Photon counting spectral CT versus conventional CT: comparative evaluation for breast imaging application

Transcription

1 Physics in Medicine & Biology Photon counting spectral CT versus conventional CT: comparative evaluation for breast imaging application To cite this article: Polad M Shikhaliev and Shannon G Fritz 2011 Phys. Med. Biol Related content - Photon counting spectral CT: improved material decomposition with K-edgefiltered x-rays Polad M Shikhaliev - Energy-resolved computed tomography: first experimental results Polad M Shikhaliev - Soft tissue imaging with photon counting spectroscopic CT Polad M Shikhaliev View the article online for updates and enhancements. Recent citations - THCOBRA X-ray imaging detector operating in pure Kr L.F.N.D. Carramate et al - Material reconstruction for spectral computed tomography with detector response function Jiulong Liu and Hao Gao - Performance of in-pixel circuits for photon counting arrays (PCAs) based on polycrystalline silicon TFTs Albert K Liang et al This content was downloaded from IP address on 19/02/2018 at 08:20

2 IOP PUBLISHING Phys. Med. Biol. 56 (2011) PHYSICS IN MEDICINE AND BIOLOGY doi: / /56/7/001 Photon counting spectral CT versus conventional CT: comparative evaluation for breast imaging application Polad M Shikhaliev 1 and Shannon G Fritz Imaging Physics Laboratory, Department of Physics and Astronomy, Louisiana State University, Baton Rouge, LA 70803, USA pshikhal@lsu.edu Received 5 October 2010, in final form 26 January 2011 Published 2 March 2011 Online at stacks.iop.org/pmb/56/1905 Abstract Spectral CT systems with photon counting detectors have more advantages compared to conventional CT systems. However, clinical applications have been hampered for a long time due to the high demands of clinical systems and limitations of spectroscopic x-ray detectors. Photon counting detector technology has gained considerable improvements in the past decade, and spectral CT has become a hot topic. Several experimental spectral CT systems are under investigation. The purpose of this work was to perform the first direct, side-by-side comparison of existing spectral CT technology with a mature clinical CT system based on a conventional energy integrating detector. We have built an experimental spectral CT system whose main parameters are similar to the parameters of a clinical CT system. The system uses a spectroscopic cadmium zinc telluride (CZT) detector. The detector includes two rows of CZT pixels with 256 pixels in each row. The pixel size is 1 1mm 2, and the maximum count rate is 2 Mcounts/pixel/s. The spectral CT system has a magnification factor of 1.62 and the source to detector and source to image distances of 85 and 53 cm, respectively. The above parameters are similar to those of the clinical CT system, Siemens Sensation 16, used for comparison. The two systems were compared by imaging spatial resolution and contrast resolution phantoms made from acrylic cylinders with 14 cm diameters. The resolution phantom included Al wires with 0.3, 0.6, and 1 mm diameters, and 0.25 g cc 1 CaCO 3 contrast. The contrast phantom included contrast elements with 1.7, 5, and 15 mg cc 1 iodine, and 1.1, 3.3, and 10 mg cc 1 gadolinium. The phantoms were imaged with the two systems using 120 kvp tube voltage and 470 mr total skin exposure. The spectral CT showed CT numbers, image noise, and spatial and contrast resolutions to be similar within 10% compared to the Siemens 16 system, and provided an average of 10% higher CNR. However, the spectral CT system had a major advantage in that the iodine, gadolinium, and CaCO 3 contrasts were decomposed by 1 Author to whom any correspondence should be addressed /11/ $ Institute of Physics and Engineering in Medicine Printed in the UK 1905

3 1906 P M Shikhaliev and S G Fritz dual-energy and K-edge subtraction methods using energy selective CT data acquired in a single CT scan and fixed tube voltage. It is concluded that photon counting spectral CT technology is close to feasibility for routine clinical applications. Furthermore, it is ready for some clinical applications such as dedicated breast CT which has relatively lower demands on photon counting detectors. 1. Introduction Spectral computed tomography (CT) has gained considerable interest in recent years. Improvements in photon counting semiconductor detectors, fast readout electronics, and computational powers have translated spectral CT close to its clinical applications. Several companies have developed prototype photon counting x-ray and CT imaging detectors based on cadmium telluride (CdTe), cadmium zinc telluride (CZT), and Si semiconductors. These detectors allow counting x-ray photons separately, and CdTe and CZT detectors also allow measuring photon energies with accuracy appropriate for clinical applications. Photon counting spectral x-ray and CT imaging has been investigated in a number of theoretical (Niederlohner et al 2005, Shikhaliev 2005, 2006, 2008a, 2010, Schmidt 2009, 2010, Bornefalk and Danielsson 2010, Herrmann et al 2010, Taguchi et al 2010) and experimental (Cajipe et al 2004, Iwanczyk et al 2007, Onishiet al 2007, Schlomka et al 2008, Shikhaliev 2008b, 2009, Le et al 2010) studies. The results of these studies have shown a major improvement in technology. However, it has also been shown that there are still problems with the technology that hamper clinical use of the spectral imaging systems. Some of the remaining limitations are insufficiently high count rate capabilities of the detectors and readout electronics (Shikhaliev 2008a, Barber et al 2009, Taguchi et al 2010), intensity-dependent image artifacts associated with defects in CZT and CdTe crystals (Shikhaliev 2009), and limited energy resolution associated with hole trapping and leakage current in CZT and CdTe materials (Cajipe et al 2004, Szeles et al 2007, Schlomka et al 2008, Shikhaliev 2008b). On the other hand, for some specific clinical applications such as dedicated spectral breast CT, commercially available photon counting CZT and CdTe detectors could be appropriate (Shikhaliev 2008a). The dedicated spectral breast CT with current detector technology is feasible due to relatively low x-ray attenuation of the breast, its uniform tissue content, and round shape when imaged in pendant geometry (Shikhaliev 2008a). As photon counting detector technology now meets many of the clinical demands, it is of interest to investigate spectral CT in a direct comparison with currently used clinical CT systems. This comparison could allow for assessing where current photon counting spectral CT technology stands with respect to mature clinical CT systems based on conventional energy integrating detectors. We have built an experimental prototype of a photon counting spectral CT system based on a CZT detector, with physical parameters comparable to that of a clinical CT system. The main parameters of our system such as detector pixel size, slice thickness, magnification, and other parameters are similar to that of a clinical CT system. Resolution and contrast phantoms have been developed and imaged using spectral CT and clinical CT systems in a similar x-ray technique and exposure conditions. We report here results of the first direct, side-by-side comparison of a photon counting spectral CT system with a state-of-the-art clinical CT system based on an energy integrating detector.

