Dual-Energy MDCT in Hypervascular Liver Tumors: Effect of Body Size on Selection of the Optimal Monochromatic Energy Level

Transcription

1 Medical Physics and Informatics Original Research Mileto et al. Effect of Body Size on Dual-Energy MDCT of Hypervascular Liver Tumors Medical Physics and Informatics Original Research Achille Mileto 1 Rendon C. Nelson 1 Ehsan Samei 2 Kingshuk Roy Choudhury 2 Tracy A. Jaffe 1 Joshua M. Wilson 2 Daniele Marin 1 Mileto A, Nelson RC, Samei E, et al. Keywords: body size, dual-energy CT, hypervascular liver tumors, MDCT, virtual monochromatic imaging DOI:1.2214/AJR Received November 13, 213; accepted after revision March 16, Department of Radiology, Duke University Medical Center, Box 388, 231 Erwin Rd, Durham, NC Address correspondence to D. Marin (danielemarin2@gmail.com). 2 Carl E. Ravin Advanced Imaging Laboratories, Duke University Medical Center, Durham, NC. Supplemental Data Available online at AJR 214; 23: X/14/ American Roentgen Ray Society Dual-Energy MDCT in Hypervascular Liver Tumors: Effect of Body Size on Selection of the Optimal Monochromatic Energy Level OBJECTIVE. The purpose of this article is to investigate the effect of body size on the selection of optimal monochromatic energy level for maximizing the conspicuity of hypervascular liver tumors during late hepatic arterial phase using dual-energy MDCT. MATERIALS AND METHODS. An anthropomorphic liver phantom in three body sizes and iodine-containing inserts simulating low- and high-contrast hypervascular lesions was imaged with dual- and single-energy MDCT at various energy levels (8, 1, 12, and 14 kvp). Dual-energy MDCT was also performed in 48 patients with 114 hypervascular liver tumors; virtual monochromatic images were reconstructed at energy levels from 4 to 14 kev. The effect of body size and lesion iodine concentration on noise and tumor-to-liver contrast-to-noise ratio was compared among different datasets for phantoms and patients. RESULTS. The highest tumor-to-liver contrast-to-noise ratio was noted at 8 kvp for all phantom sizes. On virtual monochromatic images, the minimum noise was noted at 7 kev for small and medium phantoms and at 8 kev for the large phantom. Tumor-to-liver contrastto-noise ratio was highest at kev for small and medium phantoms and at 6 kev for the large phantom (p <.1). Compared with 8-kVp images, an optimal monochromatic energy level yielded a significantly higher (p <.1) tumor-to-liver contrast-to-noise ratio for high-contrast lesions in the large body size and for low-contrast lesions in all phantom sizes. In patients, the optimal monochromatic energy level for tumor-to-liver contrast-to-noise ratio increased proportionally along with body size (p <.1). CONCLUSION. Selection of the optimal monochromatic energy level for maximizing the conspicuity of hypervascular liver tumors is significantly affected by patient s body size. V irtual monochromatic images from a dual-energy MDCT dataset are being promoted with great emphasis, because they obviate the drawbacks associated with conventional polychromatic images [1 3]. When a polychromatic x-ray beam consisting of a range of photon energies is transmitted through the human body, the low-energy photons are preferentially absorbed, producing a filtered beam with an overall increase in the mean energy [1 3]. As the mean energy of the beam moves away from the k-edge of iodine (33.2 kev), the inherent attenuation of iodinated contrast medium decreases on polychromatic images, resulting in reduced low contrast resolution between objects [1 4]. By comparison, a virtual monochromatic image reconstructed from a dual-energy CT acquisition simulates how the objects would attenuate a monochromatic x-ray beam. By interrogating the changes in attenuation over a range of discrete energy levels, the most substantial advantage of virtual monochromatic imaging is the potential for image contrast optimization, while also mitigating noise and beam hardening [1 3, ]. Preliminary experience has shown that images obtained at a monochromatic energy level of 7 kev minimize noise and improve contrast-to-noise ratio (CNR), resulting in an increase in liver lesion conspicuity and the potential for replacing conventional 12- kvp polychromatic images [1, 2]. Although an energy level of 7 kev has been invoked as the sweet spot for improving image quality, other information, including the effect of patient s body size as well as the comparison with different tube voltages, needs to be considered when attempting to optimize monochromatic imaging for specific diagnostic tasks [6]. In larger patients, more pronounced AJR:23, December

