.7 mm increments (D30 s kernel). The mixed-image set is a weighted sum of low-energy (80 or 100 kV) and high-energy (140 kV) images and simulates an equivalent 120 kV image, in a 70% and 30% ratio, respectively, for the first protocol and in an equal ratio for the second protocol.
Dual-energy postprocessing
Dual-energy postprocessing was performed using dedicated software (Syngo Dual Energy Brain Hemorrhage, Siemens Healthcare, Forchheim, Germany) on a separate workstation (Leonardo, Siemens Healthcare, Forchheim, Germany).
For ICH identification, a three-material decomposition algorithm was used based on brain parenchyma, hemorrhage, and iodine. Virtual non-contrast (VNC) and iodine overlay images were derived from the original low and high kV datasets. The different patterns of hyperattenuation on VNC and iodine overlay images were analyzed.7
Image analysis
Images were analyzed in consensus by two board-certified radiologists to determine the success of the three-material decomposition algorithm and to assess for confounding factors. A study was regarded as limited if both operators concluded that there was scope for multiple interpretations, when artifacts confounded the processing to the extent that certain portions of the study could not be interpreted or when questions arose about the validity of material characterization offered by DECT.
Phantom experiment
A phantom was created that consisted of a 6×6 matrix of 10 mL plastic tubes containing precalibrated, homogeneous mixtures of iodinated contrast (Isovue-370, Bracco Diagnostics, Princeton, New Jersey, USA), heparinized swine blood (Rockland Immunochemicals, Inc, Gilbertsville, Pennsylvania, USA), and normal saline. The amount of blood increased from 0% to 25%, in increments of 5%, as one moved up the matrix from row to row. Similarly, iodine concentrations increased from 0 to 92.5 mg/mL, in increments of 5% (18.5 mg/mL), arranged in the matrix from left to right, column by column. Two additional tubes containing undiluted blood and undiluted iodinated contrast were imaged as controls. The phantom was scanned on the same dual-source scanner, operated in dual-energy mode using the first protocol described above.
Results
Material decomposition was successful in 65/72 (90.2%) patients in the study. Perfect agreement was observed between the radiologists (Cohen's κ coefficient=1). The DECT material decomposition algorithm failed in seven patients for the reasons described below.
Failure due to the presence of a fourth material
In three cases (4.2%) the three-material decomposition algorithm failed and iodine was not identified owing to the presence of a fourth material: parenchymal calcification. Figure 1 shows a patient with mineralization of the brain parenchyma, in which calcification appears as a hyperdensity on both the VNC and iodine overlay images. Another patient exhibited intraparenchymal calcification after partial surgical resection of a meningioma (figure 2), which confounded the three-material decomposition algorithm.
There is a chronic infarct in the right middle cerebral artery territory with diffuse mineralization in this region (circled). A single-energy image (A) and virtual non-contrast image (B) show hyperdensity (mean of 58 HU) surrounding infarction of the right basal ganglia and adjacent internal capsule. There is trace corresponding hyperdensity on the iodine overlay image (C). This finding, by itself, may represent mineralization or a combination of iodine and hemorrhage. Hard-plaque removal software (D) cannot identify this region of faint, diffuse mineralization.
One week after partial surgical resection of a left falcine and parafalcine meningioma. The parafalcine hyperdense residual mass with subarachnoid and subdural hemorrhage (arrowhead) is seen in the surgical bed (A). Hyperattenuations (arrows) within the residual mass are consistent with foci of calcification. The calcifications and the hemorrhage are seen on the virtual non-contrast images (B). Iodine-only images show enhancement of the residual mass with associated calcifications (C). It is clear from these images that the calcified portions of the meningioma are seen on both virtual non-contrast and iodine-only images. ‘Hard-plaque’ dual-energy application software allows differentiation of dense calcification (in pink) from iodine contrast (in blue) (D). However, this identification works only for voxel values >100 HU. Furthermore, the iodine content of the skull base and occipital bone areas is misclassified, probably owing to beam hardening.
Failure due to beam hardening/metallic artifact
In three cases (4.2%), the three-material decomposition algorithm was affected by beam-hardening or metallic-streak artifact. In cases in which a very dense object is present, such as metallic intracranial clip (figure 3), the three-material decomposition algorithm may fail to detect superimposed hemorrhage or iodine in the vicinity of the metallic object. This artifact predominantly affects the low-energy image.
