Determination of iodine detectability in different types of multiple-energy images for a photon-counting detector computed tomography system

Abstract.

In addition to low-energy-threshold images (TLIs), photon-counting detector (PCD) computed tomography (CT) can generate virtual monoenergetic images (VMIs) and iodine maps. Our study sought to determine the image type that maximizes iodine detectability. Adult abdominal phantoms with iodine inserts of various concentrations and lesion sizes were scanned on a PCD-CT system. TLIs, VMIs at 50 keV, and iodine maps were generated, and iodine contrast-to-noise ratio (CNR) was measured. A channelized Hotelling observer was used to determine the area under the receiver-operating-characteristic curve (AUC) for iodine detectability. Iodine map CNR ($0.57\pm 0.42$) was significantly higher ($P<0.05$) than for TLIs ($0.46\pm 0.26$) and lower ($P<0.001$) than for VMIs at 50 keV ($0.74\pm 0.33$) for 0.5 mgI/cc and a 35-cm phantom. For the same condition and an 8-mm lesion, iodine detectability from iodine maps ($\mathrm{AUC}=0.95\pm 0.01$) was significantly lower ($P<0.001$) than both TLIs ($\mathrm{AUC}=0.99\pm 0.00$) and VMIs ($\mathrm{AUC}=0.99\pm 0.01$). VMIs at 50 keV had similar detectability to TLIs and both outperformed iodine maps. The lowest detectable iodine concentration was 0.5 mgI/cc for an 8-mm lesion and 1.0 mgI/cc for a 4-mm lesion.

1. Introduction

Iodine-based intravenous contrast agents are commonly used in computed tomography (CT) imaging. For instance, contrast-enhanced multiphase CT exams are considered the standard of care for detection of liver lesions.^1,2 In dual-energy CT, iodine maps and virtual monoenergetic images (VMIs) have been used to improve the visualization of iodine.³ Iodine maps have been reported to improve the diagnostic confidence for visualization of contrast-enhanced structures while suppressing noniodinated materials.⁴ VMIs at lower energy levels (keV settings) have higher iodine contrast due to the increased attenuation of iodine at energies closer to its k-edge.⁵ VMIs at 50 keV have been shown to provide superior lesion conspicuity compared to conventional CT images.⁶

Recently, photon-counting detectors (PCDs), which use semiconductor materials such as cadmium telluride, have been used for CT imaging, with research systems developed for phantom, animal, and patient imaging.^7–13 A major advantage of PCDs is that, unlike traditional energy-integrating detectors (EIDs), the detector response is relatively independent of photon energy. Another advantage of PCDs is that they provide an estimate of the energy of each detected x-ray, which can be compared against predefined energy thresholds, yielding simultaneous multi-energy data in a single acquisition.^9,14–16 This feature enables material decomposition and the generation of VMIs and iodine maps in addition to the energy-threshold and bin images.^17,18 The low-energy-threshold images (TLIs) are comparable to conventional single-energy images, in that all detected photons are used for image formation.

With the availability of multiples types of images from a single PCD-CT scan, it is important to determine which image type provides the best iodine detectability. Further, for each image type, it is important to determine the lowest detectable iodine concentration. The purpose of this study, therefore, is to determine the image type that has the highest iodine detectability, as well as to determine the lowest concentration of iodine that can be reliably detected for each image type.

2. Methods

2.1. Experimental Setup

A 10-cm diameter, water-equivalent phantom was constructed with 4-mm diameter and 8-mm diameter cylindrical holes to represent small and medium liver lesions, respectively (Fig. 1).^19,20 Iodinated contrast material (iohexol, Omnipaque 350, GE Healthcare Ireland, Cork, Ireland) was diluted to yield four different concentrations (0.2, 0.5, 1.0, and 2.0 mgI/cc). These iodine concentrations were selected so that they are relevant to clinical tasks, e.g., distinguish enhanced pathologies (as low as 0.3 mgI/cc), with the consideration of the detectable limit of current dual-energy CT technologies.^4,21–23 These iodine solutions were used to fill the phantom holes to simulate lesions of different enhancement levels. Deionized water was also used to fill the holes to represent signal-absent images for the model observer studies. The 10-cm phantom was placed into a $200\mathrm{mm}\times 300\mathrm{mm}$ anthropomorphic abdomen phantom (Quality Assurance in Radiology and Medicine, Moehrendorf, Germany), which represents the attenuation of a small adult. Extension rings with exterior dimensions of $250\mathrm{mm}\times 350\mathrm{mm}$ and $300\mathrm{mm}\times 400\mathrm{mm}$ (Quality Assurance in Radiology and Medicine) were used to represent the attenuation of a medium and large adult, respectively (Fig. 1).

