Suppression of metal artifacts using image-based monoenergetic DECT imaging Poster No.: C-0519 Congress: ECR 2011 Type: Scientific Paper Authors: B. Krauss, B. Schmidt, M. Sedlmair, T. Flohr; Forchheim/DE Keywords: Physics in radiology, CT, CT-Quantitative, Computer ApplicationsGeneral, Physics, Prostheses DOI: 10.1594/ecr2011/C-0519 Any information contained in this pdf file is automatically generated from digital material submitted to EPOS by third parties in the form of scientific presentations. References to any names, marks, products, or services of third parties or hypertext links to thirdparty sites or information are provided solely as a convenience to you and do not in any way constitute or imply ECR's endorsement, sponsorship or recommendation of the third party, information, product or service. ECR is not responsible for the content of these pages and does not make any representations regarding the content or accuracy of material in this file. As per copyright regulations, any unauthorised use of the material or parts thereof as well as commercial reproduction or multiple distribution by any traditional or electronically based reproduction/publication method ist strictly prohibited. You agree to defend, indemnify, and hold ECR harmless from and against any and all claims, damages, costs, and expenses, including attorneys' fees, arising from or related to your use of these pages. Please note: Links to movies, ppt slideshows and any other multimedia files are not available in the pdf version of presentations. www.myesr.org Page 1 of 15
Purpose Metal artifacts (figure below) are a common problem for interpreting CT-images of patients with metal prostheses. These artifacts can completely obscure delicate soft tissue and bone structures close to metal and are not removed by simply increasing the X-ray tube output. Dual Energy CT offers a way to reduce these artifacts and - as will be shown here commercially available image based processing can be used. Page 2 of 15
Fig.: Beam Hardening by hip prosthesis phantom scanned at 80kV. Methods and Materials Origin of metal artifacts It is known that metal artifacts are simultaneously caused by several mechanisms: a) beam hardening [1] (due to polychromatic X-ray spectrum) b) scattered radiation [2,3] (due to inefficiencies of the anti-scatter grid on the detector) c) photon starvation [1] (due to detector response for small signals) d) edge artifacts [1,4] (due to exponential absorption law and/or undersampling) e) cone beam artifacts [1] (due to data acquisition with X-ray cone beam) In the following we will mainly focus on effect a). In contrast to CT-images acquired with a monoenergetic photon beam at a synchrotron facility, X-ray spectra on commercially available CT-scanners are quite broad and range from about 30keV to the peak tube voltage times electron charge. The spectra of a commercially available CT-scanner (Somatom Definition Flash, Siemens AG, Healthcare Sector, Forchheim, Germany) have been simulated with DRASIM and are shown in the figure below. The typical Dual Energy Scan mode on this CT scanner operates at 100kV and 140kV with tin filter (Sn140kV). Page 3 of 15
Fig.: X-ray spectra of the SOMATOM Definition Flash (Siemens AG, Healthcare Sector, Forchheim, Germany) as simulated with DRASIM. As a consequence of the polychromatic spectrum, the logarithm of the measured signal attenuation for a certain X-ray beam is no longer proportional to the area density of the traversed material (which would be necessary to perform image reconstruction). However, for clinical CT-scanners the patient mainly consists of water and by assuming that it is again possible to calculate the effective area density of water or - more commonly 3 - the equivalent water thickness at standard density of 1g/cm. Unfortunately, this relation is material dependent: In the following figure the measured equivalent water thickness for 100kV and Sn140kV is plotted as a function of the material thickness of water and iodine for a true water thickness of up to 40cm and an iodine area 2 density of up to 400mg/cm (simulation with DRASIM). While the surface is almost flat for low amounts of iodine, there is an obvious curvature at higher iodine area densities. This curvature is the reason for beam hardening in the case of large amounts of iodine, bone or metal. Page 4 of 15