Single-energy image (A) with beam-hardening artifacts from clips on a right middle cerebral artery aneurysm. An iodine overlay image (C) is particularly impaired by the metallic artifact. The virtual non-contrast image (B) is less affected by the metallic artifact.
Failure due to motion artifact
Misregistration due to motion was seen in one patient (figure 4). Although, the two image datasets were obtained at virtually the same time (difference of one quarter of the rotation time), motion impaired the image quality in this case, and dual-energy post-processing was not possible because the 80 and 140 kV images were not co-registered.
Fused dual-energy CT material decomposition image. When motion impairs the image quality, the two sets of images are not co-registered, resulting in degradation of both image quality and dual-energy postprocessing.
Phantom study: saturation effect
In addition to streaking artifacts, the dilution experiment demonstrated a saturation effect at high iodine concentrations. CT attenuation numbers increased, as expected, as the tube voltage decreased from 140 to 80 kV. However, CT attenuation reached a plateau at a value of 3071 HU. This value was reached in the 80 kV imaging chain at ≥37 mg/mL of iodinated contrast. The plateau effect was also seen on the iodine overlay images. The tubes containing >37 mg/mL iodine were mapped as white instead of showing progressively higher intensity on the overlay image, as was expected (figure 5) and, consequently, quantitative analysis of the VNC and iodine overlay images was not reliable. Moreover, beam hardening and streaking artifacts may impair the quality of the three-material decomposition at iodine concentrations <37 mg/mL.
Dilution experiment using cylinders containing defined quantities of heparinized swine blood, normal saline, and selected iodine concentrations. The increase in CT attenuation numbers reached a plateau at 3071 HU, as the iodine concentration increased. The 80 kV imaging chain (A) reached a plateau for all tubes containing ≥37 mg/mL iodinated contrast. The 140 kV imaging chain (B) reached a plateau with an iodine concentration of 92.5 mg/mL (last column on the right). Consequently, the tubes containing >37 mg/mL iodine are mapped as white instead of showing progressively higher intensity on the overlay image (D). Note the streaking/beam-hardening artifacts, particularly visible on the virtual non-contrast (VNC) images (C).
Discussion
Compared with standard CT, DECT enables characterization and quantification of tissue composition and contrast agents, based on differences in attenuation at two X-ray energies of different spectra.9 ,10 These capabilities of DECT can be applied to reconstruction of VNC images, contrast medium accentuation, characterization of urinary stones, discrimination between calcification and contrast enhancement in solitary pulmonary nodules, differentiation between iodinated contrast and hemorrhage, and diagnosis and management of gouty arthropathy.2–6 ,11
DECT data can be obtained using different methods, including dual-source CT scanning, rapid kV switching,12 and multilayer ‘sandwich’ detectors.13–15 Dual-source DECT scanning, which was assessed in this study, is performed using two X-ray tubes and two corresponding detectors.16 ,17
Two different DECT protocols were used in this study. The material decomposition results should be the same for these two protocols. However, the signal-to-noise ratio (SNR) in the derived material-specific images (eg, the iodine image or the VNC image) may be different. The SNR depends on the dual-energy (DE) contrast—that is, the difference between the DE ratios of the materials involved. The DE contrast is determined by the spectral separation and the effective atomic numbers of the materials.15 To determine a specific DE ratio, one has to measure the mean CT number of different concentrations of a material (eg, calcium or iodine) as a function of X-ray beam energy. For each material, the slope of the line relating the CT number to material density is determined using a best-fit analysis. The DE ratio corresponds to the ratio of the high and low kV slope. According to the literature, the DE ratio for the 80/140 kV mode is 1.64 for calcium and 1.99 for iodine.15 This ratio for the 100/140(Sn) kV mode, where the higher kV tube employs tin (Sn) filtration, is 1.55 for calcium and 2.20 for iodine.15 Therefore, while the spectral separation of both protocols is sufficient for material decomposition, the SNR of the iodine map is higher using 100/140(Sn) kV mode. For calcium, this situation is reversed and the 80/140 kV mode provides a better SNR in the calcium-specific images.