2 Mileto et al. TABLE 1: MDCT Acquisition Parameters for Dual-Energy and Single-Energy Scans CT Parameter beam hardening is thought to be responsible for a substantial decline in the inherent iodine contrast and lesion conspicuity. Therefore, selection of the appropriate tube voltage for optimizing the CNR between objects at a certain radiation dose is somewhat dependent on the size of the patient [7]. Recently, a phantom experiment showed that the iodine CNR of virtual monochromatic images is lower compared with 8- kvp polychromatic images for all body sizes [8]. To our knowledge, these data have not yet been validated in patients and there is still uncertainty about how the patient s body size could affect the optimal monochromatic energy level using dual-energy CT. This gap in knowledge may hamper implementation of monochromatic imaging in clinical practice. The purpose of our study was to investigate the effect of body size on selection of optimal monochromatic energy level for maximizing conspicuity of hypervascular liver tumors during late hepatic arterial phase using dual-energy MDCT. Materials and Methods Phantom Study Phantom design An adult-sized female anthropomorphic phantom (ATOM Phantom, model 72, Computerized Imaging Reference Systems) consisting of a muscle background, with the spine and a custom-made liver insert, was used [9, 1]. The liver insert was designed to mimic the CT attenuation of the liver parenchyma during the late hepatic arterial phase of enhancement. The attenuation of the liver insert was titrated by scanning serial dilutions of iodine solutions to achieve CT numbers of 87 ± 1. HU (mean ± standard error of the mean) at 8 Dual-Energy Gemstone Spectral Imaging Scan kvp, 7 ± 1.4 HU at 1 kvp, 66 ± 1.3 HU at 12 kvp, and 6 ± 1.2 HU at 14 kvp. These values corresponded to previously reported CT numbers measured in human liver parenchyma during the late hepatic arterial phase, from 6 consecutive patients who underwent clinically indicated contrast-enhanced MDCT scans of the liver at our institution [9, 1]. To simulate the attenuation of hypervascular liver tumors, four 1.-cm spheres were embedded in the liver insert and filled using iodine solutions with a concentration of either 2. or 2. mg/ ml, corresponding to low-contrast (barely visible) and high-contrast (readily visible) lesions. These iodine concentrations allowed contrast enhancement of 111 ± 2.6 HU and 12 ± 2.4 HU, respectively, at 8 kvp, 9 ± 2.3 HU and 1 ± 2.1 HU at 1 kvp, 77 ± 2. HU and 8 ± 1.8 HU at 12 kvp, and 73 ± 1.8 HU and 79 ± 1.6 HU at 14 kvp. Lesions were placed in the liver insert in concentric rings at a distance of 3. and 7. cm from the isocenter of the phantom [9, 1]. Single-Energy Scans, by Tube Voltage 8 kvp 1 kvp 12 kvp 14 kvp Detector configuration (mm) Tube voltage (kvp) 14/ Tube current time product (mas) Gantry revolution time (s) Acquisition mode Helical Helical Helical Helical Helical Pitch Reconstruction kernel Standard Standard Standard Standard Standard Reconstructed section thickness (mm)..... Reconstructed section interval (mm)..... FOV (cm) Volume CT dose index (mgy) Fig. 1 Pictures and corresponding transverse CT images of three different sizes of anthropomorphic liver phantom. The phantom represented the abdomen of a small patient (anteroposterior diameter, 18. cm; mediolateral diameter, 22. cm; circumference, 91 cm). In addition, one or two fat-mimicking rings measuring 7.7 and 7.9 cm in thickness were fitted around the phantom to simulate medium (anteroposterior diameter, 26.2 cm; mediolateral diameter, 3. cm; circumference, 126 cm) and large (anteroposterior diameter, 34.1 cm; mediolateral diameter, 38. cm; circumference: 161 cm) patients, respectively (Fig. 1). These three sizes were derived from parameterized equations of anteroposterior and mediolateral diameters that mapped United States adult weight ranges [9, 1]. MDCT technique The phantom was positioned at the isocenter of the gantry, with its crosssection perpendicular to the z-axis of the scanner, and was imaged with a single-source dual-energy 64-MDCT scanner (Discovery CT 7 HD, GE Healthcare). For each phantom size, one dual-energy (with fast kilovoltage switching between 8 and 14 kvp) and four polychromatic single-en- 128 AJR:23, December 214

3 Effect of Body Size on Dual-Energy MDCT of Hypervascular Liver Tumors ergy scans (8, 1, 12, and 14 kvp) were performed. The radiation output, expressed by the volume CT dose index, was kept constant for all acquisitions (Table 1). No automatic tube current modulation was used for all acquisitions. Conventional polychromatic images derived from the four single-energy scans were reconstructed by using a filtered back projection convolution kernel. Virtual monochromatic images (ranging from 4 to 14 kev energy levels) were reconstructed from the dual-energy datasets by using a projection-based material-decomposition algorithm. All images were reconstructed and viewed at a section thickness of mm and stored on a dedicated workstation with the gemstone spectral imaging viewer (Advantage for Windows, AW Server 2, release., GE Healthcare). Specific details about the algorithm adopted to create virtual monochromatic images have been previously described in the literature []. Data analysis Single-energy