4 Photon counting spectral CT versus conventional CT 1907 Figure 1. Schematic of the two-slice spectral CT system with a photon counting CZT detector. 2. Materials and methods 2.1. Spectral CT system The CT gantry of the spectral CT system has a source to detector distance of 85 cm and a source to isocenter distance of 53 cm (figure 1). The magnification factor is The sensitive field of view (FOV) at the isocenter is 15.9 cm, which determines the maximum size of the object that can be imaged. The maximum fan angle of the x-ray beam is determined by the detector FOV and equals 17. The slice thickness and pixel size at the isocenter are determined by the detector pixel size and magnification factor. The square-shaped 1 1mm 2 pixels of the CZT detector provide a 0.62 mm pixel size and a 0.62 mm slice thickness at the isocenter. The isotropic resolution element (voxel) has a 0.62 mm 3 size at the isocenter. The spectral CT system uses a photon counting CZT detector that includes two rows of pixels with 256 pixels in each row. The pixel size is 1 1mm 2 and length of the detector is 25.6 cm. The details of the detector have been described in our previous reports (Shikhaliev 2008b, 2009). The current version of the detector has a larger field of view compared to the previous detector that had a 12.8 cm field of view. Increasing the length of the detector from 12.8 to 25.6 cm allowed for building a CT setup with parameters similar to that of a clinical CT system, and also for scanning phantoms with clinically relevant sizes. The CZT crystals used in the detector have a thickness of 3 mm which allows nearly complete absorption of x-ray photons at x-ray tube voltages up to 120 kvp. Each of the pixels of the CZT detector has five independent readout channels. Each channel has an independent amplifier, discriminator, and counter. Therefore, each detector pixel can simultaneously record x-ray photons from five different sub-regions (energy bins) of the x-ray spectrum thus providing energy selective CT data acquisition. The maximum count rate of the detector is 2 Mcounts/pixel/s. CT projections can be acquired at the rate of (1 1000) ms per projection, and up to 5000 projections can be acquired in a single CT scan.

5 1908 P M Shikhaliev and S G Fritz The spectral CT system uses a rotating anode x-ray tube powered by a high frequency generator. The maximum tube voltage is 120 kvp with <5% ripple. The focal spot size is variable and can be set to 0.3 or 1 mm. The tube can operate in continuous mode for several minutes at tube currents of ma. In this study we used a 120 kvp tube voltage, a 0.3 mm focal spot size, and the tube current was adjusted to provide the required x-ray exposure. The x-ray beam was shaped by a 1 mm wide Pb collimator installed at 23 cm from the tube focal spot. This collimator shaped the x-ray beam arriving to the object. A second collimator with 2 mm width was installed at close proximity to the detector surface. The function of the second collimator was to match the width of the x-ray beam to the width of the detector rows. The x-ray beam quality was 6.75 mm Al equivalent half value layer (HVL). The current experimental setup includes a CT gantry that is stationary. The CT phantoms were placed at the isocenter and rotated around the Z-axis with adjustable angular speed using a step motor. The data acquisition process using this system has been described in details elsewhere (Shikhaliev 2008b, 2009). In the current experiments the number of projections was 800 per 360 rotation and exposure time was 50 ms per projection Siemens Sensation 16 CT system State-of-the-art clinical CT systems are produced by several companies. All of these systems generally use similar x-ray techniques, detector technologies, and data reconstruction algorithms. As a result, the main parameters of these systems are similar to each other. We have selected a Siemens Sensation 16 CT system (Siemens Medical Solutions, Malvern, PA) as a state-of-the-art clinical CT system for comparison to a spectral CT system. This system is used in a local hospital and was available to us for experimental studies. Detailed technical parameters and schematics of commercially available CT systems are considered proprietary by manufacturers. However, the main parameters of these systems can be determined from the existing literature with sufficiently high accuracy. The Sensation 16 systems (figure 2) has a source to detector distance and a source to isocenter distance of approximately cm and 55 cm, respectively, providing a magnification factor of 1.76 (Kalender 2006). The system uses an ultra-fast ceramic (UFC) scintillation detector based on gadolinium oxysulfide (Gd 2 S 2 O) (Kalender 2006). The thickness of the scintillator is approximately 1.5 mm (Fuchs et al 2000, Kalender 2006). The number of detector pixels along the detector row is 768 and the fan angle of the x-ray beam is approximately 52 (Fuchs et al 2000). The detector pixel size along the row direction is determined from the above-described parameters to be 1.19 mm. The physical slice thickness measured at the isocenter is 0.75 mm according to specifications of the Sensation 16 system. Taking into account the magnification factor of 1.76, the detector pixel size in the axial (Z) direction is determined to be 1.32 mm. The physical resolution element measured at the isocenter is not uniform and has mm 2 size in the fan beam (XY) plane and 0.75 mm size along the axial direction. The system uses a rotating anode x-ray tube with available focal spot size of 0.5 mm used in this study, and tube voltage is adjustable in the range kvp. The x-ray beam quality at 120 kvp is 7 mm Al HVL (Perisinakis et al 2007). The experiments were performed at a 120 kvp tube voltage Phantoms Two CT phantoms were fabricated from tissue-equivalent material (acrylic) to evaluate system spatial and contrast resolutions, respectively. The phantoms had a cylindrical shape and

