Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Radiation burn wikipedia , lookup
Industrial radiography wikipedia , lookup
Radiosurgery wikipedia , lookup
Backscatter X-ray wikipedia , lookup
Positron emission tomography wikipedia , lookup
Nuclear medicine wikipedia , lookup
Medical imaging wikipedia , lookup
Optimization of image acquisition techniques for dual-energy imaging of the chest N. A. Shkumat Department of Medical Biophysics, University of Toronto, Toronto, Ontario, Canada M5G 2M9 J. H. Siewerdsena兲 Department of Medical Biophysics, University of Toronto, Toronto, Ontario, Canada M5G 2M9 and Ontario Cancer Institute, Princess Margaret Hospital, Toronto, Ontario, Canada M5G 2M9 A. C. Dhanantwari and D. B. Williams Ontario Cancer Institute, Princess Margaret Hospital, Toronto, Ontario, Canada M5G 2M9 S. Richard Department of Medical Biophysics, University of Toronto, Toronto, Ontario, Canada M5G 2M9 N. S. Paul Department of Medical Imaging, Princess Margaret Hospital, Toronto, Ontario, Canada M5G 2M9 J. Yorkston and R. Van Metter Carestream Health, Inc., Rochester, New York 14650 共Received 20 April 2007; revised 6 July 2007; accepted for publication 6 August 2007; published 20 September 2007兲 Experimental and theoretical studies were conducted to determine optimal acquisition techniques for a prototype dual-energy 共DE兲 chest imaging system. Technique factors investigated included the selection of added x-ray filtration, kVp pair, and the allocation of dose between low- and highenergy projections, with total dose equal to or less than that of a conventional chest radiograph. Optima were computed to maximize lung nodule detectability as characterized by the signaldifference-to-noise ratio 共SDNR兲 in DE chest images. Optimal beam filtration was determined by cascaded systems analysis of DE image SDNR for filter selections across the periodic table 共Zfilter = 1 – 92兲, demonstrating the importance of differential filtration between low- and high-kVp projections and suggesting optimal high-kVp filters in the range Zfilter = 25– 50. For example, added filtration of ⬃2.1 mm Cu, ⬃1.2 mm Zr, ⬃0.7 mm Mo, and ⬃0.6 mm Ag to the high-kVp beam provided optimal 共and nearly equivalent兲 soft-tissue SDNR. Optimal kVp pair and dose allocation were investigated using a chest phantom presenting simulated lung nodules and ribs for thin, average, and thick body habitus. Low- and high-energy techniques ranged from 60– 90 kVp and 120– 150 kVp, respectively, with peak soft-tissue SDNR achieved at 关60/ 120兴 kVp for all patient thicknesses and all levels of imaging dose. A strong dependence on the kVp of the low-energy projection was observed. Optimal allocation of dose between low- and high-energy projections was such that ⬃30% of the total dose was delivered by the low-kVp projection, exhibiting a fairly weak dependence on kVp pair and dose. The results have guided the implementation of a prototype DE imaging system for imaging trials in early-stage lung nodule detection and diagnosis. © 2007 American Association of Physicists in Medicine. 关DOI: 10.1118/1.2777278兴 Key words: dual-energy imaging, flat-panel detector, imaging performance, image quality, imaging dose, optimization, image acquisition technique, cardiac gating, thoracic imaging, lung cancer I. INTRODUCTION As the leading cause of cancer death for both men and women, lung cancer presents an enormous burden to society.1–4 Because survival is very low for advanced stage disease 共e.g., 5-year survival of 38% – 61%, 24% – 34%, 5 % – 13% and 1% at Stages I, II, III, and IV, respectively5兲, the key to survival is early detection. Conventional chest radiography has proven inadequate in the detection of earlystage disease, missing 50% of nodules measuring 10 mm or less.6 The lack of sensitivity is attributed in large part to the superposition of anatomical structures in the projection image7—i.e., the obscuration of subtle soft-tissue nodules by 3904 Med. Phys. 34 „10…, October 2007 overlying “anatomical noise,” such as the ribs and clavicles. Low-dose CT 共LDCT兲 offers a dramatic improvement in diagnostic sensitivity;8 however, diagnostic specificity 共as well as increased cost and radiation dose兲 presents a remaining challenge, limited in part by the lack of fine material characterization.9–11 Dual-energy 共DE兲 imaging has been shown to offer a potentially promising alternative or adjuvant to accurate, earlystage detection of lung disease12–18—reducing the influence of “anatomical noise” by decomposition of the image into distinct material bases 共e.g., soft-tissue and bone兲 and offering the potential for fine material characterization 共e.g., analysis of nodule calcification兲. Conventionally, DE imag- 0094-2405/2007/34„10…/3904/12/$23.00 © 2007 Am. Assoc. Phys. Med. 3904 3905 Shkumat et al.: Dual-energy optimization for chest imaging 3905 FIG. 1. Research prototype for DE chest imaging. The system is based on a radiographic chest stand 共Kodak RVG 5100, Carestream Health, Inc., Rochester, NY兲, modified to perform cardiac-gated DE imaging. ing has been limited by suboptimal clinical implementation, a relatively high radiation dose, and the lack of a highperformance detector. The availability of flat-panel detectors 共FPDs兲 offering real-time digital readout and performance consistent with the demands of chest radiography, however, promises to remove conventional limitations, permitting high-performance DE imaging at total dose equivalent to that of a single chest radiograph. Over the last ⬃5 years, clinical DE systems based on FPDs have become available,15,19 renewing interest in a broad spectrum of DE imaging applications and raising the need for investigation of optimal acquisition techniques in the context of this new technology. Further, such renewed interest in DE imaging using FPDs extends beyond chest imaging to include real-time DE fluoroscopy20 共e.g., vascular and cardiac interventions兲 and DE computed tomography.21,22 In each case, a careful examination of optimal imaging techniques is important to maximizing DE imaging performance. Several previous studies have investigated DE technique optimization in the context of mammography,23–26 and to a lesser extent, single-shot DE imaging of the chest.27 Such work provides a valuable basis for investigation of optimal image acquisition techniques in the current context of dualshot FPD imaging. The work reported below details the optimization of DE image acquisition techniques for a chest imaging prototype under development. A combination of theoretical studies 共cascaded systems analysis28兲 and experimental measurements 共in chest phantoms兲 were performed to identify the optimal DE filtration, kVp pair, and allocation of dose between low- and high-kVp projections. The objective in each case is the maximization of soft-tissue visibility of lung nodules in DE soft-tissue images. Three phantom thicknesses corresponding to “thin,” “average,” and “thick” adult chest thicknesses were investigated, with total radiation dose equivalent to that of a single chest radiograph. Novel aspects of the work reported below include: 共i.