polychromatic images (8, 1, 12, and 14 kvp) as well as 11 datasets of monochromatic images at 1-keV monochromatic energy level increments from 4 to 14 kev were displayed as -mm-thick transverse images on the workstation equipped with a 27-inch high-resolution (matrix size, pixels) color liquid crystal display monitor (PL18M, Planar Systems). The reconstructed section thickness was selected to match the setting used in our routine clinical protocol (see the Clinical Study section later in this article for details). The mean CT numbers of the four simulated hypervascular liver lesions (in Hounsfield units) were recorded by manually placing circular ROIs (pixel number, 3). At the same level, the mean CT numbers of the liver background were obtained by the mean of four circular ROIs (pixel number, ) manually placed in the hepatic parenchyma surrounding each lesion. Noise was measured as the SD of the pixel values of ROIs Noise (SD HU) Polychromatic Energy (kvp) A (pixel number, 4) drawn in the background air anterior to the phantom within the -cm scan FOV. To ensure consistency, all measurements were performed five times for each lesion and average values were calculated. The size, shape, and position of the ROIs were kept constant between all image sets for all measurements, by applying a copy-and-paste function at the workstation. All the measurements were performed by an abdominal radiology fellow with 1 year of experience in gastrointestinal and hepatobiliary imaging. The tumor-to-liver parenchyma CNR was calculated with the following formula: CNR = (ROI lesion ROI liver ) / SD noise (1), where ROI lesion is the mean attenuation of the tumor, ROI liver is the mean attenuation of the liver parenchyma, and SD noise is the image noise [11]. According to the American Association of Physicists in Medicine Task Group 24 [12, 13], the effective diameter was determined for the three sizes of the phantom, with the following equation: (ED = AP LAT) (2), where ED is the effective diameter and AP and LAT are the anteroposterior and lateral diameters, respectively. The calculated effective diameters of the three phantom sizes (small, medium, and large) were 2.4, 28.3, and 36.2 cm, respectively. Clinical Study This retrospective single-center HIPAA-compliant study was approved by the institutional review board of Duke University, and a waiver of informed consent was obtained. Patients Fifty-six patients (3 men and 21 women; mean age, 61 ± 12 years; age range, years; mean body mass index [BMI; weight in kilograms divided by the square of height in meters], 27.8 ± 6; BMI range, ) who underwent contrast-enhanced multiphase CT of the Noise (SD HU) 1 1 liver between March 211 and March 213 for known or suspected primary or secondary hypervascular liver tumors (n = 21) or on the basis of the results of a prior cross-sectional imaging study (n = 3) were considered for inclusion into this study. Eight of these 6 patients (14.3%) were excluded because of unavailability of source data for image reconstruction (n = 4), scan time deviation from the study protocol (n = 2), or lack of an adequate reference standard (n = 2). Fortyeight patients (3 men and 18 women; age, 62 ± 12 years; age range, years; BMI, 27.2 ± 6; BMI range, ) composed the final clinical study population [11, 14] (Fig. S1, a flowchart showing study enrollment, can be seen in the AJR electronic supplement to this article, available at MDCT technique All multiphase contrast-enhanced CT scans of the liver were performed on the same single-source dual-energy 64-MDCT scanner used for the phantom study. All patients were positioned supine with feet first on the scanning table. After acquisition of anteroposterior and mediolateral digital scout radiographs, each patient was scanned in the craniocaudal direction after IV contrast medium administration during the late hepatic arterial, hepatic portal venous phase, and, for 21 patients with chronic liver disease, the equilibrium phase. To determine the scanning delay for late hepatic arterial phase imaging, the time to peak aortic enhancement was assessed by using an automatic bolus-tracking technique with automated scan-triggering software (SmartPrep, GE Healthcare). The late hepatic arterial phase scanning was started automatically 12 seconds after the trigger threshold (1 HU at 12 kvp) was reached at the level of the supraceliac abdominal aorta. The hepatic venous phase was acquired 4 seconds after the end of the arterial phase; the equilibrium phase was started at a fixed time delay of 18 seconds after the beginning of the injection. All patients received 1 ml Monochromatic Energy (kev) Fig. 2 Image noise for three phantom sizes. A, On polychromatic image, minimum noise was observed at 14 kvp for all phantom sizes. B, On virtual monochromatic images, minimum noise was observed at 7 kev for small and medium phantoms and at 8 kev for large phantom. B AJR:23, December