6 Photon counting spectral CT versus conventional CT 1909 Figure 2. Schematic of the 16-slice CT system, Siemens Sensation cm diameter. The diameter of the phantoms was selected such that it fit the FOV at the isocenter. The phantoms can be considered clinically relevant because their diameter matches the average diameter of the breast used in dedicated breast CT (Boone et al 2001, Shikhaliev 2008a). The first phantom was constructed to evaluate spatial resolution (figure 3(a)). It includes a set of resolution elements made from aluminum wires with diameters of 0.3, 0.6, and 1 mm. Four cylindrical holes with 2 cm diameters were cut in the acrylic phantom. Three of these holes were filled with a liquid gel solution which was then solidified by cooling. The gel solution was prepared by mixing gelatin powder in pure water. The gel solution had nearly the same attenuation characteristics as pure water because the weight fraction of the gelatin in the solution was approximately 5% and includes primarily carbohydrates. Three sets of aluminum wires were placed in the gel matrix such that the wires are directed parallel to the axis of the cylindrical phantom. The fourth hole in the phantom was filled with a CaCO 3 powder with 10 μm grain size, suspended in the gel matrix. The CaCO 3 was used to evaluate the CNR of calcifications when they are volume averaged due to large pixel size. The weight fraction of the CaCO 3 powder in the gel matrix was 0.25 g cc 1, which is approximately 9% of the density of solid CaCO 3. The second phantom (figure 3(b)) was fabricated to compare the contrast resolution of the two CT systems when iodine and gadolinium contrast uptakes are measured in soft tissue. Eight holes with 2 cm diameters were cut in the phantom. Two holes were filled by gel solutions with no contrast. These two elements were used as water contrast for calibration purposes. Three of the remaining holes were filled by gel solutions with iodine, and the other three were filled by gel solutions of gadolinium. The iodine contrasts were prepared by mixing iodine contrast agent (Omnipaque, GE Healthcare AS, Oslo, Norway) with gel solutions. The concentrations of the iodine were 1.7, 5, and 15 mg cc 1. The iodine concentrations used in the phantoms are consistent with known iodine uptakes of 1 5 mg cc 1 observed in malignant tumors of the breast (Chang et al 1978, Jong et al 2003). The gadolinium contrast agent (Omniscan, GE Healthcare, Princeton, NJ) was mixed in a gel solution in concentrations of 1, 3.3, and 10 mg cc 1. While iodine is widely used as a contrast agent in x-ray and CT imaging, gadolinium is used primarily in MRI. In MRI, however, the amount of gadolinium injected into

7 1910 P M Shikhaliev and S G Fritz CaCO mg/cc 2 cm Al wire 0.3 mm Acrylic Water Al wire 1 mm 14 cm Al wire 0.6 mm Acrylic 1.7 mg/cc Water 2 cm I Gd 10 mg/cc 3.3 mg/cc 1.1 mg/cc 14 cm 5 mg/cc 15 mg/cc Water (a) (b) Figure 3. Schematics of the resolution phantom including aluminum wires with mm diameters and CaCO 3 contrast (a), and a contrast phantom including iodine and gadolinium contrasts with different concentrations (b). a patient is very low due to the high sensitivity of MRI to this contrast agent. Therefore, the gadolinium uptakes used in the phantom are somewhat arbitrary and were selected to provide CT signals comparable to the signals from iodine Data acquisition with adaptive filters Current x-ray and CT imaging detectors based on CZT and CdTe suffer from intensitydependent line artifacts. The line artifacts appear after flat field correction over areas of the image with large variations in x-ray intensity (Shikhaliev 2009). This problem is associated with defect distributions and space charge arrangements in the detector material, the study of which is not in the scope of the current work. To bypass the line artifacts problem we proposed using an adaptive filter, which equalizes x-ray intensity along the detector FOV (Shikhaliev 2008a). Using an adaptive filter also decreases count rate requirements to the detector. The experimental results on spectral CT imaging with an adaptive filter were previously reported (Shikhaliev 2008b). It should be noted that the adaptive filter is not optimal for imaging objects with complex shapes. It is also not optimal when objects include parts with large variations of attenuation coefficients, such as the human chest. However, for some specific applications such as breast CT, an adaptive filter can be helpful. The adaptive filter was fabricated from tissue-equivalent material (acrylic). It had a cylindrical hole with 14.4 cm diameter. The resolution and contrast phantoms were placed within the adaptive filter and there was a 2 mm gap between the adaptive filter and phantom (figure 4). The phantom was rotated during data acquisition and the adaptive filter was stationary. Two spectral CT data were acquired with identical parameters. The first data was acquired with the phantom placed in the adaptive filter. The second data was acquired when the phantom was replaced with a blank acrylic cylinder. The data with the blank cylinder was used for flat field correction of the data with the phantom. The CT projections of the phantom after flat field correction included the attenuation profiles of the contrast elements, but they