兲 identification of optimal added filtration based on theoretical analysis of image signal and noise; 共ii.兲 selection of optimal kVp pair for FPDbased systems; and 共iii.兲 investigation of optimal dose allocation between low- and high-energy images. Taken together, Medical Physics, Vol. 34, No. 10, October 2007 these constitute a fairly complete characterization of technique factors, suitable for definition of an optimal “technique chart” for use in clinical studies. II. MATERIALS AND METHODS II.A. High-performance, cardiac-gated dual-energy imaging system A DE imaging system has recently been developed in our laboratory29,30 and translated to patient imaging trials designed to test the sensitivity and specificity of lung nodule diagnosis. The system is illustrated in Fig. 1. Based on a Kodak RVG-5100 digital radiography chest stand 共Carestream Health, Inc., Rochester, NY兲, the system includes a high-frequency, three-phase generator 共Indico 100, CPI, Georgetown, Ontario兲, a 400 kHU x-ray tube 共Varian Rad-60, Salt Lake City, UT兲, and a 10:1 antiscatter Bucky grid 共Advanced Instrument Development Inc., Melrose Park, NJ兲. Modifications to the RVG-5100 platform include: 共1兲 a collimator 共Ralco R302 ACS/A, Biassono, Italy兲 incorporating a computer-controlled filter wheel; 共2兲 a highperformance FPD 共Trixell Pixium-4600, Moirans, France兲; 共3兲 a custom-built cardiac gating system based on a fingertip pulse oximeter 共Nonin Ipod, Plymouth, MN兲; and 共4兲 the associated image acquisition and processing/display workstations. The filter wheel supports four positions for differential filtration of low- and high-kVp beams. The selection of added filtration for DE imaging is described in detail below, with results suggesting low-kVp filtration equivalent to 2.5 mm Al 共equal to the inherent filtration of the x-ray tube and collimator兲 and high-kVp filtration by an additional 2 mm Al+ 0.6 mm Ag. The added filtration in the two remaining filter wheel positions are used for conventional DR image acquisition 共1 mm Al+ 0.2 mm Cu兲 and quality assurance tests 共2 mm Al兲. The Pixium-4600 is a large-area 共⬃43⫻ 43 cm2兲 indirect-detection 共250 mg/ cm2 CsI: Tl兲 FPD composed of 3121⫻ 3121 pixels 共143 m pitch兲 with a 68% fill-factor based on an a-Si: H photodiode plus doublediode pixel readout architecture.31 To minimize the misregistration associated with cardiac motion between low- and 3906 Shkumat et al.: Dual-energy optimization for chest imaging high-kVp projections, a cardiac gating system was implemented to trigger x-ray exposure within the quiescent phase of the heart cycle.29,30 DE DE DE soft-tissue and bone-only images 共Isoft and Ibone , respectively兲 were decomposed by weighted log-subtraction: DE ln共Isoft 兲 = ln共I 兲 − ws ln共I 兲 H 共1a兲 L DE ln共Ibone 兲 = − ln共IH兲 − wb ln共IL兲, 共1b兲 where IL represents the low-energy image, IH the highenergy image, and ws and wb are weighting parameters for soft-tissue and bone, respectively. Weighted log-subtraction was employed throughout this study, chosen due to its applicability to cascaded systems modeling and computational simplicity, with weighting parameters chosen either theoretically 共from the ratio of attenuation coefficient at low- and high-kVp兲 or experimentally 共iteratively selected to cancel a given material兲. II.B. Dosimetry and imaging performance metrics II.B.1. Imparted energy Radiation dose was characterized in terms of the imparted energy: = 冕 E qE共E兲 · 共E;t兲dE, 共2兲 0 where has units of J / cm , qE共E兲 is the incident x-ray energy fluence, and 共E ; t兲 is the fraction of energy absorbed as a function of x-ray energy, E, and patient 共water兲 thickness, t.32 The imparted energy associated with typical DR chest imaging was determined by computing x-ray spectra for typical clinical techniques33 共kVp, mAs, and filtration兲 integrated over the absorption fraction for patient 共water兲 thickness approximating various body habitus and for a given source-to-patient distance. Throughout this work, unless stated otherwise, the total imparted energy for a DE acquisition 共Total = L + H兲 was equal to that of a single DR radiograph 共within ±5%兲 for the same chest thickness. For example, for an average-sized chest 共24 cm兲 Total = 0.91 J / cm2, consistent with the mean DR imaging dose reported in the Nationwide Evaluation of X-Ray Trends 共NEXT兲 Survey of 2001.33 In comparison to alternative detector technologies, the dose for a DR image is slightly lower than for a computed radiograph 共CR兲 or film-screen 共400 speed兲 acquisition. Therefore, fixing the total DE imaging dose to that of a DR radiograph represents a conservative operating point for the studies described below. To quantify differences in dose across detector types, the entrance surface dose 共ESD兲 was computed using the f-factor34 共f water兲 and backscatter fraction35 共BSF兲 averaged over the incident x-ray spectrum: 3906 the TASMIP algorithm37兲, and 共q / X兲共E兲 is the fluence per unit exposure. ESD was computed to ensure consistency with the values reported in the NEXT Survey.33 II.B.2. Dose allocation An important technique factor in DE imaging is the proportion of total dose imparted by the low- and high-kVp projections, referred to as dose allocation. For a fixed total imparted energy, Total, the dose allocation, A, is ESD = 冋冕 0 册 qo共E兲 dE BSF · f water , 共q/X兲共E兲 共3兲 where ESD has units of mGy, qo共E兲 is the incident x-ray spectrum 共computed using the SPEKTR36 implementation of Medical Physics, Vol. 34, No. 10, October 2007 共4兲 L where L and H are the energies imparted in low- and highkVp projections, respectively. Dose allocation ranges from 0 共all dose allocated to the high-kVp projection兲 to 1 共all dose allocated to the low-kVp projection兲. II.B.3. Dual-energy image signal, noise, and SDNR II.B.3.a. Dual-energy image signal. A simple metric used below to characterize DE imaging performance in the visualization of soft-tissue structures is the signal-differenceto-noise ratio 共SDNRDE兲 in a lung nodule relative to background 共lung兲. For the soft-tissue image, henceforth denoted IDE, the signal in the DE image may be written IDE = 2 E L , + H A = IH . 