4 Mileto et al. of an IV nonionic contrast medium with an iodine concentration of 3 mg I/mL (iopamidol, Isovue 3, Bracco Diagnostics). The bolus of contrast medium was injected through an 18- to 2-gauge cannula inserted into a vein in the antecubital fossa by means of a dual-chamber mechanical power injector (Empower, E-Z-Em) at a flow of 4 ml/s. The late hepatic arterial phase was performed in the dual-energy mode, by using the same acquisition parameters used for the phantom study, with no automatic tube current modulation (Table 1). The hepatic portal venous and equilibrium phases were acquired with single energy at standard tube settings (i.e., tube voltage, 12 kvp; tube current, ma). These two image sets were not considered in this study because they were single-energy acquisitions. The CT images derived from late hepatic arterial phase dual-energy scans were reconstructed by using projection-based material-decomposition algorithm with standard filtered back projection reconstruction kernel. The reconstruction thickness was mm to balance image noise and spatial resolution. For each patient, 11 datasets of monochromatic images ranging from 4 to 14 kev energy levels were reconstructed and then stored on the workstation with the gemstone spectral imaging viewer. Data analysis All the measurements were performed on the workstation with the gemstone spectral imaging viewer by the same author who Tumor-to-Liver CNR Tumor-to-Liver CNR Polychromatic Energy (kvp) A performed the quantitative measurements in the phantom study. Eleven datasets of monochromatic images, at 1-keV monochromatic energy level increments from 4 to 14 kev, were displayed for each patient with a default soft-tissue window setting (window width, 3 HU; window level, 4 HU). Before the beginning of patients data sampling, two main criteria for hypervascular liver lesion inclusion were established: first, a craniocaudal diameter of 1 mm or greater was used to avoid partial volume averaging artifacts, because a -mm section thickness was used in this study; second, in patients with multiple focal liver lesions, only the three largest lesions were selected on the basis of tumor diameter to avoid data repeated-measurement bias in a single patient (i.e., the clustering effect) [1]. The mean CT numbers of hypervascular liver tumors (in Hounsfield units) were recorded by manually placing circular or ovoid ROIs, which were drawn to encompass as much of the hyperenhancing portion of the lesion as possible (mean pixel number, 32; range, 11 1). At the same level, the mean CT numbers of the liver were obtained by the mean of four circular ROIs manually placed in the hepatic parenchyma surrounding each lesion (mean pixel number, ; range, 4 7). Areas of focal change in parenchymal attenuation, large vessels, and prominent artifacts, if any, Tumor-to-Liver CNR Polychromatic Energy (kvp) Tumor-to-Liver CNR 1 1 B were avoided. Noise was measured as the SD of the pixel values (mean pixel number, 4; range, 1 7) from a circular or ovoid ROI drawn in a homogeneous region of the subcutaneous fat of the anterior abdominal wall [11]. As in the phantom study, to maximize the consistency of the results all measurements were performed five times for each lesion and average values were calculated. Size, shape, and position of the ROIs were kept constant between all image sets for all measurements, by applying a copy-and-paste function at the workstation. The tumor-to-liver parenchyma CNR was calculated. As was performed with the phantom, the effective diameter of patients was also determined (mean, 32.1 ± 4.1 cm; range, cm) according to the American Association of Physicists in Medicine Task Group 24 protocols [12, 13]. Statistical Analysis Numeric values of continuous variables were expressed as mean (± SD), and categoric variables were expressed as frequencies or percentages. In the phantom experiment, the optimal monochromatic energy level for tumor-to-liver CNR was calculated for each lesion as the maximum tumorto-liver CNR between 4 and 14 kev for monochromatic images, as well as for 8, 1, 12, and 14 kvp for polychromatic images. Similarly, the optimal monochromatic energy level for noise was Fig. 3 Tumor-to-liver contrast-to-noise ratio (CNR) for high- and low-contrast hypervascular lesions for three phantom sizes. A and B, On polychromatic images, for both highcontrast (A) and low-contrast (B) lesions, tumor-toliver CNR was highest at 8 kvp for all phantom sizes. C and D, On virtual monochromatic images, optimal tumor-to-liver CNR of both high-contrast (C) and lowcontrast (D) lesions changed depending on phantom body size, from kev for small and medium sizes, to 6 kev for large size Monochromatic Energy (kev) C Monochromatic Energy (kev) D 126 AJR:23, December 214