Fig.: Measured equivalent water thickness as a function of true water thickness and iodine area density for 100kV (left) and Sn140kV (right); simulation with DRASIM. Artifact Suppression To describe the observed curvature more quantitatively, an expansion into a power series can be done. This is shown in the next figure, where the measured equivalent water thickness dlow (dhigh) for the low (high) energy spectrum is plotted as a function of the iodine area density #I at a true water thickness of 20cm. A third order polynomial of the form 2 dlow=20cm+a1#i+a2#i +a3#i 2 3 3 dhigh=20cm+b1#i+b2#i +b3#i Page 5 of 15
provides an almost perfect fit of the measured equivalent water thickness for both voltages. A fairly good approximation is obtained if terms up to second order are kept, while the first order approximation mainly works for iodine area densities below 75mg/ 2 cm. Fig.: Measured equivalent water thickness as a function of iodine area density at a true water thickness of 20cm; a third order polynomial fit (P3) is shown as well as its truncation to two orders (P2) or one order (P1). Simulation with DRASIM. Since image reconstruction is a linear process, the resulting images Ilow (Ihigh) for the low (high) energy spectrum can be similarly expressed after reconstruction: Ilow=Iwater+a1II+a2II2+... Ihigh=Iwater+b1II+b2II2+... Page 6 of 15
where the artifact-free iodine image II is related to #I, while the higher order images II2, II3,... are pure artifact images related to powers of #I. The image II is exactly the same for low and high voltage only if the measured X-ray trajectories are the same (e.g. for sequence scans on Dual Source scanners). Otherwise this relation is approximately true, if the object has little variation along the scan axis, which may be fulfilled for metal prostheses. It is now possible to eliminate higher order terms by performing a weighted average of the low and high kv image: Iw=w Ilow+(1-w) Ihigh. In order to eliminate the second order artifact contribution the following condition is required: w=b2/(b2-a2). As image based calculation of monoenergetic images also results in weighted average images, this weight can be transformed into a monoenergetic energy as shown by the blue curve in the following figure. The red line indicates that the optimum weight for 100/Sn140kV corresponds to a monoenergetic energy of 128keV. This statement is independent of the beamhardening material, as the X-ray absorption of all relevant materials can be described by an equivalent mixture of iodine and water [5]. Page 7 of 15
Fig.: Monoenergetic energy as a function of the image weight. The equivalent energies of the 100kV and Sn140kV spectrum are represented by green lines. The weight for vanishing artifacts (iodine contrast) is shown by a red (black) line. An estimated patient diameter of 20cm is assumed. Validation with Simulation In order to show that this prediction is self-consistent, the optimum image weight for all three voltage combinations was computed and DRASIM was used to simulate images of a water phantom with 20cm diameter and two Aluminum inserts of 2cm diameter, both at a distance of 2cm from the scan axis. By placing an ROI at the phantom center between the Aluminum rods, it was verified that the beam hardening artifact is indeed removed with high precision. Validation with Scanner Page 8 of 15
As the real X-ray spectra differ slightly from the simulated ones, phantom measurements were done in addition. The phantom had to fulfill the following criteria: - body/prosthesis-equivalent shape and materials (realistic beam-hardening artifacts) - small phantom diameter (avoid photon starvation) - preferably cylindrical metal rod facing along the scan axis (minimize cone beam and edge artifacts) Finally, a CTDI head phantom (lucite, 16cm diameter) was chosen, in which lucite rods were replaced with either aluminum (10mm diameter) or stainless steel (13mm diameter). As a more realistic case, a hip prosthesis phantom was used. Scans were performed on a SOMATOM Definition Flash Scanner with the following voltage combinations: 140kV/80kV, 80kV/Sn140kV, 100/Sn140kV. While the CTDI phantom was scanned in sequence mode (default scan protocol DE_AbdomenSeq: collimation 32x0.6mm; slice thickness 5mm, kernel D30f, rotation time 0.5s, CTDIvol=15.2mGy), the hip prosthesis phantom was scanned in spiral mode (default scan protocol DE_Abdomen_LiverVNC: collimation 32x0.6mm; slice thickness 0.75mm, kernel B30f, rotation time 0.5s, pitch 0.6, CTDIvol=18.2mGy). Images were analyzed using the commercially available software syngo Dual Energy (application class "Monoenergetic"). For each phantom and combination of spectra the monoenergetic energy was determined that gave the best visual image impression. The procedure is demonstrated in the attached movie on page. Results For the simulation the following results were obtained, where errorw is the remaining error in the weighted average image: spectra weight energy errorlow errorhigh errorw 100/ Sn140kV -0.3824 128 kev -68.8 HU -19.4 HU -0.3 HU 80/Sn140kV -0.2904 144 kev -84.1 HU -19.4 HU -0.4 HU 140/80kV (no equival.) -86.4 HU -56.2 HU -3.2 HU -1.7516 Page 9 of 15