Fig. 1.

Anthropomorphic abdomen phantom with extension rings to mimic small (lateral width = 30 cm), medium (lateral width = 35 cm), and large (lateral width = 40 cm) adults. The water-equivalent central 10-cm phantom insert (blue) consisted of 4- and 8-mm diameter cylindrical holes that were filled with iodine solutions to mimic enhancing lesions.

2.2. Computed Tomography Acquisition and Reconstruction

Phantoms were scanned on a whole-body research PCD-CT scanner (Definition CounT, Siemens Healthcare, Forchheim, Germany). Detailed descriptions of this system can be found elsewhere.^9,24 With four iodine concentrations and a scan where the holes were filled with water (5 conditions in total) and three phantom sizes (3 conditions in total), there were a total of 15 unique conditions to be scanned. For each condition, 15 scans were performed with the following parameters: 140 kV, 25 and 75 keV energy thresholds, and 0.6 helical pitch. For each of the three phantom sizes, tube current was adjusted so that radiation dose (in terms of volume CT dose index, CTDIvol) was matched to our routine abdominal CT protocol. To determine this dose level, the same phantom was scanned on a second-generation dual-source CT scanner (SOMATOM Flash, Siemens Healthcare) using 120 kV and 200 quality reference mAs, with automatic exposure control (AEC) on (CareDose4D, Siemens Healthcare). The CTDIvol values for small, medium, and large phantoms were 6.9, 10.8, and 17.7 mGy, respectively. The tube current was selected on the PCD-CT to achieve this same dose level for each of the phantom sizes. The scans were performed without AEC as it was not available on the PCD-CT. All CT images were reconstructed using a filtered-back projection (FBP) algorithm with a quantitative, medium smooth kernel (D30), 100-mm field of view, $512\times 512$ matrix size, 5-mm slice thickness, and 8-mm slice increment yielding 150 images for each condition.

2.3. Multi-Energy Processing

For each PCD acquisition, two energy-bin images were used for the multi-energy processing (bin $1=25$ to 75 keV and bin $2=75$ to 140 keV). VMIs at 50 keV were generated using manufacturer-provided research software (eXamine, Siemens Healthcare). Iodine maps were generated using commercial dual-energy postprocessing software (Syngo Via, Siemens Healthcare).^25,26

2.4. Contrast-to-Noise Ratio Analysis

Contrast was measured as the CT number (or iodine concentration in iodine maps) difference between 8-mm diameter simulated lesions and the water background regions; noise was measured as the square root of the CT number variance of both the lesion and water regions (averaged). Contrast-to-noise ratio (CNR) values were calculated and compared between TLIs, VMI, and iodine maps.

2.5. Model Observer Analysis

A channelized Hotelling observer (CHO) was used to assess iodine detectability in each of the three image types: TLIs, VMIs, and iodine maps.^27–29 In this method, the test variable $\lambda$ is given by

$$\lambda ={\omega }_{\mathrm{CHO}}^t{g}_C,$$

where $g_C$ is the channel output of the test image and ${\omega }_{\mathrm{CHO}}^t$ is the CHO template defined as,

$${\omega }_{\mathrm{CHO}}=S_c^{-1}[\overline{g_{\mathrm{s}\mathrm{c}}}-\overline{g_{\mathrm{b}\mathrm{c}}}]$$

where $S_C$ is the intraclass channel scatter matrix calculated as the average of the channel output covariance matrix with signal-present [signal channel (sc)] and signal-absent [background channel (bc)] images. $\overline{g_{\mathrm{s}\mathrm{c}}}$ and $\overline{g_{\mathrm{b}\mathrm{c}}}$ Here, and denote the average channel values of signal-present and signal-absent images. In this study, a $150\times 150\text{pixel}$ region of interest (ROI) was used, with lesion centered within each ROI.