5 Effect of Body Size on Dual-Energy MDCT of Hypervascular Liver Tumors calculated as the minimum for each location across these same measurements. For phantom data, we analyzed the difference in tumor-to-liver CNR between polychromatic and monochromatic images using the following ANOVA model: cd ijk = μ + s i + d j + b k + sb ik + ε ijk (3), where cd ijk is the difference in optimal CNR value (polychromatic minus monochromatic) for the ith size of phantom (i = small, medium, and large), jth distance from the isocenter (j = 3. or 7. cm), and kth level of iodine concentration (k = 2. or 2. mg I/mL); µ is the baseline level of difference (for a small phantom, 3. cm in from the isocenter and with an iodine concentration of 2. mg I/ ml); s i is the effect of the size of phantom; d j is the effect of distance from isocenter; b k is the effect of the kth iodine concentration; sb ik is the interaction effect between iodine concentration and size of phantom; and ε ijk represents the error in measurement. The ANOVA model was fit using ordinary least squares. For patients, the optimal monochromatic energy level for both noise and tumor-to-liver CNR were patient-wise modeled using a similar method. The optimal monochromatic energy level was calculated as follows: p i = μ + CNR i + ED i + Age i + S i + ε i (4), where p i is the optimal reconstruction energy for tumor-to-liver CNR for the ith patient, μ is the overall mean, CNR i is the optimal tumor-to-liver CNR value for the ith patient, ED i is their effective diameter, Age i is patient age, S i is patient sex, and ε i is error in measurement. The baseline patient is a 62-year-old man with an effective diameter of 2.4 cm and optimal tumor-to-liver CNR of 9.8. All predictors in the above model were adjusted for baseline values. An analogous model was also fitted for the calculation of the optimal monochromatic energy level for noise. This model was fit using ordinary least squares. The correlation between patients effective diameter and BMI, as well as the agreement between phantom and patients results, were assessed in terms of Pearson correlation. For all comparisons, statistical significance was assumed to be p <.. All statistical analyses were performed by using the R computing platform (The R Project). Results Phantom Study Our phantom data showed that for all effective diameters, noise was lowest at 14 kvp compared with 8-, 1-, and 12-kVp images (Fig. 2A). Using virtual monochromatic images, the minimum noise was observed at 7 kev for the small and medium Fig. 4 Graph shows trend of difference in tumor-to-liver contrast-to-noise ratio (CNR) between polychromatic and monochromatic images for high- and low-contrast lesions across three different phantom sizes. Two lines represent average difference in tumor-to-liver CNR between polychromatic and monochromatic images at optimal energy levels for highcontrast (black line) and low-contrast (gray line) lesions, across different phantom sizes. Note that difference in tumor-to-liver CNR between polychromatic and monochromatic images decreases with increase of phantom size. phantoms and at 8 kev for the large phantom (Fig. 2B). Noise steeply increased at progressively lower monochromatic energies, with the highest noise at 4 kev for all phantom sizes. For all phantom effective diameters, tumor-to-liver CNR was highest at 8 kvp compared with 1-, 12-, and 14-kVp single-energy images (Figs. 3A and 3B). The highest tumor-to-liver CNR was observed at kev for the small and medium phantoms and at 6 kev for the large phantom (p <.1) (Figs. 3C and 3D). No change in the optimal energy level was observed with different iodine concentrations or distance from the isocenter of the phantom. Although the tumor-to-liver CNR of highcontrast lesions was significantly higher at 8 kvp compared with the optimal monochromatic energy levels for the small phantom effective diameter (p <.1) (Fig. 4 and Table 2), it was significantly lower for the large phantom effective diameter (p <.1). Though the tumor-to-liver CNR of high-contrast lesions was slightly higher at the optimal monochromatic energy levels for Tumor-to-Liver CNR Difference Polychromatic vs Monochromatic Images Small High-contrast lesions (2. mg I/mL) Low-contrast lesions (2. mg I/mL) Medium Large the medium phantom, no statistically significant difference was observed compared with 8 kvp. The tumor-to-liver CNR of low-contrast lesions was significantly higher on the optimal monochromatic energy compared with 8-kVp images for all phantom effective diameters (p <.1, for all comparisons). Our data showed a trend toward a decreased difference in tumor-to-liver CNR between polychromatic and monochromatic images proportional to the increase of phantom size (p <.1). Clinical Study The average optimal monochromatic energy levels for minimizing noise or maximizing tumor-to-liver CNR were 72.7 ± 4. and 2.6 ± 4.4 kev, respectively (an example can be seen in Fig. S2, which is available in the AJR electronic supplement to this article, available at The optimal monochromatic energy level for the tumor-to-liver CNR ranged between and 6 kev, and it significantly increased with the patient s effective diameter (p <.1) (Fig. A and Table 3). A similar statistically TABLE 2: Analysis of Optimal Tumor-to-Liver Contrast-to-Noise Ratio Differences Between Polychromatic and Monochromatic Imaging in the Phantom Experiment Variable Estimate Standard Error t p Baseline < < <.1 High contrast <.1 7. cm from isocenter Large, high contrast Medium, high contrast <.1 Note Baseline refers to a low-contrast lesion, 3. cm distance from the isocenter, and small phantom size. The analyzed factors are iodine concentration (low or high contrast), phantom size (small, medium, or large), and distance from isocenter (3. or 7. cm). The interaction effect between iodine concentration and phantom size is also included. AJR:23, December