共IL兲ws 共5兲 The relative signal difference between the nodule and background was measured as the difference in mean signal between the two regions, normalized by the mean signal level DE = SDrel DE DE Inodule − Ibackground DE Imean , 共6兲 DE DE where Inodule and Ibackground are the mean signal in nodule and background regions of a DE image, respectively, and the mean signal is simply DE DE DE = 21 共Inodule + Ibackground 兲. Imean 共7兲 Signal difference was used as a measure of contrast in both experimental 共phantom兲 measurements and theoretical calculations. Cascaded systems analysis provides an analytical description of signal and noise propagation in an imaging system, has been applied successfully to several imaging systems,38–45 and was employed in this work to compute the DE signal and noise across a broad range of energy, dose, filtration, etc. The detector signal in either the low- or highenergy image is proportional to the linear combination of gain factors associated with the imaging chain: I=X 冉冊 qo 2 ¯ ¯ ¯ a g 1g 2g 4 , X pix 共8兲 where detector signal, I, has units of electrons per pixel. X-ray spectra were computed using SPEKTR,36 X is the exposure at the detector, qo is the mean fluence of 共Poisson- 3907 Shkumat et al.: Dual-energy optimization for chest imaging q distributed兲 incident x rays, and Xo is the mean fluence per 2 unit exposure. The sensitive area of the pixel aperture is apix 共including the fill factor兲. The gain parameters, g1 共quantum detection efficiency兲, g2 共scintillator gain兲, and g4 共coupling efficiency of secondary quanta兲 were computed as described previously.46,47 II.B.3.b. Dual-energy image noise. Noise in DE images was measured in terms of the variation in pixel values in regions of the nodule and background, with relative noise given by the mean standard deviation divided by the mean signal: DE rel = 1 DE DE 2 共nodule + background兲 , DE Imean DE nodule 共9兲 DE background where and are the standard deviations in signal in the nodule and background regions. Theoretically, the noise in DE images was computed using the noise-power spectrum 共NPS兲 for the low- and highkVp projections, combined to yield the dual-energy relative NPS as47 DE H L NPSrel = NPSrel + ws2NPSrel . 共10兲 The NPS was computed using cascaded systems analysis, including effects such as K fluorescence, scintillator blur, noise aliasing, and electronic noise.46,47 The subscript “rel” indicates relative NPS 共i.e., the absolute NPS divided by the square of the mean signal兲. The pixel variance was computed by integrating the NPS over the Nyquist region of the 2D Fourier domain, yielding the relative DE pixel noise: DE 2 H 2 L 2 兲 = 共rel 兲 + ws2共rel 兲 , 共rel 共11兲 H 2 L 2 兲 and 共rel 兲 are the relative variances in highwhere 共rel and low-kVp images, respectively, and ws is the weighting parameter for bone cancellation calculated from the ratio of the effective low- and high-kVp linear attenuation coefficients: ws = = H H ¯ bone ¯ bone xbone = L L ¯ bone ¯ bone xbone H H /Ibone,0 ln共Ibone 兲 ln共 L L Ibone /Ibone,0 兲 , 共12兲 DE SDrel DE rel . 共13兲 共14兲 Similarly for theoretical calculations, SDNRDE was computed as the ratio of relative signal difference and noise as computed by cascaded systems analysis 关Eqs. 共5兲, 共8兲, 共10兲, and 共11兲兴. Medical Physics, Vol. 34, No. 10, October 2007 II.C. Filtration The effect of differential added filtration between lowand high-kVp projections was examined as a function of the material type 共atomic number, Zfilter兲 and thickness 共sfilter兲 of added filtration. Performance was evaluated in terms of SDNR as well as patient dose and tube loading characteristics. The contrast between nodule and lung in a DE image was calculated from the difference in attenuation coefficients at low- and high-kVp: 冋 H H − lung 兲− CDE = 共nodule 册 H bone L L 共nodule − lung 兲 dnodule , L bone 共15兲 where is the effective attenuation coefficient for nodule, lung, or bone averaged over the low- or high-kVp spectra,46 and dnodule is the thickness of the nodule. This equation indicates that increasing the spectral separation improves nodule contrast, accomplished by hardening the high-kVp beam or softening the low-kVp beam 共e.g., with a K-edge filter兲. Previous studies30 indicate that effects of the low-kVp filter 共e.g., softening the beam with a ⬃0.1– 0.2 mm Ce兲 are fairly small due to subsequent hardening of the beam by the patient. The results below focus on the high-kVp filter, keeping the low-kVp filter fixed at 2.5 mm Al. Calculations were performed on the basis of a hypothetical chest model composed of 10 cm water and 10 cm inflated lung.34 Ribs were modeled as 5 mm cortical bone and pulmonary nodules as 9.5 mm polyethylene.34 The signal difference, noise, and SDNR in DE images were calculated analytically as in Sec. II B 3 as a function of the atomic number 共Zfilter = 1 – 92兲 and thickness 共sfilter = 0 – 2.5 g / cm2兲 of added filtration. For each filter selection, the exposure at the detector was fixed at 1 mR, and patient dose was calculated in terms of the imparted energy. As typical of clinical practice, therefore, Total was allowed to vary in these calculations such that the detector exposure was 1 mR. II.D. Optimal acquisition techniques II.D.1. Imaging phantom where Ibone,0 denotes the signal without bone attenuation. II.B.3.c. Dual-energy image SDNR The SDNR was measured in DE images of a chest phantom 共detailed below兲 as the ratio of relative signal difference and noise 关Eqs. 共6兲 and 共9兲, respectively兴: SDNRDE = 3907 Optimal acquisition techniques, including kVp pair and dose allocation were investigated experimentally using a chest phantom modeled after the ANSI patient-equivalent phantom,48 as illustrated in Fig. 2共a兲. Lung nodules 共9.5 mm right-circular cylinders兲 were simulated using materials ranging from micro-bubble-infused polyurethane 共−500 HU兲 to nylon 共⬃ + 75 HU兲. Ribs were simulated by Al slats 共3 and 6 mm thick兲. The correspondence between phantom thickness and patient thickness was established by measuring the transmitted exposure 共i.e., the exposure at the surface of the Bucky grid兲 for “thin,” “average,” and “thick” DR technique stations, varying the thickness of acrylic such that the transmitted exposure was ⬃1 mR in each case. The phantom 共acrylic兲 thicknesses corresponding to “thin” 共18 cm兲, “average” 共24 cm兲, and “thick” 共30 cm兲 patient thicknesses were 7.5, 10, and 12 cm acrylic, respectively. The selection of 3908 Shkumat et al.: Dual-energy optimization for chest imaging FIG. 2. 共a兲 Schematic of the chest “slab” phantom containing simulated lung nodules 共9.5 mm diameter right-circular polyethylene cylinders兲 and simulated ribs 共3 and 6 mm thick Al slats兲. Chest thickness is variable through the addition or removal of acrylic slabs. 