Gabor filters were applied in the CHO, which have been shown to provide similar detectability performance as human visual system:²⁷

$$G(f)=\exp [-4(\ln 2)({(x-x_0)}^2+\frac{(y-y_0{)}^2}{{\omega }_s^2}\left)\right]\times \cos [2\pi f_c((x-x_0)\cos \theta +(y-y_0)\sin \theta )+\beta ],$$

where ${\omega }_s$ is the channel width, $f_c$ is the central frequency, $\theta$ is the orientation angle, and $\beta$ is a phase offset. A total of 20 channels (Fig. 2) were used in this study with 6 frequency passbands ($f_c=3/256$, 3/128, 3/64, 3/32, 3/16, 3/8 cycles/pixel; ${\omega }_s=84.72$, 42.36, 21.31, 10.66, 5.33 cycles/pixel), 4 orientations ($\theta =0$, $\pi /4$, $\pi /2$, $3\pi /4$), and 1 phase ($\beta =0$).³⁰

Fig. 2.

Gabor filters with six spatial channel passbands and four orientations that were used in the CHO model.

Internal noise was added to the test variables:³¹

$${\lambda }^{\prime }=\lambda +\alpha ·x,$$

where $\alpha$ is the weighting coefficient and $x$ is a variable following a normal distribution with mean equal to zero and standard deviation equal to the standard deviation of the test variables in signal-absent images. The area under the receiver-operating-characteristic curve (AUC) was calculated based on the 150 signal-present and 150 signal-absent images for each condition using a Wilcoxon nonparametric method. AUC was selected as the figure of merit for the detectability task.²⁸ To determine the threshold of iodine concentration, a minimum AUC value of 0.9 was considered acceptable detectability performance of each image type. This is consistent with previous clinical reports for detecting abdominal lesions.^32–34

2.6. Human Observer Studies

The internal noise ($\alpha$) was determined by matching the AUC calculated from the model observer (CHO) to the human observer two-alternative forced choice (2AFC) outcome at the condition of 4-mm lesion size and 0.5-mgI/cc concentration for VMIs at 50 keV and the 35-cm phantom size.

Although the performance of CHO has been shown to be correlated with that of human observer in multiple studies, these prior studies focused on single-energy images acquired with EID-CT systems.^28,29 Since this study involved PCD-CT and multi-energy image types, including derived VMIs and iodine maps, we performed human observer studies using a subset of images to validate the CHO performance before applying the CHO to all images to determine iodine detectability. A 2AFC study was performed with three medical physicists serving as the human observers. Each observer evaluated six conditions (two concentrations: 0.5 and 1.0 mgI/cc; three image types: TLIs, VMIs, and iodine maps). These two concentrations were chosen as it is more important to assess the detectability of challenging, low contrast lesions than obvious, high-contrast lesions to determine the overall performance of the system and diagnostic outcome. Each condition contained 100 trials corresponding to 100 repeated images, and each trial consisted of an ROI with signal centered and an ROI without signal. The two alternatives were presented side-by-side to the readers using our default abdomen viewing setting [window level = 40 Hounsfield unit (HU) and window width = 300 HU] and readers were asked to choose the side with iodine signal. Reader studies were performed using a calibrated diagnostic monitor in a dark room with controlled ambient light and a customized user interface built with MATLAB (MathWorks, Natick, Massachusetts). The detectability is assessed with an ensemble of images (100 in this study), and percent correct (PC) for the 2AFC task was calculated as the ratio between the number of correct decisions and the total number of trials. Under the assumption of a Gaussian distribution, the PC value in 2AFC tasks is equal to the AUC, which was used as the figure of merit in model observer studies.³⁵ A PC = 1.0 indicates a perfect detectability, e.g., detection of large size and high-contrast lesions; PC = 0.5 represents a random guess, such as detection of very small size and low-contrast lesions; and $0.5<\mathrm{PC}<1$ indicates a task that lesions can be accurately detected in some images but not others. For clinical tasks, the specific PC value varies but usually much higher than 0.5. Spearman correlation was performed to determine the correlation between the PC of the human observer and the AUC of the model observer. Statistical analysis was performed using a free statistical package (R Project, Version 3.4.0³⁶). $P<0.05$ was considered to be statistically significant.