6 Mileto et al. significant trend was observed for noise (p <.1), which ranged from 7 to 8 kev (Fig. B and Table 4). Excellent agreement was found in the optimal monochromatic energy levels for CNR between phantom and the predicted values from patients data at corresponding effective diameter values (r =.87) (optimal monochromatic energy levels for different effective diameters can be seen in Table S1, which is available in the AJR electronic supplement to this article, available at Discussion Our study results show that the optimal monochromatic energy level for maximizing the conspicuity of hypervascular liver tumors is significantly affected by the patient s body size, likely reflecting an incomplete correction of the greater beam hardening that occurs with increased patient s body size. Our clinical data showed that the optimal monochromatic energy level for maximizing tumor-to-liver CNR increases by.6 kev/cm of patient s effective diameter, with an average increase of 2% comparing our smallest (effective diameter, 22.2 cm) to the largest (effective diameter, 47.7 cm) patient s body size. Our results substantiate the data from a prior experiment by Yu and colleagues [8] showing that the optimal monochromatic energy level for maximizing the CNR of iodine increases with larger phantom sizes. However, there are discrepancies in the optimal monochromatic energy levels proposed in our study Optimal Monochromatic Energy Level (kev) TABLE 3: Estimated Coefficients From Multivariate Regression Model of Optimal Monochromatic Energy Level for Tumor-to-Liver Contrast-to-Noise Ratio (CNR) in Patients Variable Estimate Standard Error t p Baseline Optimal tumor-to-liver CNR Effective diameter <.1 Age Sex Note Baseline refers to a 62-year-old male patient, with an effective diameter of 2.4 cm, and an optimal tumor-to-liver CNR value of 9.8. The fitted model implies that the optimal monochromatic energy level for maximizing tumor-to-liver CNR increases by.6 kev for every centimeter increase in patient s effective diameter (9% CI, kev). TABLE 4: Estimated Coefficients From Multivariate Regression Model of Optimal Monochromatic Energy Level for Noise in Patients Variable Estimate Standard Error t p Baseline Optimal noise Effective diameter <.1 Age Sex Note Baseline refers to a 62-year-old male patient, with an effective diameter of 2.4 cm, and an optimal noise value of 9.3. The fitted model implies that the optimal monochromatic energy level for minimizing noise increases by.94 kev for every cm increase in patient s effective diameter (9% CI, kev). ( kev for small and medium sizes and 6 kev for the large size) and those recommended from Yu and colleagues (66 kev for the small, 68 kev for the medium, and 7 kev for the large) [8]. These differences may be related to differences in size between our and Yu s phantom, as well as the different dual-energy platform used [8]. For example, the use of Optimal Monochromatic Energy Level (kev) different technical solutions for virtual monochromatic image synthesis (projection-space vs image-space domain) may have affected the optimal energy level. Because dual-energy data are acquired at the same projection angle, inherent preprocessing beam-hardening corrections as well as the synthesis of virtual monochromatic images in the projec Effective Diameter (cm) Effective Diameter (cm) A B Fig. Optimal virtual monochromatic energy level as function of patients effective diameter. A and B, Scatterplots show optimal virtual monochromatic energy level as function of patients effective diameter for tumor-to-liver contrast-to-noise ratio (CNR) (A) and noise (B). Open circles represent patients. Gray circles denote estimated energy levels at three effective diameters of phantom. Vertical gray lines through gray circles are 9% CIs for predicted values. Straight line represents fitted regression of optimal monochromatic energy levels on patients effective diameter for tumor-to-liver CNR and noise. Note that optimal virtual monochromatic energy level for both tumor-to-liver CNR and noise increased with patient s effective diameter AJR:23, December 214