8 Photon counting spectral CT versus conventional CT 1911 Acrylic Acrylic Acrylic Acrylic X-ray (a) X-ray (b) Figure 4. Schematics of the CT phantom (a) and a blank cylinder (b) surrounded with an adaptive filter. The data acquired with a blank cylinder was used for flat field correction of the CT projections of the phantom. did not include the attenuation profiles of the base acrylic cylinder of the phantom. In order to reconstruct consistent CT images the attenuation profiles of the cylindrical base material had to be restored. For this purpose, as in previous work (Shikhaliev 2008b), digital (noiseless) CT profiles of the acrylic cylinder were generated using a 120 kvp tube voltage and a 6.75 mm Al HVL x-ray technique. The digital profiles were combined with the CT projections of the phantom after logarithmic normalization thus providing projection data that represented both contrast elements and cylindrical base material. The CT data was then reconstructed using a filtered back projection algorithm. CT acquisitions without an adaptive filter were also performed. The purpose was to compare CT images acquired with and without an adaptive filter when the x-ray technique and skin exposure are kept the same. Note that CT acquisition without an adaptive filter is not clinically feasible because it requires very high count rates over areas close to the periphery of the phantom. For CT acquisition without an adaptive filter, the blank acrylic cylinder was used similarly to that used with an adaptive filter to eliminate intensity-dependent line artifacts (figure 4). CT data was first acquired with the resolution and contrast phantoms, while the second (blank) data was acquired using an acrylic cylinder without contrast elements. The blank CT data was used for flat field correction of the phantom data. Flat field correction with the blank CT data eliminated the attenuation profiles of the cylindrical base in the phantoms, while the attenuation profiles of the contrast elements were maintained. The attenuation profile of the base cylinder was restored using the digital profiles of the acrylic cylinder as described above. Although using the blank CT data of the acrylic cylinder allows for correcting line artifacts, it does not decrease count rate requirements to the detector. The adaptive filter shown in figure 4 is appropriate for phantom studies and was used in previous (Shikhaliev 2008b) and current studies. However, it is not optimal from a dose minimization point of view because it absorbs part of the x-rays that pass through the imaged object. For the clinical prototype of the spectral CT system the adaptive filter should be installed between the x-ray source and object such that no x-ray transmitted by the object is attenuated in the filter. Two possible designs satisfying the above conditions are shown in figure 5. The first design uses a semi-circular filter fabricated from Teflon (figure 5(a)). The semi-circular Teflon filter should provide x-ray attenuation comparable to a circular acrylic filter due to the larger density of Teflon compared to acrylic. The semi-circular filter is an

9 1912 P M Shikhaliev and S G Fritz Acrylic Acrylic Teflon Acrylic X-ray (a) X-ray (b) Figure 5. Schematics of the semi-circular adaptive filter fabricated from Teflon (a), and a hollow acrylic adaptive filter. Both filters provide flat (equalized) x-ray beams at the detector plane. The Teflon filter provides slightly higher beam hardening compared to the acrylic filter. analog of the bow-tie filter used in clinical CT systems. The difference is that the bow-tie filters are mounted at the x-ray source and they provide approximate equalization of the x- ray intensity, which may be insufficient to prevent intensity-dependent line artifacts of CZT detectors. The x-ray exposure passed through a circular acrylic filter and semi-circular Teflon filter were simulated in this work to verify whether the semi-circular Teflon filter provides x-ray attenuation profiles similar to that of a circular acrylic filter. The other design of the adaptive filter uses an acrylic slab with a hollow part at the center (figure 5(b)). The hollow part of the filter has a prolonged shape rather than a circular shape. The prolonged shape of the hollow part has a pre-determined geometry that provides exact compensation of the variation of the thickness of circular object. Note that a filter with a circular hollow part cannot provide exact thickness compensation because the diameter of the hollow part should be smaller than that of the imaged object due to the demagnification effect Comparison of two CT systems System parameters. For the comparison of two CT systems, the following conditions should be met. First, the respective physical parameters of the systems should be similar. Second, the x-ray technique, CT phantoms, and radiation exposure to the phantoms used for both systems should be similar. Third, CT data acquired with the spectral CT and commercial CT systems should be reconstructed using the same algorithm and software. Once the above