共b兲–共d兲 Example low-energy 共LE兲, high-energy 共HE兲, and DE images of a simulated lung nodule obscured by a 3 mm rib. ROIs for SDNRDE analysis are shown as squares superimposed in 共d兲, six within the background 共cancelled bone兲 and one within the nodule. three sizes was motivated by clinical technique charts typically classifying patients in this manner. As the central lung field is the main area of interest when imaging for the detection of solitary lung nodules, the phantom thickness corresponds to this region. Therefore, the phantom thickness 共e.g., 7.5– 12 cm acrylic兲 corresponds to the effective thickness in the region of the lung for a given patient thickness 共e.g., 18– 30 cm chest兲 as measured from the spine to the sternum. II.D.2. DR technique factors and dose DR technique factors for “thin,” “average,” and “thick” patient sizes were obtained from a review of the literature and clinical technique charts at our institution. The resulting kVp and mAs are shown in Table I, along with the transmitted exposure measured behind the corresponding thickness of acrylic 共XDetector兲 and total imparted energy 共Total兲. Other factors relating to x-ray scatter and glare were held fixed at nominal selections for the prototype system—e.g., use of a 10:1 bucky grid and a fixed geometry 共⬃10 cm objectdetector air gap兲. II.D.3. Dose allocation and kVp pair Measurements of SDNRDE were performed using the phantom of Fig. 2 across a range of low-kVp 共60– 90 kVp兲, high-kVp 共120– 150 kVp兲, dose allocation 共A = ⬃ 0 – 1兲, and 3908 patient dose 共Total = 0.20– 1.73 J / cm2兲. Beam filtration was fixed at 2 mm Al ⫹ 0.6 mm Ag for high-kVp projections and the equivalent of 2.5 mm Al 共inherent兲 for low-kVp projections. To acquire DE images at various low-kVp, high-kVp, and allocation but at the same total dose, imparted energies were computed at all available kVp and mAs stations permitted by the x-ray generator. For each patient thickness and kVp pair, combinations of L共mAs兲 and H共mAs兲 were identified that yielded a given total dose, Total, within ±5%. For example, at a kVp pair of 关70/ 130兴 kVp, mAs settings of 关3.2/ 16兴 mAs give Total = 0.88 J / cm2 with an allocation of A = 0.29, whereas mAs settings of 关10/ 2兴 mAs deliver the same total dose 共Total = 0.90 J / cm2兲, but with allocation of A = 0.91. In this manner, ⬃ten stations were identified for each patient thickness, kVp pair, and total dose that resulted in allocation in the range A ⬃ 0.1– 0.9. SDNR was evaluated in soft-tissue DE images of the phantom, with the tissue weighting parameter, 共ws兲 determined automatically to minimize the signal difference between regions of simulated rib and background, ensuring optimal bone cancellation in the DE soft-tissue images. As illustrated in Fig. 2共d兲, seven ROIs 共41⫻ 41 pixels兲 were DE identified, one within the polyethylene nodule 共Inodule 兲 and DE six in the adjacent background 共Ibackground兲. Signal difference, noise, and SDNR were computed as in Eqs. 共6兲, 共11兲, and 共14兲, respectively. The mean and standard deviations in each measurement were determined from ten repeat image acquisitions. Measurements were performed for a total of 16 kVp pairs and three phantom thicknesses. In addition, measurements were performed as a function of imparted energy 共viz., 11 dose levels ranging from about one fifth to twice that of a conventional DR chest exam, 0.20– 1.73 J / cm2兲 at 关70/ 130兴 kVp. Slight variations in the dose 共constant to within ±5% for fixed patient thickness and kVp pair兲 were corrected by normalizing the measured noise by the square root of the ratio of calculated and target level of Total. Curves of SDNRDE versus dose allocation 共for a given kVp pair and Total兲 were fit using a three-parameter empirical function. Curve fits were intended to guide the reader’s eye in the results below and to identify optimal dose allocation, denoted A*, as indicated by the maximum of the fitted curve. Fits were found to give a better representation of the data under a change of variables, where a modified independent variable, A⬘, was defined as A⬘ = A / 共1 − A兲. Nonlinear TABLE I. Summary of DR technique factors for thin, average, and thick patient sizes. Patient Thickness tchest tacrylic kVp Added Filtration mAs XDetector Total Thin Average Thick 18 cm 7.5 cm 120 kVp 1 mm Al+ 0.2 mm Cu 2.0 mAs 共1.10± 0.004兲 mR 0.44 J / cm2 24 cm 10 cm 120 kVp 1 mm Al+ 0.2 mm Cu 3.2 mAs 共1.14± 0.005兲 mR 0.91 J / cm2 30 cm 12 cm 120 kVp 1 mm Al+ 0.2 mm Cu 6.4 mAs 共1.34± 0.003兲 mR 2.08 J / cm2 Medical Physics, Vol. 34, No. 10, October 2007 3909 Shkumat et al.: Dual-energy optimization for chest imaging 3909 FIG. 3. Effect of added filtration on DE imaging performance. Calculations are shown as a function of high-kVp filter material 共Zfilter兲 and thickness 共sfilter兲. 共a兲 Tissue weighting parameter, ws, for decomposition of a soft-tissue-only image. 共b兲 DE image signal difference 共nodule contrast兲 computed as in Eqs. 共5兲–共8兲. 共c兲 DE image signal-difference to noise ratio, SDNRDE. 共d兲 Tube mAs required to deliver an exposure of 1 mR to the detector 共in the high-kVp projection兲. Note the logarithmic scale. 共e兲 Imparted energy. 共f兲 SDNRDE per unit patient dose 共imparted energy兲. All results are for an average patient thickness. fitting using the Levenberg–Marquardt method was used to minimize the 2 value between fitted data and measurement. II.E. Anthropomorphic phantom An anthropomorphic chest phantom 共Model 55-8PL, Radiology Support Services, Long Beach, CA兲 was imaged as a function of dose allocation 共A = 0.06, 0.30, 0.63, and 0.91兲 at 关70/ 130兴 kVp to illustrate the effect of allocation on image quality. As in the experiments described above, the total dose delivered to the phantom was fixed, and only the dose allocation was varied. The phantom was imaged at techniques corresponding to an average patient, and images were interpreted by an expert chest radiologist 共NSP兲 on a diagnostic workstation 共dual-head, 1536⫻ 2048 pixel, 8-bit grayscale displays; AXIS III, National Display Systems, Morgan Hill, CA兲. eter, ws 共Zfilter, sfilter兲, as filter thickness and atomic number increase 共up to Zfilter ⬃ 65兲, corresponding to reduced bone contrast for harder beams. The increase in ws in the region Zfilter = 65– 80 is due to the filter K-edge falling close to the mean energy of the high-kVp beam, effectively softening the beam. A sharp decrease in ws occurs as the K-edge increases at higher atomic numbers, Zfilter ⬎ 80. The effect of filtration on SDDE is similar, as shown in Fig. 3共b兲. A harder beam results in increased spectral separation, giving increased DE signal difference at Zfilter III. RESULTS III.A. Differential beam filtration The dependence of DE imaging parameters and performance metrics on beam filtration 共described in Sec. II C兲 is illustrated in Fig. 3. In each case, calculations are shown as a function of high-kVp filter material type 共Zfilter兲 and thickness 共sfilter兲, with the low-kVp beam fixed at 70 kVp 共+2.5 mm Al added filtration兲 and a high-kVp of 130 kVp. Figure 3共a兲 shows the reduction in tissue weighting paramMedical Physics, Vol. 34, No. 10, October 2007 FIG. 4. Dual-axis plot displaying peak SDNRDE and required filter thickness 共converted to millimeters兲 as a function of filter material. The plateau in the range Zfilter ⬃ 25– 50 suggests a range of filters providing nearly equivalent peak SDNRDE. 