7 Effect of Body Size on Dual-Energy MDCT of Hypervascular Liver Tumors tion-space domain are enabled with the single-source platform. In contrast, with a dualsource system, there is an approximately 9 phase difference between the low- and highenergy projections that precludes the reconstruction of virtual monochromatic images in the projection-space domain. As a result, it could be argued that the beam-hardening correction may be fairly penalized using the image-space domain [3]. Our phantom data showed that, compared with conventional 8-kVp polychromatic images, the optimal monochromatic energy levels allow improved conspicuity of high-contrast hypervascular liver tumors in the large body size and of low-contrast hypervascular tumors for all body sizes. Our results, therefore, provide new insights in the context of existing knowledge with respect to those of Yu and colleagues [8], who observed that the optimal monochromatic energy level for the iodine CNR is significantly lower compared with 8-kVp images, for all phantom sizes. It is conceivable that variations in the phantom design may have determined the differences between our results and those of Yu and colleagues [8]. A thoracic phantom containing only elevated iodine concentrations (3. and 7. mg I/mL) was adopted by Yu and colleagues [8]; we used an anthropomorphic liver phantom with two levels of iodine concentrations (2. and 2. mg I/mL) that closely reflected human hypervascular liver tumors. The observations we made in comparing monochromatic and polychromatic images at different tube potentials have important clinical implications. The improved conspicuity of hypervascular liver tumors achieved with monochromatic images may be exploited in large patients or in clinically challenging scenarios, such as the detection of subtle hypervascular tumors in cirrhotic livers with heterogeneous parenchymal enhancement or distorted hepatic anatomy. Our study results confirm the data recently reported by Lv and coworkers [2], who observed that monochromatic energy levels between 4 and 7 kev yield improved CNR as well as increased detectability of small hepatocellular carcinomas. However, it is important to note that, although the CNR represents an objective measure of image quality, it is just a first-level indicator that does not completely encompass all of the components that influence the reader s subjective perception of image quality [16, 17]. It has been recognized that the noise, particularly the spatial correlation of noise patterns within an image, referred to as correlated noise [18 22], critically affects both the image quality and detectability of hypervascular liver tumors [17]. In some diagnostic tasks, such as the detection of readily hypervascular tumors in a healthy liver, the increased iodine contrast may be sufficient to compensate for a certain noise constraint. However, in other scenarios that involve assessment of liver with heterogeneous parenchymal texture or that require analysis of more subtle hypervascular tumors, the benefit of the higher iodine conspicuity may not offset the negative impact of the dramatically increased noise levels at lower kiloelectron voltage settings. Given the discrepancy both in our study and in the literature concerning the optimal monochromatic energy levels for CNR and noise, further investigation to elucidate the best trade-off between the two indicators is warranted. Our study has some potential limitations. First, our study did not include a diagnostic performance study of readers with different levels of experience. Because CNR is a simple approximation of the radiologists performance for lesion detection, a future reader-based qualitative study is needed to confirm our quantitative data. It should be noted, however, that the comparison between virtual monochromatic and polychromatic images is currently limited in patients, using the single source with fast-kilovoltage-switching platform, because of the unavailability of polychromatic datasets other than 14 kvp during the dual-energy scan. Second, to keep the radiation output constant among all scans, pitch was decreased from 1.37 to.984 and gantry rotation time increased from.6 to 1. seconds during the 8-kVp single-energy scan to compensate for the x-ray tube output limitation. One could advocate that lowering the helical pitch to allow the use of lower tube potentials can be tolerated only when the total scan time remains within the acceptable limits, considering the potential for a higher incidence of motion artifacts. Another potential criticism of our research is the lack of utilization of an automatic exposure control during the acquisitions in single-energy mode. Although this approach allowed for matching the tube output between single- and dual-energy acquisitions, it could have diminished the clinical value of the comparison we made between polychromatic and virtual monochromatic datasets. Finally, variances in the technical approach for creating virtual monochromatic images between the two dual-energy hardware platforms (single and dual source) currently available clinically may restrict the