3. Results

Phantom results were compared among different image types. Representative images of TLIs, VMIs at 50 keV, and iodine maps are shown in Fig. 3 for the 8-mm lesion at iodine concentrations of 0 (signal absent), 0.2, 0.5, 1.0, and 2.0 mgI/cc for the 35-cm phantom. As expected, the conspicuity of enhanced lesions improved with increasing iodine concentration for each image type. Note that the visibility of the displayed lesion at low concentrations (0.2 and 0.5 mgI/cc) in Fig. 3 might be affected by the process of converting DICOM images to figures. In addition, viewing conditions, such as ambient light, have significant impact on visibility. Note that images were reviewed on a calibrated diagnostic monitor in a dark room with controlled ambient light during the reader studies. When compared across image types for the condition of 0.5 mgI/cc and a 35-cm phantom, the CNR of iodine maps ($0.57\pm 0.42$) was significantly higher ($P<0.05$, Wilcoxon signed-rank test) than TLIs ($0.46\pm 0.26$) but lower ($P<0.001$, Wilcoxon signed-rank test) than VMIs at 50 keV ($0.74\pm 0.33$). A similar trend was observed for CNR across different concentrations and phantom sizes (Fig. 4).

Fig. 3.

Representative images of the 8-mm lesion for the TLIs, VMIs at 50 keV, and iodine maps for various iodine concentrations (0, 0.2, 0.5, 1.0, and 2.0 mgI/cc) in the 35-cm wide phantom. Window level/window width = 30/160 HU.

Fig. 4.

Iodine CNR of an 8-mm lesion comparisons between TLIs, VMIs at 50 keV, and iodine maps across different phantom sizes (30, 35, and 40 cm) and iodine concentrations (0.2, 0.5, 1.0, and 2.0 mgI/cc).

The weighting coefficient ($\alpha$) of internal noise was determined to be 4.03 (Fig. 5) based on the calibration process. Figure 6 illustrates that for selected conditions (three image types with 0.5 and 1.0 mgI/cc concentrations in the 35-cm phantom), the performance of CHO agreed well with that of the human observers, with a Spearman correlation coefficient $\rho =0.93$ and a corresponding $P$ value $<0.01$. This validated the use of the model observer (CHO) to characterize iodine detectability for multi-energy CT images derived from PCD-CT acquisitions.

Fig. 5.

Calibration of the internal noise was performed for the condition of 4-mm lesion size, VMI at 50 keV, 0.5-mgI/cc concentration, and the 35-cm phantom. The final internal noise coefficient ($\alpha$) was determined to be 4.03.

Fig. 6.

Comparison between human observer detection performance (solid triangles) with predicted performance by a channelized Hoteling observer (empty squares) for six selected conditions: 4-mm lesion size, two concentrations (0.5 and 1.0 mg/cc), and three image types (TLIs, VMIs, and iodine maps).

Iodine detectability was calculated with the calibrated model observer (CHO) and compared among the different image types (Fig. 7). For the 8-mm lesion, TLIs demonstrated comparable detectability to VMIs at 50 keV for all four iodine concentrations and three phantom sizes, whereas iodine maps showed lower AUC values compared to the other two image types. For example, detectability from iodine maps ($\mathrm{AUC}=0.95\pm 0.01$) was significantly lower ($P<0.001$, Wilcoxon signed-rank test) than both TLIs ($\mathrm{AUC}=0.99\pm 0.00$) and VMIs ($\mathrm{AUC}=0.99\pm 0.01$) for the condition of 0.5 mgI/cc and the 35-cm phantom. For the 4-mm lesion, VMIs at 50 keV had a significantly higher AUC ($0.88\pm 0.02$, $P<0.01$, Wilcoxon signed-rank test) than TLIs ($0.84\pm 0.02$) for the condition of 0.5 mgI/cc and the 35-cm phantom, whereas AUC values were comparable between the two image types at other conditions. When comparing to iodine maps, both TLIs and VMIs demonstrated superior iodine detectability for the 4-mm lesion. At the condition of 1.0 mgI/cc and the 35-cm phantom, iodine detectability from iodine maps ($\mathrm{AUC}=0.88\pm 0.02$) was significantly lower ($P<0.001$, Wilcoxon signed-rank test) than for both TLIs ($\mathrm{AUC}=0.97\pm 0.01$) and VMIs ($\mathrm{AUC}=0.97\pm 0.01$).

Fig. 7.

AUC for detection of (a)–(c) 8-mm and (d)–(f) 4-mm lesions, as calculated by the CHO model. TLI, low-energy-threshold image; VMI, virtual monoenergetic image.