Table 1: Beam hardening artifact at the center of the phantom for the original CTimages and weighted average images. In addition the optimum image weight and corresponding monoenergetic energy is listed. The maximum beam hardening error is observed for the 80kV spectrum, the error in the 140kV image without tin filter is 35% lower. Adding the tin filter reduces the artifact by an additional 65%, while the error in the weighted average is again lower by 98%. For the voltage combination 140/80kV it is mathematically impossible to calculate an energy corresponding to the optimum weight; this is in contradiction to the phantom measurement and hints at inaccuracies of the simulated spectra. For the CTDI phantom the following optimum energies were determined, irrespective of the metal insert (Al or steel): - 100/Sn140kV:130keV - 80/Sn140kV: 140keV - 140/80kV:140keV The results are shown in the following two figures, where the rows correspond to the voltage combinations and the columns correspond to the low tube voltage, the high tube voltage and the optimized image, respectively. Image windowing was kept constant at center=40hu and width=300hu, because soft tissue CT-values are almost independent of the spectrum. Page 10 of 15
Fig.: Metal artifacts by stainless steel rods for all voltage combinations (rows), the two original images (left, middle) and the weighted average image (right). Page 11 of 15
Fig.: Metal artifacts by aluminum rods for all voltage combinations (rows), the two original images (left, middle) and the weighted average image (right). For the hip prosthesis phantom the optimum energy was 130keV (100/Sn140kV) or 120keV (80/Sn140kV); results are shown in the figure below using the same image order as before. One should note that bone is even visible for very high monoenergetic energies as it has a higher effective atomic number than water but also a higher electron density. Page 12 of 15
Fig.: Metal artifacts by hip prosthesis phantom for all voltage combinations (rows), the two original images (left, middle) and the weighted average image (right). To quantify the results for the CTDI phantom, one ROI was placed in the area between the metal rods, while a second ROI was placed at the top of the CTDI phantom. The difference between the two measurements was taken as error. Results are shown in the following table: spectrum error for steel error for aluminum 80kV 678 HU 20 HU 100kV 627 HU 12 HU 140kV 243 HU 9 HU Sn140kV 160 HU 2 HU mono 100/Sn140kV 11 HU 1 HU mono 80/Sn140kV 28 HU 3 HU Page 13 of 15
mono 140/80kV 21 HU 1 HU Table 2: Beamhardening artifact at the middle between two rods for steel and aluminum inserts in the CTDI head phantom. The steel rods are actually big enough to make higher order artifact images (II3, II4,...) relevant. In addition photon starvation and scattered radiation should have some impact. It is remarkable that by choosing the optimum monoenergetic energy useful images are obtained from this mathematically simple approach (in comparison with published metal artifact reduction methods [6]). Conclusion Monoenergetic images derived from Dual Energy CT images have substantially lower beam hardening artifacts than the original images if the optimal photon energy is selected (120...140keV for the used phantoms). In clinical cases the optimum monoenergetic energy has been found to be within a larger range (100keV... >190keV). References 1) J.F. Barrett and N. Keat, "Artifacts in CT: Recognition and Avoidance", RadioGraphics 24: 1679-1691, 2004 2) Vetter JR, Holden JE, "Correction for scattered radiation and other background signals in dual-energy computed tomography material thickness measurements", Med Phys 15(5):726-731, 1988 3) P.M. Joseph and R.D. Spital, "The effects of scatter in x-ray computed tomography", Med Phys 9(4): 464-472, 1982 4) P.M. Joseph and R.D. Spital, "The exponential edge-gradient effect in x-ray computed tomography", Phys Med Biol 26(3): 473-487, 1981 Page 14 of 15
5) R.E. Alvarez and A. Macovski, "Energy-selective Reconstruction in X-ray Computerized Tomography", Phys Med Biol 21(5): 733-744,1976 6) E. Meyer et al., "Normalized metal artifact reduction (NMAR) in computed tomography", Med Phys 37(10):5482-5493, 2010 Personal Information All authors are employees of Siemens AG, Healthcare Sector. Page 15 of 15