applicability of our study results to only users of our dual-energy system. However, given the commonality between our study and the one by Yu and coworkers [8], we think that our results may be of benefit to users of both dual-energy CT platforms. Our findings may serve as a starting point for future investigations on the optimal monochromatic imaging with other vendors. In conclusion, the results of our study indicate that selection of the optimal monochromatic energy level for maximizing the conspicuity of hypervascular liver tumors is significantly affected by the patient s body size. The use of an optimal monochromatic energy level can improve the conspicuity of hypervascular liver tumors compared with conventional polychromatic images. Acknowledgments We thank John Schiereck for the invaluable help. We also thank Baiyu Chen and Carolyn Lowry for providing help with the scans. References 1. Matsumoto K, Jinzaki M, Tanami Y, Ueno A, Yamada M, Kuribayashi S. Virtual monochromatic spectral imaging with fast kilovoltage switching: improved image quality as compared with that obtained with conventional 12-kVp CT. Radiology 211; 29: Lv P, Lin XZ, Chen K, Gao J. Spectral CT in patients with small HCC: investigation of image quality and diagnostic accuracy. Eur Radiol 212; 22: Yu L, Leng S, McCollough CH. Dual-energy CTbased monochromatic imaging. AJR 212; 199(suppl ): S9 S1 4. Wu X, Langan DA, Xu D, et al. Monochromatic CT image representation via fast switching dual kvp. Proc SPIE 29; 728:1 9. Goodsitt MM, Christodoulou EG, Larson SC. Accuracies of the synthesized monochromatic CT numbers and effective atomic numbers obtained with a rapid kvp switching dual energy CT scanner. Med Phys 211; 38: Venema HW. Virtual monochromatic spectral imaging with fast kilovoltage switching should not be used as standard CT imaging modality. Radiology 211; 26: Kalender WA, Deak P, Kellermeier M, van Straten M, Vollmar SV. Application- and patient size-dependent optimization of x-ray spectra for CT. Med Phys 29; 36: AJR:23, December

8 Mileto et al. 8. Yu L, Christner JA, Leng S, Wang J, Fletcher JG, McCollough CH. Virtual monochromatic imaging in dual-source dual-energy CT: radiation dose and image quality. Med Phys 211; 38: Husarik DB, Marin D, Samei E, et al. Radiation dose reduction in abdominal computed tomography during the late hepatic arterial phase using a modelbased iterative reconstruction algorithm: how low can we go? Invest Radiol 212; 47: Chen B, Marin D, Richard S, Husarik D, Nelson R, Samei E. Precision of iodine quantification in hepatic CT: effects of iterative reconstruction with various imaging parameters. AJR 213; 2:[web]W47 W Marin D, Nelson RC, Samei E, et al. Hypervascular liver tumors: low tube voltage, high tube current multidetector CT during late hepatic arterial phase for detection initial clinical experience. Radiology 29; 21: American Association of Physicists in Medicine. Size-specific dose estimates (SSDE) in pediatric and adult body CT examinations (task group 24). College Park, MD: American Association of Physicists in Medicine, Brady SL, Kaufman RA. Investigation of American Association of Physicists in Medicine report 24 Size-specific dose estimates for pediatric CT implementation. Radiology 212; 26: Bossuyt PM, Reitsma JB, Bruns DE, et al. Towards complete and accurate reporting of studies of diagnostic accuracy: the STARD Initiative. Radiology 23; 226: Sica GT. Bias in research studies. Radiology 26; 238: Schindera ST, Tock I, Marin D, et al. Effect of beam-hardening on arterial enhancement in thoracoabdominal CT angiography with increasing patient size: an in vitro and in vivo study. Radiology 21; 26: Marin D, Nelson RC, Schindera ST, et al. Lowtube-voltage, high-tube-current multidetector abdominal CT: improved image quality and decreased radiation dose with adaptive statistical iterative reconstruction algorithm initial clinical experience. Radiology 21; 24: Samei E. Performance of digital radiography detectors: factors affecting sharpness and noise. In: Samei E, Flynn MJ, eds. RSNA 23. Oak Brook, IL: Radiological Society of North America, 23; Boedeker KL, McNitt-Gray MF. Application of the noise power spectrum in modern diagnostic MDCT. Part II. Noise power spectra and signal to noise. Phys Med Biol 27; 2: Samei E, Ranger NT, MacKenzie A, Honey ID, Dobbins JT, Ravin CE. Detector or system? Extending the concept of detective quantum efficiency to characterize the performance of digital radiographic imaging systems. Radiology 28; 249: Richard S, Husarik DB, Yadava G, Murphy SN, Samei E. Towards task-based assessment of CT performance: system and object MTF across different reconstruction algorithms. Med Phys 212; 39: Solomon JB, Li X, Samei E. Relating noise to image quality indicators in CT examinations with tube current modulation. AJR 213; 2:92 6 FOR YOUR INFORMATION A data supplement for this article can be viewed in the online version of the article at: AJR:23, December 214