10 Photon counting spectral CT versus conventional CT 1913 Table 1. Parameters of the spectral CT and clinical CT systems. Parameters Spectral CT Sensation 16 Source to detector distance (cm) Source to object distance (cm) Magnification factor Detector row pixel Detector pixel size (physical) (mm) Detector pixel size at the isocenter (mm) Slice thickness (mm) Fan angle Number of projections/ X-ray tube voltage (kvp) Focal spot size (mm) HVL (Al) at 120 kvp (mm) Total entrance exposure (mr) Reconstruction pixel (mm) Reconstruction slice (mm) Reconstruction filters #1;2;3 B20 B60 Matched filters #1 B50 conditions are satisfied, the main parameters of the two systems including spatial resolution, CT numbers, CNR, noise, and image artifacts can be compared. The parameters of the two CT systems used in this study are presented in table 1. The spatial resolution of the CT system is primarily determined by detector pixel size at the isocenter, focal spot blurring, and blurring due to the particular cutoff frequency of the reconstruction filter. For the spectral CT system the detector pixel size and focal spot blurring projected to the isocenter were 0.62 and 0.11 mm, respectively. The FWHM (full width at half maximum) spatial blurring at the isocenter determined from pixel size and focal spot blurring was 0.63 mm. Similarly, the FWHM spatial blurring for the Sensation 16 system projected to the isocenter was 0.71 mm. The FWHM spatial blurring of the Sensation 16 system is 14.5% larger than that of the spectral CT system. As mentioned above, the detailed technical information of clinical CT systems is considered proprietary, and we had no access to the data processing and reconstruction software of the Sensation 16 system. Particularly, the details of the reconstruction filters used in the Sensation 16 system were unknown. Therefore, CT data acquired with the Sensation 16 and spectral CT systems were reconstructed using different software. However, this inconsistency can be addressed. A variety of reconstruction filters with different cutoff frequencies were available for both systems. We reconstructed CT data acquired with the Sensation 16 system using reconstruction filters B20, B30, B40, B50, and B60. The CT data acquired with the spectral CT system was reconstructed using filters #1, #2, and #3. The CT numbers (HU) of the resolution elements were measured in the images acquired with the two systems using different filters, and compared. The CT numbers of the resolution elements for the two systems were nearly the same when reconstruction filters B50 and #1 were used for Sensation 16 and spectral CT systems, respectively. Because spatial blurring due to detector pixel and focal spot are similar for the two systems, it was concluded that filters B50 and #1 provide similar

11 1914 P M Shikhaliev and S G Fritz blurring components in two respective CT images. These two filters were selected as matched filters and used for all images acquired in this study Exposure conditions. The total entrance skin exposures applied to the phantoms were measured for the two CT systems using a 6 cc cylindrical ion chamber (Radcal Corp., Monrovia, CA) positioned at the isocenter. The conditions for measuring skin exposures were different for the Sensation 16 and spectral CT systems. However, total skin exposures were matched based on known system parameters. In the Sensation 16 system, the x-ray beam is shaped by a software-controlled collimator located at the x-ray tube. The thickness of the beam depends on the number of detector rows, which is set by the user prior to CT acquisition. In this study we used 16 detector rows and each detector row had a 0.75 mm width measured at the isocenter. The total width of the detector rows in the axial direction projected to isocenter was 12 mm. The total skin exposure was measured when the ion chamber was placed at the isocenter and scanned with the exactly same parameters as the phantom. In helical CT with multi-row detectors the thickness of the x-ray beam should be set larger than the total width of the detector rows. This is required because the edges of the x-ray beam are blurred due to non-zero focal spot size (penumbra effect) resulting in non-uniform beam intensities over the margins, which in turn creates image artifacts in helical CT. These non-uniform edges of the beam are rejected by a collimator installed close to the surface of the detector rows. Therefore, part of the beam past through the object is rejected and does not contribute to the image formation (Kalender 2006). The rejected fraction of the beam is characterized by a parameter called geometric efficiency, which is determined as the x-ray exposure arriving to the detector related to the total exposure in the beam. The geometric efficiency increases as the number of detector rows increases. For the Sensation 16 system the geometric efficiency is 93% when 16 central rows are used (MHRA 2004), which means only 93% of the measured exposure is used for image formation. Total skin exposure for the spectral CT system was measured with the ion chamber placed at the isocenter. In this case, the beam shaping collimator was opened sufficiently large such that the x-ray beam covered the entire surface of the ion chamber. Although the spectral CT system includes only two detector rows, the x-ray exposure measured by the ion chamber is used for image generation with 100% efficiency. This is because the exposure was measured with an open beam, and a 2 mm wide collimator was installed close to the surface of the CZT detectors with 2 mm total width of the rows. Therefore, the beam thickness was matched to the width of the pixel rows, and the beam intensity that arrived to the detector rows was flat. It should also be considered that for the Sensation 16 system, the area of the detector pixel measured at the isocenter is a factor of 1.33 larger than that for the spectral CT system ( mm 2 versus mm 2 ). Taking into account both differences in geometric efficiencies and pixel areas, the Sensation 16 system uses a factor of 1.23 larger x-ray exposure per detector pixel compared to the spectral CT system when total skin exposure is the same for the two systems. Despite this difference, in this study we used the same 470 mr total entrance skin exposure for the two systems. In the case of spectral CT acquisition with an adaptive filter the exposure to the object is decreased due to attenuation in the adaptive filter. More attenuation occurs farther from the center toward the edge of the phantom. The exposure profile arriving to the object after passing through the adaptive was simulated, and a correction factor was determined. The exposure at the surface of the adaptive filter was set such that the mean total exposure to the object remains at 470 mr.