3910 Shkumat et al.: Dual-energy optimization for chest imaging 3910 FIG. 5. DE image SDNR measured as a function of dose allocation for 共a兲 thin, 共b兲 average, and 共c兲 thick phantom thicknesses. The results plotted here correspond to a fixed high kVp 共130 kVp兲, with the low-energy technique varied from 60– 90 kVp. Curve fits are as described in the text. For each patient thickness, an optimal allocation of A ⬃ 0.3 is suggested. ⬃ 42– 63 and Zfilter ⬎ 84 at thicknesses greater than 1.5 g / cm2. The K-edge effect at Zfilter = 65– 80 significantly reduces SDDE at all thicknesses due to softening of the highkVp beam. While a harder high-kVp beam increases nodule contrast, the trade-off in image noise and SDNRDE is illustrated in Fig. 3共c兲, suggesting optimal filtration in the region Zfilter = 25– 50 共depending on filter thickness兲, and a second region of even higher SDNRDE above Zfilter ⬎ 77. The filters thus implied were considered in relation to tube loading and patient dose as in Figs. 3共d兲 and 3共e兲. Figure 3共d兲 shows the mAs required to deliver 1 mR to the detector as a function of high-kVp filtration, implying an enormous heat load for thick, high-Z filters. Such loading effectively rules out the upper-right quadrant of 共Zfilter, sfilter兲 for which mAsH ⬎ ⬃ 100 mAs. The patient dose 共imparted energy兲 for the high-kVp beam is shown in Fig. 3共e兲, showing increased dose for softer beams and suggesting a region in the range Z ⬃ 30– 65 consistent with lower patient dose. Because the calculations were performed with a fixed detector exposure of 1 mR, the patient dose varies significantly over the range of 共Zfilter, sfilter兲 investigated. The SDNRDE per unit dose 共imparted energy兲 is shown in Fig. 3共f兲. Similar to Fig. 3共c兲, the results illustrate the degradation in performance at low atomic number 共Zfilter ⬍ 20兲, the influence of the K-edge 共Zfilter = 65– 80兲, and the enhancement at very high atomic number 共Zfilter ⬎ 80兲. The effects within the optimal range Zfilter ⬃ 25– 50 implied by Fig. 3共c兲 exhibit a shift toward higher filter thickness 关due to reduced patient dose, as in Fig. 3共e兲兴. Such is consistent with the generally recognized notion that increasing filter thickness improves SNR per unit dose, but at the cost of tube loading 关Fig. 3共d兲兴. A realistic, optimal filter selection must therefore account for the trade-offs among SDNRDE, tube loading, and patient dose. For example, considering SDNRDE 关Fig. 3共c兲兴, an optimal filter selection of Zfilter = 47 共Ag兲 and sfilter = 0.5 g / cm2 共0.48 mm兲 is implied, and such is consistent with reasonable tube output 关mAsH = 17 mAs in Fig. 3共d兲兴. Considering SDNRDE per unit dose, on the other hand 关Fig. 3共f兲兴, the optimal filter thickness increases to 0.85 g / cm2 共0.81 mm Ag兲 with an increased high-kV tube output of Medical Physics, Vol. 34, No. 10, October 2007 35 mAs. A reasonable compromise among competing factors of SDNRDE, tube output, and dose, for example, is 0.63 g / cm2 共0.6 mm兲 Ag, corresponding to an acceptable tube output 共mAsH = 25 mAs兲 without significant trade-off in SDNRDE or SDNRDE / Total. FIG. 6. DE soft-tissue images of a polyethylene lung nodule. Images were acquired at a fixed kVpH = 130 kVp and various kVpL for three phantom thicknesses. Nodule contrast is highest at lower kVp 共60 kVp兲 and for the thin phantom. Reduced contrast and SDNRDE in thicker phantoms is offset in part by increased dose 共reduced noise兲 as in Table II. 3911 Shkumat et al.: Dual-energy optimization for chest imaging 3911 FIG. 7. Optimal kVp pairs. 共a兲 Tissue weighting parameter, ws, for optimal bone cancellation. 共b兲 Relative signal difference 共contrast兲, 共c兲 image noise, 共d兲 peak SDNRDE, and 共e兲 optimal dose allocation in the resulting soft-tissue DE images. All results are shown for the average patient thickness. The results of Fig. 3 imply a fairly broad range of filter materials that, given an appropriate thickness, represent equivalently “optimal” filter selections. To illustrate this point, the peak SDNRDE from Fig. 3共c兲 and the associated DE is filter thickness are shown in Fig. 4. A plateau in SDNRpeak found in the range Zfilter = 25– 50, suggesting a fairly broad range of choices for high-kVp filtration. The increase in DE at Zfilter ⬃ 80 was ruled out due to unacceptably SDNRpeak high tube loading. For filters in the range Zfilter ⬃ 45– 52, optimal filtration is achieved with filter thickness less than ⬃1 mm, aiding practical implementation. Reasonable filter selections include ⬃2.1 mm Cu 共Zfilter = 29兲, ⬃1.2 mm Zr 共Zfilter = 40兲, ⬃0.7 mm Mo 共Zfilter = 42兲, ⬃0.4 mm Pd 共Zfilter = 46兲, and ⬃0.5 mm Ag 共Zfilter = 47兲. III.B. Optimal acquisition techniques III.B.1. Dose allocation and kVp pair Varying the proportion of dose between low- and highkVp images had a substantial effect on SDNRDE. Figures 5共a兲–5共c兲 show SDNRDE as a function of A at a fixed highkVp 共130 kVp兲 for three phantom thicknesses. The four curves in each figure correspond to low kVp of 60, 70, 80, and 90 kVp, respectively, each corresponding to the same total dose level ±5%. As guided by the optimal filter results of Fig. 3, added filtration in these experiments was fixed at 2 mm Al ⫹ 0.6 mm Ag for the high-kVp projection and 2.5 mm Al 共inherent兲 for low-kVp projections. For each curve, the peak SDNRDE is found at an allocation of A ⬃ 0.3, suggesting optimal image quality when one third of the total dose is imparted by the low-kVp beam. A significant Medical Physics, Vol. 34, No. 10, October 2007 increase in SDNRDE is observed with increasing spectral separation 共i.e., reduced low-kVp兲. These results are qualitatively illustrated in Fig. 6, showing DE images of a simulated 共polyethylene兲 nodule acquired at optimal allocation 共denoted A*兲 for each of the 12 curves shown in Fig. 5. For a given phantom thickness, nodule contrast is seen to improve with reduced low-kVp. The reduction in nodule contrast for thicker phantoms is attributed to x-ray scatter, offset somewhat by a reduction in noise 共an increase in total dose兲 such that SDNRDE is similar for each phantom thickness. Measurements as in Fig. 5 were repeated for all 16 kVp pairs, summarized in Fig. 7, where each parameter plotted corresponds to the peak SDNRDE 共i.e., optimal allocation兲. As shown in Fig. 7共a兲, the weighting parameter giving optimal bone cancellation decreases with increasing high-kVp 共reduced bone contrast兲. Figure 7共b兲 illustrates the trend toward lower low- and high-kVp, suggesting maximum DE soft-tissue signal difference at 关60/ 120兴 kVp. The results suggest a trade-off between spectral separation 共i.e., increased contrast for lower low-kVp兲 and x-ray scatter 共i.e., reduced nodule contrast at higher high-kVp兲. As shown in Fig. 7共c兲, image noise was highest at 90 kVp 关likely due to decreased quantum detection efficiency 共QDE兲兴. Taken together, the effects of kVp selection on nodule contrast and noise are shown in Fig. 7共d兲 where SDNRDE is found to be highest at 关60/ 120兴 kVp, reduces sharply with increasing low-kVp 共reduced spectral separation兲, and reduces slightly with increasing high-kVp 共increased x-ray scatter兲. Finally, as shown in Fig. 7共e兲, the selection of kVp pair was found to have a small effect on the optimal dose alloca- 3912 Shkumat et al.: Dual-energy optimization for chest imaging 3912 FIG. 8. Effect of dose on DE imaging performance. 共a兲 DE image SDNR measured as a function of dose allocation for four total dose levels 共1 = 0.20, 2 = 0.45, 3 = 0.86, and 4 = 1.73 J / cm2兲. 