DECT material decomposition is usually successful in neuroradiology. Its limitations are rare and uncommon. The intention of this paper is to sensitize radiologists to the failure modes and how one can guard against misinterpretations in DECT. However, it is important to mention that this work does not generally apply to the area of spectral CT but is specific to dual-source DECT scanning and its software for postprocessing, which were assessed in this study. New technological developments in CT technology (eg, counting detectors) might be able to deal with the current limitations of dual-source DECT that are further discussed below.
Presence of a fourth material
Dual-source DECT allows one to split every voxel in the acquired (80, 140 kV) image-pair along any two preselected base materials; any residual ‘error term’ that does not fit the model can then be attributed to a third material.
The software calculation relies heavily on the assumptions made about the three materials under consideration. However, if a fourth (or more) material, such as calcium, is present at a certain concentration in a voxel, DECT cannot separate the constituent materials and will misclassify them. In cases of parenchymal calcification, prior imaging, when available, helps in detecting this kind of failure.18 Another method of distinguishing between calcium and contrast staining is to evaluate the pattern of hyperdensity on the VNC and iodine overlay images. In general, hyperattenuation due to calcium on a VNC image will closely mirror that on the iodine overlay image. This is because the software ascribes each voxel containing calcium erroneously as containing an iodine component as well as a component that reflects some combination of brain parenchyma and ICH. However, since iodine staining and hemorrhagic conversion are two separate processes, the distribution of iodine and parenchymal calcium is not expected to be identical in the presence of combined hemorrhage and iodine staining. Therefore, the regions of hyperdensity are not expected to be exactly superimposable.7
An alternative method for the determination of parenchymal calcium as a fourth material involves additional processing steps. Using the specific absorption of different materials, it is possible to focus on the differentiation between calcium and iodine. This is accomplished by the hard-plaque application in the dual-energy Syngo software (Siemens Medical Solution, Forchheim, Germany).19 Figure 2 shows an example of an analysis using this application, in which some foci of calcification were identified and color-coded. However, this application works only for dense calcifications such as those seen in the choroid plexus, dura, falx cerebri, or tentorial leaflets. Faint, diffuse mineralization may be so close to brain parenchyma in its spectral properties that it eludes detection by the algorithm (figure 1). Furthermore, the hard-plaque application is optimized for vessel analysis and, therefore, uses a lower density limit to assess the presence of calcium or iodine (a minimum of 100 HU). Any voxel with attenuation below this value will appear as normal on brain parenchymal images.
Figure 6 shows a more general processing algorithm using the brain mineralization application (Siemens Medical Solution, Forchheim, Germany. If the presence of calcium is suspected in the VNC and iodine overlay images, reprocessing the DECT images with this application will show hyperattenuation due to calcium and iodine on the ‘mineralization’ overlay map but not on the brain tissue overlay (figure 7). Further research is needed to characterize the behavior of diffuse calcification in ICH imaging with DECT.
A proposed algorithm for assessing intraparenchymal calcification using dual-energy CT processing. The original 80 and 140 kV images are decomposed into two alternate base-pairs: brain parenchyma and calcium. A hyperdensity disappearing on the brain overlay can be regarded as a calcification. ICH, intracranial hemorrhage.
Two types of hyperattenuation seen on a mixed image (A, D) obtained by dual-energy CT in a patient who underwent recanalization therapy. Contrast staining (oval) in the right basal ganglia is also depicted in the iodine overlay image (C) but not in the virtual non-contrast (VNC) image (B). A faint focal mineralization is seen on the left lentiform nucleus (arrow). The iodine-specific material decomposition algorithm cannot identify this fourth material which is seen on both VNC (B) and iodine overlay image (C). After postprocessing using the brain mineralization application, this hyperdensity disappears on the brain overlay (E), confirming a calcification. Note that both iodine content and calcifications are seen on the ‘calcium overlay’ (F).
Saturation effect
The material decomposition algorithm assumes accurate CT attenuation and is likely to fail for very high iodine contrast concentrations, as shown in the dilution phantom experiment.20 DECT analysis fails for highly attenuating objects (eg, undiluted iodinated contrast) that saturate the measurement (first seen in the 80 kV imaging chain). A conventional CT scanner has a dynamic range of 12 bits (0–4095), which spans the interval from −1024 HU (air) to +3071 HU (densest observable object). In other words, the maximal CT attenuation number measurable with the regular CT technology is 3071 HU.