12 Photon counting spectral CT versus conventional CT Data analysis. The data from the spectral CT was acquired in five energy bins with a single CT scan. The borders of the energy bins were 26 38, 38 50, 50 70, 70 90, and kev. The raw CT projection data was first corrected using flat field correction. For this purpose multiple flat field CT projections were acquired with no object in the beam, and averaged to suppress statistical noise. The flat field projections were then normalized and used for correction of the CT projections of the phantoms. Once the energy bin thresholds are set, as needed, five flat field data corresponding to five energy bins are generated in a single scan. The multi-energy flat field data are then used for correction of the CT projections acquired in respective energy bins. The energy selective data was used to reconstruct (1) multi-energy, (2) simple photon counting, (3) energy weighted, and (4) material decomposed CT images. The multi-energy CT images were reconstructed from CT projections acquired in different energy bins. The projection data in each energy bin was corrected, processed, and reconstructed independently. The simple photon counting images were reconstructed from the projection data that is composed from the data in five energy bins by simple stacking of all counts in respective pixels. The energy weighted images were generated similarly to simple photon counting with the difference that before stacking the projections from five energy bins each projection data was weighted with an optimal weight factor. The weight factors are task dependent and should be calculated for each type of the contrast element separately. For example, the weight factors that are optimal for iodine contrast may be not optimal for gadolinium. Therefore, the energy weighted images optimized for iodine and for gadolinium are presented separately. The multi-energy projection data was used for material decomposition based on differences in atomic numbers of different contrast elements. The material decomposition is performed by dual-energy subtraction. Theoretically, if an object that includes two types of materials with different atomic numbers is imaged using two different x-ray energies (so-called low and high energies), then it is possible to subtract these two images in such a way that one of the materials is cancelled out (Bushberg et al 2002). The low-energy CT projections were generated by combining energy bins 1 and 2, and high-energy data was generated by combining bins 3, 4, and 5. Therefore, the CT projections acquired in and kev ranges were used as low-energy and high-energy data, respectively, for decomposition of the calcium and Al wires from the background material. Dual-energy subtraction is also possible using the K-edge energy of the contrast element. In this case the low-energy and high-energy regions are selected immediately before and after the K-edge of the material. Because the linear attenuation coefficient of the contrast element changes step-wise at the K-edge, more efficient decomposition can be achieved. The K-edge of gadolinium was used to decompose gadolinium and iodine contrasts. The low-energy and high-energy CT projections were acquired for and kev energy ranges, respectively, taking into account the gadolinium K-edge position at kev. The same data was used to reconstruct the iodine-canceled gadolinium image and gadolinium-canceled iodine image. Note that K-edge subtraction was not feasible using the iodine K-edge with the current version of the CZT detector because the lowest energy threshold, 26 kev, was too close to the K-edge of iodine (33.17 kev), and it would be difficult to generate a sufficient number of counts between the threshold and K-edge energies. The CT images were reconstructed with and mm 2 pixel sizes. The slice thickness was 0.62 and 0.75 mm for spectral CT and Sensation 16 systems, respectively. The smaller image pixels were used to evaluate spatial resolution. The linear attenuation

13 1916 P M Shikhaliev and S G Fritz coefficients of the reconstructed images were converted to CT numbers (Hounsfield Units HU) as (Bushberg et al 2002) CT(x, y) = μ(x,y) μ water μ water 1000 where CT(x, y) is the CT number measured in HU, μ(x, y) is the reconstructed linear attenuation coefficient, and μ water is the known linear attenuation coefficient of water. The phantoms used in this study included contrast elements with pure water, and the values of μ water determined from reconstructed images were used. The water included a small amount of gelatin, approximately 5% by weight. The gelatin consists of primarily hydro-carbonates and some nitrogen, and its influence on μ water was neglected. The water elements in the resolution phantom also included Al wires. When calculating μ water in this case, the ROI was selected within the water element such that the Al wires were excluded. The mean values and variations of the reconstructed attenuation coefficients were measured within circular regions of interest (ROI) of identical diameter and location in the CT images acquired with the two systems. The ROI had 15 mm diameter and included N = 1264 pixels. The standard deviations of the pixel values as well as the standard deviation of the average signal within the ROI were calculated using the error propagation method (Knoll 2000). The relatively large numbers of the pixels within ROI allowed for reducing the standard deviation of an average pixel value by a factor of N as compared to the standard deviation of a pixel value. Similarly, errors added during the conversion of the attenuation coefficients to HU were calculated. The CT noise was measured as the standard deviations of the pixel values within 15 mm ROI. The noise was measured for ROI in ten different locations distributed at close proximities of the contrast elements over the outer peripheries of the contrast elements. The mean values and standard deviations of the noise were calculated based on these ten measurements. For resolution elements, the central profiles of the Al wires were drawn, and FWHM of the profile was determined. Circular ROIs with diameter equal to corresponding FWHMs were used to measure the mean CT numbers of the images of Al wires. The mean CT numbers of Al wires were further averaged over three wires with the same diameters included in each set of wires. The mean values of the CNR were calculated as a difference of the mean HU inside and outside of the contrast elements related to the mean values of the CT noise. The errors of the CNR were calculated from the measured standard deviation of CT numbers and CT noise using the error propagation method (Knoll 2000) Energy response of the detector Advantages of the photon counting spectral CT for improved CNR and material selective image acquisition depend on the energy resolution of the detector. In this work the energy spectra of the CZT detector were measured using isotope sources and x-rays. The energy resolution of CZT and CdTe detectors is affected by many factors, including hole trapping, charge sharing between the pixels, and leakage current of the detector material. Furthermore, effects of the above factors vary from pixel to pixel. Therefore, the energy spectra of the detector were measured for different pixels and compared. The measurement of the energy spectrum was described in detail elsewhere. Because CZT pixels do not have independent output for spectroscopic measurements the energy spectrum was measured by the threshold scan method. A single energy threshold was first set at 24 kev for all 512 pixels, an exposure was made, and the threshold was increased by either 2 or 4 kev steps and the next exposure was made, etc. After the energy range of interest was scanned, the energy spectrum was derived by differentiating the threshold scan data. Threshold scans were performed with 2 kev steps for an Am-241 source with 59.6 kev photon energy and with 4 kev steps for a Co-57 source