The minimum detectable iodine concentrations (for an AUC value of 0.9 or higher) for each image type are given in Table 1 for each experimental condition.

Table 1.

The minimum iodine concentration required to achieve an AUC no smaller than 0.9 across different image types and phantom sizes. TLI, low-energy-threshold image; VMI, virtual monoenergetic image.

Lesion size	Image type	Phantom size
		30 cm	35 cm	40 cm
8 mm	VMI at 50 keV	0.5 mgI/cc	0.5 mgI/cc	0.5 mgI/cc
	TLI [25 140] keV	0.5 mgI/cc	0.5 mgI/cc	0.5 mgI/cc
	Iodine map	0.5 mgI/cc	0.5 mgI/cc	1.0 mgI/cc
4 mm	VMI at 50 keV	0.5 mgI/cc	1.0 mgI/cc	1.0 mgI/cc
	TLI [25 140] keV	0.5 mgI/cc	1.0 mgI/cc	1.0 mgI/cc
	Iodine map	1.0 mgI/cc	1.0 mgI/cc	2.0 mgI/cc

4. Discussion

In this study, we determined the iodine detectability for the three commonly used types of CT images (TL, VMIs, and iodine maps). Previous investigations have demonstrated that these image types, when acquired using dual-energy CT systems, could improve the detection of pulmonary artery thrombi,³⁷ the conspicuity of hepatocellular carcinoma lesions,³⁸ and the evaluation of treatment response for gastrointestinal stromal tumors.³⁹ To the best of our knowledge, this is the first study to systematically evaluate iodine detectability in multiple image types, including TLIs, VMIs, and iodine maps, for PCD-CT.

As expected, iodine detectability generally decreased with the increasing phantom size, where the increased noise, beam hardening, and scatter would be expected to degrade iodine detectability.¹⁷ Among different types of images, TLIs and VMIs demonstrated similar iodine detectability and both outperformed the iodine maps, even though the iodine maps and VMIs had higher iodine CNR than TLIs. There are several reasons for the observed discrepancies between CNR and detectability across the image types. First, detectability could be affected by spatial resolution and noise correlation changes during data processing such as material decomposition,⁴⁰ which were incorporated into the CHO but excluded in the calculation of CNR. A similar discrepancy has also been observed when comparing CNR and low-contrast detectability between iterative reconstruction and FBP images.⁴¹ Second, CNR measurements do not take into account lesion size, whereas detectability does.

Our results suggest that TLIs and VMIs at 50 keV will provide the best sensitivity for detecting subtly enhancing lesions. One possible reason that iodine maps have inferior AUC values for detecting iodine at low concentrations and small sizes could be due to the noise magnification during the material decomposition process. As such, substantial filter and noise reduction techniques are usually used, which could change the spatial resolution and noise correlation and consequently impact detectability. Although material-decomposed images (iodine maps) did not provide the best iodine detectability for low-iodine concentrations or small lesions, their ability to provide quantitative and material-specific information provide other clinical benefits, such as use of a quantitative threshold for iodine enhancement for discriminating enhancing renal masses from cysts⁴² and measuring liver fat by eliminating iodine and iron signals.⁴³ Additionally, this study was performed using a uniform background and a known lesion location where enhancement was unequivocally due to the presence of iodine. For nonuniform backgrounds, where apparent enhancement could be due to other factors, e.g., contrast enhancement or background variation due to higher atomic number or higher density materials, material-specific maps might have an advantage over TLIs or VMIs. Studies using more realistic phantoms or in patients are required to address this possibility.

For realistic lesion sizes, our results demonstrated that PCD-CT system was able to resolve 0.5 mgI/cc for 8-mm lesions and 1.0 mgI/cc for 4-mm lesions across different phantom sizes. Our results provided the potential of using PCD-CT to distinguish enhanced pathology with optimal threshold lower than 1.0 mgI/cc.^4,22 A previous phantom study concluded that the limit of iodine detection was below 0.5 mgI/cc for different dual-energy CT platforms, all of which used EIDs.²³ This does not imply that PCD-CT has inferior iodine detectability compared to dual-energy EID-CT. The difference is mainly that a much larger lesion size ($>20\mathrm{mm}$) was used in the EID-CT study, as was a different figure of merit.