共b兲 DE images of a polyethylene nodule acquired at condiDE for tions corresponding to SDNRpeak 1, 2, 3, and 4. 共c兲 Peak SDNRDE as a function of total imparted energy, plotted in comparison to a square-root fit. 共d兲 Optimal dose allocation measured as a function of total imparted energy with a linear fit superimposed and suggesting a slight decrease with higher total dose. All results were acquired at 关70/ 130兴 kVp. tion, with A ⬃ 0.3 presenting a smooth optimum across all conditions. Although the trends are comparable to the experimental error, higher allocation was required for reduced low and high-kVp, suggesting: 共i兲 adequate transmission through the patient required a larger proportion of dose at the lower low-kVp; and 共ii兲 increasing the high-kVp necessitates lower allocation to reduce quantum noise associated with reduced QDE at higher kVp. When low-kVp increases from 80 to 90 kVp, A increases, indicating a trade-off between imparted energy, transmitted exposure, and quantum noise. In particular, the increased noise at 90 kVp combined with the larger weighting parameter suggest an increase in the optimal dose allocation. III.B.2. Dose allocation and total dose For a fixed kVp pair and patient thickness, the behavior of ws, SDDE, DE, SDNRDE, peak SDNRDE, and optimal dose allocation was investigated as a function of the total imparted energy. SDNRDE measured as a function of A for imparted energy ranging from approximately one fifth to double that of a conventional DR radiograph are shown in Fig. 8共a兲. DE images of the polyethylene nodule acquired at optimal allocation are shown in Fig. 8共b兲. The tissue weighting parameter and signal difference did not appreciably vary with dose, although image noise decreased in proportion to the inverse square-root of dose as expected, resulting in the square-root DE shown in Fig. 8共c兲. Reduction of dependence in SDNRpeak Medical Physics, Vol. 34, No. 10, October 2007 DE rel was the driving factor for the increase of peak SDNRDE. The optimal dose allocation decreased slightly with dose as shown in Fig. 8共d兲. III.B.3. Dual-energy imaging technique chart The optimal DE imaging techniques identified above guided the formation of a technique chart for use of the DE imaging prototype in patient studies, including optimal filtration, kVp, and mAs for low- and high-kVp projections as well as dose allocation. Table II summarizes the optimal techniques along with energy imparted and entrance surface dose for three patient thicknesses. Interestingly, the optimal technique parameters 共Table II兲, including kVpL, kVpH, and A* show little dependence on patient thickness across the range investigated. III.C. Anthropomorphic phantom Figure 8 illustrates the effect of dose allocation on DE imaging performance. In each case, a magnified view of the right lung of an anthropomorphic phantom is shown for DE soft-tissue images acquired across a broad range in dose allocation. The results are generally consistent with Fig. 5共b兲, suggesting strong degradation in image quality at extreme values of allocation 共e.g., A = 0.06 and A = 0.91兲, with a fairly weak dependence in the range A ⬃ 0.2– 0.6. Interpretation by an expert thoracic radiologist indicates that the vis- 3913 Shkumat et al.: Dual-energy optimization for chest imaging 3913 FIG. 9. DE soft-tissue images of an anthropomorphic phantom acquired at four levels of dose allocation. Images were acquired at 关70/ 130兴 kVp and at equivalent total dose 共⬃0.9 J / cm2, corresponding to the energy imparted for an average patient兲. Optimal image quality is obtained at A = 0.30. A noticeable increase in image noise is evident at very low 共A = 0.06兲 and very high 共A = 0.91兲 allocation. ibility of spherical nodules in the lung is highest for the case A = 0.30, slightly reduced at A = 0.63, and significantly degraded at allocation extremes. IV. DISCUSSION AND CONCLUSIONS DE imaging can reduce the contribution of anatomical clutter within a chest radiograph, which has shown to be a significant impediment in the visualization of soft-tissue structures.7 To achieve maximum DE image quality, careful consideration of trade-offs in soft-tissue contrast and image noise must be taken into account. The analysis presented above points to DE imaging techniques that maximize softtissue visibility in DE soft-tissue images, specifically in the context of chest imaging. The results pertain to DE image decomposition by log-weighted subtraction, with future work to include investigation and management of x-ray scatter, optimization in association with various postprocessing techniques 共e.g., noise reduction兲,49 and alternative imaging tasks 共e.g., visualization of bony detail in the bone-only image兲. Previous work investigated the important role of differential filtration between low- and high-kVp beams,20,27 showing that strong filtering of the high-kVp beam is important to technique optimization. The work presented above is consistent with these findings, demonstrating further the trade-offs between increased spectral separation 共improved nodule con- trast兲 and image noise. Optimal filter material types and thickness emerge that balance the trade-offs between contrast and noise, presenting techniques that are achievable at acceptable tube loading and patient dose. A range of high-kVp filters providing comparable imaging performance is suggested—e.g., as shown in Figs. 3 and 4, metals in the range Zfilter ⬃ 40– 47 with thickness less than 1 mm. The optimal kVp pair in DE imaging has been shown to be task dependent20,46,50 with optima ranging from 关60/120兴 to 关80/ 110兴 kVp. The results above indicate an optimal softtissue imaging performance at a kVp pair of 关60/ 120兴 kVp for all patient thicknesses investigated and with total dose equivalent to that of a single chest radiograph. Low-kVp exhibited a stronger effect on SDNRDE, with 60 kVp providing improved nodule contrast and higher detector efficiency. The effect of high-kVp was less significant, suggesting competing effects among energy separation 共contrast兲, image noise, and x-ray scatter in relation to soft-tissue visibility. As expected, the tissue weighting parameter was observed to be dependent on phantom thickness—slightly larger for thinner phantoms. This effect was accounted for in the experimental studies by selecting ws independently in each DE image to automatically minimize the signal difference 共contrast兲 between simulated bone and soft-tissue background. The softtissue DE images therefore exhibit optimal bone cancellation TABLE II. Dual-energy technique chart describing optimal acquisition techniques for three patient thicknesses. Patient Thickness Thin 共18 cm chest兲 kVp mAs 共J / cm2兲 ESD 共mGy兲 A* Average 共24 cm chest兲 Thick 共30 cm chest兲 Low High Total Low High Total Low High Total 60 3.2 0.13 0.03 – 120 16 0.30 0.04 – – – 0.43 0.07 0.31 60 5 0.25 0.05 – 120 25 0.64 0.07 – – – 0.89 0.12 0.28 60 10 0.54 0.11 – 120 50 1.54 0.13 – – – 2.08 0.24 0.26 Medical