14 Photon counting spectral CT versus conventional CT 1917 (a) (b) (c) (d) (e) (f) Figure 6. Spectral CT images of the resolution phantom acquired in five energy bins (a) (e), and the final CT image reconstructed from five energy bins composed together (f). with 122 kev photon energy. The energy spectrum of the 120 kvp x-rays passed through the 14 cm acrylic phantom was measured with 4 kev steps. The experimental x-ray spectra were then compared to the known spectrum (Poludniowski et al 2009). 3. Results Figure 6 shows the spectral CT images of the resolution phantom acquired in five energy bins, along with the CT image composed from these multi-energy images. All three smallest resolution elements with 0.3 mm diameter are visualized despite the detector pixel size is not optimal. The image artifacts are seen and more artifacts are present in the lowest energy image. Figure 7 shows material selective images acquired with the spectral CT system, including tissue (water)-canceled calcium and calcium-canceled tissue images. Some residual signals from the Al wires remain in the calcium-canceled image. The possible reason for the presence of the Al signal in the tissue image is that dual-energy subtraction was performed to cancel CaCO 3, which has an effective atomic number different from that of Al. Also, the Al wire with 1 mm diameter provides a strong signal that might be difficult to cancel completely. Figure 8 compares CT images acquired with the Sensation 16 and spectral CT systems at 470 mr total skin exposure. Both images are similar with respect to spatial resolution, contrast resolution, and image noise.

15 1918 P M Shikhaliev and S G Fritz (a) (b) (c) Figure 7. Material decomposition with spectral CT: image of the resolution phantom (a), tissuecancelled image (b), and CaCO 3 cancelled image (c). Figure 9 shows magnified images of the resolution phantom acquired with the Sensation 16 and the spectral CT systems and reconstructed with 0.19 mm image pixel size. The profiles of all resolution elements are also shown. The differences between the images acquired with the two systems are nearly indistinguishable except for a slightly sharper image of the 0.3 mm Al wire in the spectral CT image, and the presence of some slight image non-uniformities and artifacts with the spectral CT system that will be discussed below. Figure 10(a) shows linear attenuation coefficients (HU) of Al wires determined from CT images acquired with the Sensation 16 system and reconstructed with filters B30, B40, B50, and B60. Corresponding linear attenuation coefficients determined from CT images acquired with the spectral CT and reconstructed with filter #1 are also shown. It is clear that reconstruction spatial blurring of the resolution elements is nearly the same for both systems when the B50 filter is used in the Sensation 16 system, and filters B50 of the Sensation 16 and filter #1 of the spectral CT system can be considered as matched filters. Figure 10(b) shows image noise measured in the CT images acquired with the Sensation 16 system with different reconstruction filters, and also noise in the spectral image reconstructed with filter #1. The two systems provide similar image noise when matched reconstruction filters are used. Therefore, Sensation 16 and spectral CT systems provide similar spatial resolution and noise when similar x-ray technique, exposure, and reconstruction filters are used. Finally, figure 11 compares the CNR of Al wires and CaCO 3 in CT images acquired with Sensation 16 and spectral CT systems. The CT images of the two systems were reconstructed using matched filters. The CNR of the spectral CT is approximately 10 25% higher than that of the Sensation 16 system. Energy-selective CT images of the contrast phantom including iodine and gadolinium are shown in figure 12. The images acquired in five energy bins and the final image composed from all five energy bins is shown. The presence of image artifacts is clearly seen. The image artifacts are primarily ring artifacts, and they are emphasized over areas of highest iodine contrast concentration. These artifacts are a clear appearance of intensity-dependent line artifacts. They were created due to the non-uniform x-ray attenuation of high concentration iodine of 15 mg cc 1, and this non-uniformity cannot be compensated using an adaptive filter. This issue will be discussed below in more detail. Figures 13(a) (f) represent images acquired with the Sensation 16 CT and the spectral CT systems, respectively. The images presented in figures 13(b) (f) were reconstructed from a single data acquisition with the spectral CT and reconstructed with different methods.

16 Photon counting spectral CT versus conventional CT 1919 (a) (b) (c) (d) Figure 8. CT images of the resolution phantom acquired with (a), (b) Sensation 16 and (c), (d) spectral CT systems at 470 mr entrance skin exposures. Respective images of the homogeneous acrylic phantom are shown in (b), (d). Figure 13(b) was reconstructed using simple photon counting data. The images in figures 13(c) and (d) were reconstructed with photon energy weighting optimized for gadolinium and iodine, respectively. The images in figures 13(e) and (f) are material decomposed images generated by dual-energy subtraction of the data acquired before and after the K-edge of gadolinium. Note that the effects of the differences between simple counting and energy weighting images are small due to several factors that will be discussed below. Figure 14 presents linear attenuation coefficients of iodine and gadolinium measured from CT images acquired with Sensation 16 and spectral CT systems. The images with the spectral CT were acquired with and without an adaptive filter. All images acquired with Sensation 16 and spectral CT systems were acquired using the same total skin exposures.

17 1920 P M Shikhaliev and S G Fritz (HU) 2000 Sensation mm wires (HU) 2000 Spectral CT mm wires coeff. atten. Lin mm wires 0.3 mm wires coeff. atten. Lin mm wires 0.3 mm wires Distance (mm) (a) Distance (mm) (b) Figure 9. Images of the resolution phantom acquired with (a) Sensation 16 and (b) spectral CT systems and reconstructed with a smaller (0.19 mm) voxel size. Corresponding image profiles of the Al wires are also shown. Figure 15 presents the CNR of iodine and gadolinium measured from CT images acquired with Sensation 16 and spectral CT systems. All images were acquired with the same total skin exposures. The CNR of iodine contrast with the spectral CT is 10 25% higher than that with the Sensation 16 system. With gadolinium a similar difference is seen except for the unexpectedly high CNR for the highest concentration of gadolinium, 10 mg cc 1, which can be explained by the unexpectedly high attenuation coefficient shown in figure 14(b) for this particular contrast. Figure 16 shows energy spectra measured with the CZT detector of the spectral CT system. The spectra were measured using isotope sources with gamma ray energies of 59.6 and 122 kev. The FWHM energy resolutions are approximately 25% at 59.6 kev and 17% at 122 kev energies. The energy spectra were measured for several consecutive pixels in the detector pixel rows to show pixel-to-pixel variation of the energy response. Figure 17 shows the energy spectra of the 120 kvp x-ray beam with 6.75 mm Al HVL passed through a 14 cm acrylic slab. The spectra were recorded for several consecutive pixels, and presented without normalization to show pixel-to-pixel variation of the energy response. More details on the