In this study, a spatial-domain model observer was used for AUC calculation, after validation with human observers. Although previous studies have showed that both Fourier and spatial-domain models were able to predict human performance,^28,44 the spatial domain method was used in this study. This is due to the fact that Fourier domain models tend to overestimate the detectability performance for CT images derived from nonlinear operations,⁴⁵ which in this study refers to the denoising process performed during material decomposition. To confirm the appropriateness of the CHO for these image types, we validated the agreement between model and human observers; our results indicated that the spatial domain CHO with Gabor channels approximated human performance and can be used to evaluate detectability among the three types of images evaluated in this study.

The limitations of this study include its ex-vivo nature. Factors such as patient respiratory motion, complex background anatomy, and noncylindrical lesions may impact the detectability of enhancing lesions in the clinical environment. Image artifacts such as beam hardening, although were not observed in this phantom study, may also degrade the iodine detection results in the clinical scenarios. Second, this study only included clinical dose level and FBP reconstruction without application of AEC as it is not available on the PCD system. Since all image types were generated from exactly the same source images at the same dose, we believe the impact of AEC is minimal. Future investigations should investigate multiple dose levels, iterative reconstruction methods, and AEC (once it is available), which may also impact the iodine detection results on the PCD-CT system. Finally, future studies should also compare the multi-energy performance between different tube potential and energy threshold settings on the PCD-CT to determine optimal iodine detectability.

5. Conclusion

In this study, we demonstrated that the most sensitive image type for iodine detectability in multi-energy CT can be determined and the lowest detectable iodine concentration can be estimated using CHO. The VMIs at 50 keV had similar detectability compared to the full spectrum TLIs, both of which outperformed the iodine maps. The lowest detectable iodine concentration was 0.5 mgI/cc for an 8-mm lesion and 1.0 mgI/cc for a 4-mm lesion across different patient sizes.

Biographies

Wei Zhou received his BS degree in physics from Nanjing University in 2012 and his PhD in medical physics from the University of Texas Health Science Center, San Antonio, in 2016. He is a postdoc research fellow at Mayo Clinic. His research interests include computed tomography (CT) physics, multi-energy CT, and quantitative MRI.

Gregory Michalak received his BS degree in biomedical engineering from the Milwaukee School of Engineering in 2005 and his PhD in biomedical engineering from Louisiana Tech University in 2010. He is currently a research scientist at the Mayo Clinic. His research interests include clinical and preclinical applications of CT and the use of CT in animal models.

Jayse Weaver received his BA degree in physics and mathematics from Luther College in 2016. He is a research technologist at the Mayo Clinic.

Andrea Ferrero graduated from the Polytechnic of Turin, Italy, with a degree in physics engineering, followed by a master’s degree in medical imaging from the Royal Institute of Technology, Stockholm, Sweden, and a PhD in biomedical engineering from the University of California, Davis. Currently, he is a medical physicist in the Department of Radiology, Mayo Clinic, with a research focus on diagnostic and cone-beam CT imaging.

Hao Gong received his doctoral degree from the School of Biomedical Engineering and Sciences, Virginia Tech–Wake Forest University, in 2017. His research areas include but are not limited to x-ray CT, deep learning, and image processing/reconstruction.

Kenneth A. Fetterly received his BS degree in physics in 1995 and PhD in biophysical sciences and medical physics in 2005, both from the University of Minnesota. He is an associate professor of medicine in the Department of Cardiovascular Diseases, Mayo Clinic, Rochester, Minnesota.

Cynthia H. McCollough is a professor of medical physics and biomedical engineering at the Mayo Clinic, where she directs the CT Clinical Innovation Center. Her research interests include CT dosimetry, advanced CT technology, and new clinical applications, such as dual-energy and multispectral CT. She is an NIH-funded investigator and is active in numerous professional organizations. She is a fellow of the AAPM, ACR, and AIMBE. She received her doctorate from the University of Wisconsin in 1991.

Shuai Leng received a BS degree in engineering physics in 2001, an MS degree in engineering physics from Tsinghua University in 2003, and a PhD in medical physics from the University of Wisconsin Madison in 2008. He is a professor of medical physics at the Mayo Clinic in Rochester, Minnesota. He is an AAPM fellow. He has authored over 150 peer-reviewed articles. His research interest is in technical development and clinical application of x-ray and CT imaging.

Disclosures

Dr. Cynthia H. McCollough receives an industry grant from Siemens Healthcare, who provided the research scanner used for this study. No other authors have anything to disclose.