Physics, Vol. 34, No. 10, October 2007 3914 Shkumat et al.: Dual-energy optimization for chest imaging 共i.e., ws selection兲 in all cases. The optimal dose allocation for this imaging task was also found to be fairly constant 共A* ⬃ 0.3兲 for all patient thicknesses investigated. The majority of patient dose is allotted to the high-kVp image to reduce noise associated with the high-kVp image. Conventionally, DE imaging has been somewhat constrained by the need for increased total imaging dose,51,52 but the optimal techniques investigated above correspond to a total dose equivalent to that of a single chest radiograph. Such studies will facilitate deployment of DE imaging systems at clinically accepted dose levels. Furthermore, the insensitivity of certain optima 共e.g., kVp pair and dose allocation兲 to patient thickness is desirable from the standpoint of simplified system implementation—i.e., the optima are applicable to a fairly broad range of patient body types. This allowed the technique factors 共mAsH and mAsL兲 to be simply interpolated for any patient thickness within the range of measurements—e.g., at 1 cm increments in the range 18– 30 cm for the technique chart developed for clinical trials with this prototype. It is important to consider how DE imaging systems such as the one described above could be implemented clinically with respect to conventional PA and lateral 共LAT兲 radiographic exams. It is unlikely that a PA DE image would replace the conventional two-view chest exam—e.g., to visualize the retro-hepatic lung. Moreover, the imaging performance and diagnostic value of a LAT DE image remains to be fully investigated. Hence, a likely short-term clinical implementation would involve a PA DE image, followed by a conventional LAT DR. However, to the extent that a truly equivalent DR image can be decomposed from the low- and high-kVp projections 共e.g., by log-weighted addition兲, both the PA and LAT views could be acquired as DE images. In such implementation, the PA and LAT views could be rendered at the clinician’s discretion as either an equivalent radiograph or a soft-tissue/bone-only decomposition. Such potential implementations should, of course, be considered with respect to further clinical research. Furthermore, it is interesting to consider the extent to which DE imaging could provide a useful modality for the numerous functions of general radiography—e.g., assessment of heart failure, line placement, foreign objects, and rib fractures. For these applications and others beyond thoracic imaging 共e.g., musculoskeletal and interventional imaging兲, the potential role of DE imaging depends on its performance in PA and LAT views, the capacity to decompose an equivalent DR image from the low- and high-energy projections, and the ability to provide a high degree of material discrimination. Such questions are the subject of future investigation and preclinical trials. ACKNOWLEDGMENTS The authors extend thanks to R. Asento and M. Haines 共Carestream Health, Inc., Rochester, NY兲 for assistance with the implementation of the clinical prototype. N. A. Shkumat was supported by scholarships from the Canadian Institutes of Health Research 共CIHR兲, the University of Toronto, and the Ontario Student Opportunity Trust Fund 共OSOTF兲. This Medical Physics, Vol. 34, No. 10, October 2007 3914 research was conducted in collaboration with Carestream, Inc. 共Rochester, New York兲 and funded in part by National Institutes of Health 共NIH兲 Grant No. R01-CA112163-01. a兲 Author to whom correspondence should be addressed. Electronic mail: [email protected]. Telephone: 416-946-4501共x5516兲; Fax: 416-946-6529. 1 Canadian Cancer Institute\National Cancer Institute of Canada, “Canadian Cancer Statistics 2006,” 共2006兲. Toronto. 2 G. Gavelli and E. Giampalma, “Sensitivity and specificity of chest X-ray screening for lung cancer: Review article,” Cancer 89, 2453–2456 共2000兲. 3 R. Etzioni, N. Urban, S. Ramsey, M. Mclntosh, S. Schwartz, B. Reid, J. Radich, G. Anderson, and L. Hartwell, “The case for early detection,” Nat. Rev. Cancer 3, 243–252 共2003兲. 4 P. B. Bach, M. J. Kelley, R. C. Tate, and D. C. McCrory, “Screening for lung cancer: A review of the current literature,” Chest 123, 72S–82S 共2003兲. 5 C. F. Mountain et al., “Lung cancer classification: The relationship of disease extent and cell type to survival in a clinical trials population,” J. Surg. Oncol. 35, 147–156 共1987兲. 6 S. Diederich and D. Wormanns, “Impact of low-dose CT on lung cancer screening,” Lung Cancer 45, S13–S19 共2004兲. 7 E. Samei, M. J. Flynn, and W. R. Eyler, “Detection of subtle lung nodules: Relative influence of quantum and anatomic noise on chest radiographs,” Radiology 213, 727–734 共1999兲. 8 C. I. Henschke et al., “Early lung cancer action project: Overall design and findings from baseline screening,” Lancet 354, 99–105 共1999兲. 9 R. MacRedmond, G. McVey, M. Lee, R. W. Costello, D. Kenny, C. Foley, and P. M. Logan, “Screening for lung cancer using low dose CT scanning: Results of 2 year follow up,” Thorax 61, 54–56 共2006兲. 10 R. M. Pieterman et al., “Preoperative staging of non-small-cell lung cancer with positron-emission tomography,” N. Engl. J. Med. 343, 254–261 共2000兲. 11 T. C. McLoud et al., “Bronchogenic carcinoma: Analysis of staging in the mediastinum with CT by correlative lymph node mapping and sampling,” Radiology 182, 319–323 共1992兲. 12 A. Macovski, R. E. Alvarez, J. L. Chan, J. P. Stonestrom, and L. M. Zatz, “Energy dependent reconstruction in x-ray computerized tomography,” Comput. Biol. Med. 6, 325–336 共1976兲. 13 L. A. Lehmann, R. E. Alvarez, A. Macovski, W. R. Brody, N. J. Pelc, S. J. Riederer, and A. L. Hall, “Generalized image combinations in dual KVP digital radiography,” Med. Phys. 8, 659–667 共1981兲. 14 F. Fischbach, T. Freund, R. Rottgen, U. Engert, R. Felix, and J. Ricke, “Dual-energy chest radiography with a flat-panel digital detector: Revealing calcified chest abnormalities,” AJR, Am. J. Roentgenol. 181, 1519– 1524 共2003兲. 15 J. Ricke, F. Fischbach, T. Freund, U. Teichgraber, E. L. Hanninen, R. Rottgen, U. Engert, H. Eichstadt, and R. Felix, “Clinical results of CsI-detector-based dual-exposure dual energy in chest radiography,” Eur. Radiol. 13, 2577–2582 共2003兲. 16 L. T. Niklason, N. M. Hickey, D. P. Chakraborty, E. A. Sabbagh, M. V. Yester, R. G. Fraser, and G. T. Barnes, “Simulated pulmonary nodules: Detection with dual-energy digital versus conventional radiography,” Radiology 160, 589–593 共1986兲. 17 F. Kelcz, F. E. Zink, W. W. Peppler, D. G. Kruger, D. L. Ergun, and C. A. Mistretta, “Conventional chest radiography vs dual-energy computed radiography in the detection and characterization of pulmonary nodules,” AJR, Am. J. Roentgenol. 