18 Photon counting spectral CT versus conventional CT 1921 Lin. atten. coeff. (HU) Spectral CT Sensation 16: B60 B50 B40 B30 CT noise (HU) Sensation 16 Spectral CT Aluminum wire diameter (mm) (a) 0 B60 B50 B40 B30 CT reconstruction filter (b) Figure 10. (a) Linear attenuation coefficients of the Al wires determined from CT images acquired with Sensation 16 and spectral CT systems. The CT data acquired with the Sensation 16 system was reconstructed using four filters with a gradually decreasing cutoff frequency that gradually increases image blurring and decreases mean attenuation coefficients of the Al wires. The filter B50 provided CT images with nearly the same degree of blurring of the Al wires and the same attenuation coefficients as CT images acquired with the spectral CT. The filter B50 was selected as the matched filter and used for reconstruction of all images acquired with the Sensation 16 system. (b) The noise in CT images acquired with the Sensation 16 system plotted against the reconstruction filter. The noise in the images acquired with the spectral CT system is shown for comparison. The filter B50 provides CT noise of the Sensation 16 system similar to the noise in the spectral CT images. 30 Sensation 16, Al wire Spectral CT, Al wire Sensation 16, CaCO 3 Spectral CT, CaCO 3 CNR Aluminum wire diameter (mm) Figure 11. Contrast to noise ratios of CaCO 3 in CT images acquired with Sensation 16 and spectral CT systems at 470 mr entrance skin exposures. limitations of the spectroscopic performance of the spectral CT system will be discussed below. Figure 18 shows the results of the simulation of x-ray transport through adaptive filters. Two types of adaptive filters representing a circular acrylic filter and a semicircular Teflon filter were used. The profiles of the x-ray exposure after passing through the filters are similar

19 1922 P M Shikhaliev and S G Fritz (a) (b) (c) (d) (e) (f) Figure 12. Spectral CT images of the contrast phantom acquired with five energy bins (a) (e) and final CT image reconstructed from five energy bins composed together (f). to each other. This comparison is important because a circular filter was used in this study, which, as noted above, is not optimal for patient dose minimization. It is important, therefore, to show that the semicircular Teflon filter provides a similar exposure profile while being more suitable for dose efficiency. 4. Discussion and conclusion The results of the first experimental comparison of a spectral CT and a clinical CT system were reported. The comparative evaluation of the CT images acquired with the two systems suggests that the spectral CT and clinical CT system Sensation 16 provide similar spatial and contrast resolutions. The results of the quantitative comparison of the two systems on spatial resolution and CNR are summarized in tables 2 and 3. The experiments were performed with the two systems with similar physical parameters, x-ray technique, and entrance skin exposure. Although entrance skin exposures for the two systems were the same, the Sensation 16 system used approximately 23% higher exposure for image formation compared to the spectral CT system. This was due to the slightly larger detector pixel size of the Sensation 16 system. For an ideal CT system a 23% increase in entrance exposure would result in 11% higher CNR. However, the spectral CT system provided an average of 10% higher CNR compared to the Sensation 16 system. The measured FWHM spatial resolutions of the two systems are also similar although the Sensation 16 system has slightly larger detector pixels.

20 Photon counting spectral CT versus conventional CT 1923 (a) (b) (c) (d) (e) (f) Figure 13. CT images of the contrast phantom acquired with the Sensation 16 system (a) and the spectral CT system (b) (f). The spectral CT images (b) (f) were reconstructed from a single scan acquired with the spectral CT system. Simple photon counting (b), energy weighting optimized for iodine (c), and energy weighting optimized for gadolinium are presented. The material decomposed images with cancelled iodine (e) and cancelled gadolinium (f) are also presented. Table 2. CT numbers and CNR of spectral CT and clinical CT systems. CT system\ CaCO 3 Iod-1 Iod-2 Iod-3 Gd-1 Gd-2 Gd-3 contrast (0.25 g cc 1 ) (1.7mgcc 1 ) (5mgcc 1 ) (15mgcc 1 ) (1.1mgcc 1 ) (3.3mgcc 1 ) (10mgcc 1 ) Sensation 16 CT numbers (HU) Spectral CT numbers (HU) HU Spect /HU Sens Sensation 16 CNR ± ± ± ± ± ± ± 0.26 Spectral CT CNR ± ± ± ± ± ± ± 0.36 CNR Spect /CNR Sens There were some variations in the CT numbers of the resolution and contrast phantoms for the two systems. In general, current CT systems have known limitations related to signal and noise non-uniformities (Kalender 2006). The degree of the non-uniformities may depend on the CT system, x-ray spectrum, size and shape of the object, type of the contrast material, magnitude of the CT numbers, etc. Some of the reasons for non-uniformities are cupping and streak artifacts due to beam hardening, cupping effect due to scatter, detector imperfections,