This attenuation number is generally not seen in the intracranial and upper cervical regions because of dilution. For example, CT numbers reach a maximum average of about 505 HU in the aorta during an injection of iopromide 300 (0.3 g/kg body weight) at the rate of 5 mL/s.21 However, this limitation may pose challenges when we attempt to extend the analysis to the lower cervical region close to the subclavian vein when it is filled by undiluted iodinated contrast, or to brain regions close to highly attenuating objects, such as intracranial metallic devices (eg, proximity to a platinum coil).
The saturation effect was also shown by Takahashi et al,22 who found that the DECT iodine subtraction algorithm failed in their in vitro experiment, which had been designed to evaluate the feasibility of using VNC images to depict urinary stones in an iodine bath.
Beam-hardening/metallic spray artifact
Artifacts in CT, such as beam-hardening or metallic artifacts (arising from aneurysm clips or coils) can also confound dual-energy analysis. The attenuation value in Hounsfield units of a CT image is a product of multiple different corrections. Any over- or undercompensation (eg, improper application of cupping correction to account for beam hardening) can change the dual-energy index and lead to erroneous material decomposition. This type of artifact was noted in three patients in this study and in the phantom experiment in the tubes in which iodine concentration was >37 mg/mL. The iodine overlay reconstructions are especially sensitive to beam hardening, as is seen in figure 4 (see online supplemental file). This limitation has been seen in pulmonary DECT. Indeed focal iodine defects not related to any pulmonary embolism are often found in DECT angiography scans, most of which are caused by contrast material in the superior vena cava and appear in the apices of both upper lobes.23 In our study, patients with metal clips or coils were scanned using the 100/140 kV protocol with tin filtration, so that the low-energy imaging chain was less affected by beam hardening.
Spatial and temporal registration
Misregistration artifacts in dual-source DECT have been reported in thoracic imaging. These are mostly due to cardiac motion. They are negligible in the DECT technique for neuroradiology since the two image datasets are obtained almost simultaneously (difference of one quarter of the rotation time). In some instances of extensive motion (eg, pulsation with in an aneurysm sac), exact co-registration of the 80 and 140 kV images is not feasible and quantitative analysis is not possible.
Other limitations
Previous studies24 ,25 have reported other limitations of DECT that we did not see in our patients. In the following sections we discuss these limitations:
Cross-scattering
Single-source CT images are already degraded by X-ray scattering, in which an X-ray changes direction within the patient and is absorbed by an incorrect detector. This process is exacerbated with two sources, as X-rays from the first source could be scattered sufficiently for absorption in the second detector, and vice versa.25 Model-based scatter correction algorithms in DECT and measurement-based scatter correction performed in second-generation dual-source DECT can mitigate the negative effects of cross-scatter.24
Limited field of view
Another potential limitation of DECT is that the maximum field of view is limited to 25 or 33 cm (respectively, for first- and second-generation dual-source DECT scanners). However, this limitation is not important in neuroradiology if the patient is placed in the center of the field of view. When imaging the thoracic and lumbar spine the spinal column is adequately covered, even though some of the peripheral anatomy may be outside the field of view of DECT.
Conclusion
DECT is particularly helpful in the differentiation of intracranial hemorrhage from iodinated contrast material staining after intra-arterial stroke therapy. However, DECT can only distinguish up to three preselected materials. A fourth material such as parenchymal calcium may confound dual-energy analysis. This is particularly true in cases of diffuse mineralization, which should be sought on prior imaging. Moreover, certain artifacts in CT such as beam hardening or metallic artifacts may also impair the accuracy of material decomposition.
References
Footnotes
Contributors All authors fulfill the International Committee of Medical Journal Editors’ requirements having made substantial contributions to the study, participated in the design, execution, and analysis of the manuscript, and approved the final version. They agree to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.
Competing interests None.
Ethics approval Local institutional review board MGH/partners (IRB protocol #2008P002351).
Provenance and peer review Not commissioned; externally peer reviewed.