162, 271–278 共1994兲. 18 W. R. Brody, G. Butt, A. Hall, and A. Macovski, “A method for selective tissue and bone visualization using dual energy scanned projection radiography,” Med. Phys. 8, 353–357 共1981兲. 19 R. E. Alvarez, J. A. Seibert, and S. K. Thompson, “Comparison of dual energy detector system performance,” Med. Phys. 31, 556–565 共2004兲. 20 J. L. Ducote, T. Xu, and S. Molloi, “Optimization of a flat-panel based real time dual-energy system for cardiac imaging,” Med. Phys. 33, 1562– 1568 共2006兲. 21 J. F. Williamson, S. Li, S. Devic, B. R. Whiting, and F. A. Lerma, “On two-parameter models of photon cross sections: Application to dualenergy CT imaging,” Med. Phys. 33, 4115–4129 共2006兲. 22 T. R. Johnson, B. Krauss, M. Sedlmair, M. Grasruck, H. Bruder, D. Morhard, C. Fink, S. Weckbach, M. Lenhard, B. Schmidt, T. Flohr, M. F. Reiser, and C. R. Becker, “Material differentiation by dual energy CT: 3915 Shkumat et al.: Dual-energy optimization for chest imaging Initial experience,” Eur. Radiol. 17, 1510–1517 共2006兲. P. C. Johns and M. J. Yaffe, “Theoretical optimization of dual-energy x-ray imaging with application to mammography,” Med. Phys. 12, 289– 296 共1985兲. 24 P. C. Johns, D. J. Drost, M. J. Yaffe, and A. Fenster, “Dual-energy mammography: Initial experimental results,” Med. Phys. 12, 297–304 共1985兲. 25 J. M. Boone, G. S. Shaber, and M. Tecotzky, “Dual-energy mammography: A detector analysis,” Med. Phys. 17, 665–675 共1990兲. 26 D. P. Chakraborty and G. T. Barnes, “An energy sensitive cassette for dual-energy mammography,” Med. Phys. l6, 7–13 共1989兲. 27 D. M. Gauntt and G. T. Barnes, “X-ray tube potential, filtration, and detector considerations in dual-energy chest radiography,” Med. Phys. 21, 203–218 共1994兲. 28 I. A. Cunningham, M. S. Westmore, and A. Fenster, “A spatial-frequency dependent quantum accounting diagram and detective quantum efficiency model of signal and noise propagation in cascaded imaging systems,” Med. Phys. 21, 417–427 共1994兲. 29 N. A. Shkumat, J. H. Siewerdsen, A. C. Dhanantwari, D. B. Williams, S. Richard, D. J. Tward, N. S. Paul, J. Yorkston, and R. Van Metter, “Development and implementation of a high-performance, dual-energy imaging system,” in: Proceedings of SPIE, Physics of Medical Imaging 共SPIE, 2007兲, Vol. 6510, pp. 651006-1–651006-12. 30 J. H. Siewerdsen, N. A. Shkumat, A. C. Dhanantwari, D. B. Williams, S. Richard, M. J. Daly, N. S. Paul, D. J. Moseley, D. A. Jaffray, J. Yorkston, and R. Van Metter, “High-performance dual-energy imaging with a flatpanel detector: imaging physics from blackboard to benchtop to bedside,” in Proceedings of SPIE Physics of Medical Imaging, SPIE, 2006, Vol. 6142, pp. 61421E-1–61421E-10. 31 C. Chaussat, J. Chabbal, T. Ducourant, V. Spinnler, G. Vieux, and R. Neyret. “New CsI/ a-Si 17⬙ ⫻ 17⬙ X-Ray flat panel detector provides superior detectivity and immediate direct digital output for general radiography systems,” in Proceedings of SPIE, Physics of Medical Imaging, SPIE, 1998 Vol. 3336, 45–56. 32 J. M. Boone, “Parametrized x-ray absorption in diagnostic radiology from Monte Carlo calculations: Implications for x-ray detector design,” Med. Phys. 19, 1467–1473 共1992兲. 33 D. C. Spelic, “Nationwide Evaluation of X-Ray Trends 共NEXT兲: Tabulation and graphical summary of 2001 survey of adult chest radiography,” 共2005兲. 34 J. H. Hubbell, “Photon mass attenuation and energy-absoption coefficient form 1 keV to 20 MeV,” J. Appl. Radiation Isotope 33, 1269–1290 共1982兲. 35 H. E. Johns, J. W. Hunt, and S. O. Fedoruk, “Surface back-scatter in the 100 kV to 400 kV range,” Br. J. Radiol. 27, 443–448 共1954兲. 36 J. H. Siewerdsen, A. M. Waese, D. J. Moseley, S. Richard, and D. A. Jaffray, “Spektr: A computational tool for x-ray spectral analysis and imaging system optimization,” Med. Phys. 31, 3057–3067 共2004兲. 37 J. M. Boone and J. A. Seibert, “An accurate method for computergenerating tungsten anode x-ray spectra from 30 to 140 kV,” Med. Phys. 24, 1661–1670 共1997兲. 38 J. H. Siewerdsen, L. E. Antonuk, Y. El Mohri, J. Yorkston, W. Huang, J. M. Boudry, and I. A. Cunningham, “Empirical and theoretical investiga23 Medical Physics, Vol. 34, No. 10, October 2007 3915 tion of the noise performance of indirect detection, active matrix flatpanel imagers 共AMFPIs兲 for diagnostic radiology,” Med. Phys. 24, 71–89 共1997兲. 39 J. H. Siewerdsen, L. E. Antonuk, Y. El Mohri, J. Yorkston, W. Huang, and I. A. Cunningham, “Signal, noise power spectrum, and detective quantum efficiency of indirect-detection flat-panel imagers for diagnostic radiology,” Med. Phys. 25, 614–628 共1998兲. 40 J. H. Siewerdsen and L. E. Antonuk, “DQE and system optimization for indirect-detection flat-panel imagers in diagnostic radiology,” in Proceedings of SPIE, Physics of Medical Imaging 共SPIE, 1990兲, Vol. 3336, pp. 546–555. 41 J. P. Bissonnette, I. A. Cunningham, D. A. Jaffray, A. Fenster, and P. Munro, “A quantum accounting and detective quantum efficiency analysis for video-based portal imaging,” Med. Phys. 24, 815–826 共1997兲. 42 A. Ganguly, S. Rudin, D. R. Bednarek, and K. R. Hoffmann, “Microangiography for neuro-vascular imaging. II. Cascade model analysis,” Med. Phys. 30, 3029–3039 共2003兲. 43 K. Doi, M. L. Giger, R. M. Nishikawa, K. R. Hoffmann, H. Macmahon, and R. A. Schmidt, “Potential usefulness of digital imaging in clinical diagnostic radiology: Computer-aided diagnosis,” J. Digit Imaging 8, 2–7 共1995兲. 44 S. Vedantham, A. Karellas, and S. Suryanarayanan, “Solid-state fluoroscopic imager for high-resolution angiography: Parallel-cascaded linear systems analysis,” Med. Phys. 31, 1258–1268 共2004兲. 45 J. G. Mainprize and M. Yaffe, “Image quality of direct conversion detectors for mammography and radiography: A theoretical comparison,” in Proceedings of SPIE, Vol. 5030, 564–574 共2003兲. 46 S. Richard and J. H. Siewerdsen, “Optimization of dual-energy imaging systems using generalized NEQ and imaging task,” Med. Phys. 34, 127– 139 共2007兲. 47 S. Richard, J. H. Siewerdsen, D. A. Jaffray, D. J. Moseley, and B. Bakhtiar, “Generalized DQE analysis of radiographic and dual-energy imaging using flat-panel detectors,” Med. Phys. 32, 1397–1413 共2005兲. 48 AAPM Report No. 31, “Standards methods for measuring diagnostic x-ray exposures,” 共1991兲. New York, American Institute of Physics. 49 R. J. Warp and J. T. Dobbins, III, “Quantitative evaluation of noise reduction strategies in dual-energy imaging,” Med. Phys. 30, 190–198 共2003兲. 50 S. Serai, J. M. Sabol, and R. Magin, “Optimization of dual-energy 共DE兲 radiographic techniques for the detection of cardiac calcifications,” Radiological Society of North America 共RSNA, 92nd Scientific Assembly, 2006兲. 51 J. M. Sabol, G. B. Avinash, F. Nicolas, B. Claus, J. Zhao, and J. T. Dobbins, “The development and characterization of a dual-energy subtraction imaging system for chest radiography based on CsI: Tl amorphous silicon flat-panel technology,” in Proceedings of SPIE, Physics of Medical Imaging 共SPIE, 2001兲, Vol. 4320, pp. 399–408. 52 T. Freund, F. Fischbach, U. Teichgraeber, E. L. Haenninen, H. Eichstaedt, R. Felix, and J. Ricke, “Effect of dose on image quality in a detectorbased dual-exposure, dual-energy system for chest radiography,” Acta Radiol. 46, 41–47 共2005兲.