Nuclear Instruments and Methods in Physics Research A 569 (2006) 404 408 www.elsevier.com/locate/nima Monte-Carlo simulation for scatter correction compensation studies in SPECT imaging using GATE software package N.G. Sakellios a,, E. Karali a, D. Lazaro b, G.K. Loudos a, K.S. Nikita a a Nuclear Imaging Medical Group, Biomedical Simulations and Imaging Applications Laboratory, School of Electrical and Computer Engineering, National Technical University of Athens, 9 Iroon Polytechniou Street,GR15780 Athens, Greece b UMR 678 INSERM UPMC, CHU Pitié Salpêtière,91 Boulevard de l Hôpital, 75634 Paris Cedex 13, France Available online 23 October 2006 Abstract Simulations using Monte-Carlo packages are increasingly used in Nuclear Imaging, both for PET and SPECT applications. Either for modeling imaging systems or developing algorithms for analysis and improvement of image quantification. The aim of this work is the study of several scatter phenomena and the evaluation and improvement of two scatter correction compensation methods in pixelized scintillators. Firstly, the Dual Energy Window Subtraction Technique (DEWST) and then the Multi Energy Window and Filtering Technique (MEWFT) are investigated. Minimization Method has been used in order to achieve the best weighting co-efficient and the appropriate position and width of each low energy window. The position of them was found to be around 80 120 kev. Following scatter rejection, both quality and quantitative accuracy of the images improved. r 2006 Elsevier B.V. All rights reserved. PACS: 87.58.Pm Keywords: Image correction; Single photon emission computed tomography; Scatter; Minimization method 1. Introduction The aim of single photon emission computed tomography is to provide accurate images of the radionuclide distribution present in the object being scanned. The inclusion of scatter radiation is one of the main sources of error in SPECT imaging. Apparently, the detection of scattered events affects the contrast of the lesions and the resolution of the images. Hence, it will introduce extraneous activity in both cold and hot phantoms. During Compton scattering, photons lose part of their original energy and change their incident direction. So, energy loss can be used in order to identify scattered photons and eliminate them from detection. Several methods for the effects of scattered events have been proposed [1 8]. The difference between methods is the way of estimating the scatter contribution. For example, Corresponding author. Tel.: +30 210 7722149; fax: +30 210 7723557. E-mail address: snicol@biosim.ntua.gr (N.G. Sakellios). some methods involve events collection in one or more energy windows lower than the photopeak one, and these data can be subtracted from the true non-scattered data, before or after reconstruction. Other techniques try to estimate the shape of the scatter component within the photopeak window as a convolution of the photopeak projection with the scatter distribution function. Trying to evaluate and improve those methods, we use Monte-Carlo simulation code. The advantage of the simulation is the absolute control of any parameters involved and the wide range of analysis from the user while knowing the history of each photon created and the interactions in which it has participated. The aim of this work is to evaluate and improve two scatter correction compensation methods using simulation data obtained with GATE package. Dual Energy Window Subtraction Technique (DEWST), Multi Energy Window- Filtering Technique (MEWFT) and Convolution Subtraction Technique (CST) have been studied. 0168-9002/$ - see front matter r 2006 Elsevier B.V. All rights reserved. doi:10.1016/j.nima.2006.08.056
N.G. Sakellios et al. / Nuclear Instruments and Methods in Physics Research A 569 (2006) 404 408 405 2. Materials and methods 2.1. GATE simulation platform The GATE simulation platform is based on GEANT4 toolkit [9]. It comprises a large number of software components enabling the modeling of various aspects of nuclear medicine experiments, including the detector geometry and movement with source kinetics, the radiation detection, data output, etc. Using an extended version of GEANT4 scripting language the user can choose among these options without performing any C++ coding. New scanner geometries can be defined and simulations can be set up interactively according to the user requirements. The SPECT scanner that was simulated was a scintillation camera (3 mm thick CsI(Tl)crystal array) based on a PSPMT (Hamamatsu R2486) dedicated to small animal imaging and radiopharmaceuticals testing [10]. The prototype is equipped with a removable low-energy and highresolution collimator (2.75 cm thick, 0.4625 mm septal thickness, 1.12 mm diameter hexagonal holes). 2.2. Methods of scatter compensation The methods that were studied are as follows. 2.2.1. DEWST The DEWST proposed by Jaszczak et al. [1] involves data collection in a lower energy window, which provides an estimate for the scatter component in the photopeak window. These data are appropriately weighted and subtracted from the photopeak data. Two images are used for scatter compensation; the photopeak image P(x) and the lower energy window image L(x). C(x) is obtained as CðxÞ ¼PðxÞ klðxþ (1) where x is the vector of image pixels and k is a weighting factor. Since this method can be applied to projection data before reconstruction of tomographic slices it can be used for scatter compensation of planar images as well. In the case of pixelized scintillators a lower energy window has to be determined for each crystal cell. In a previous work [11] we assumed that each crystal element can be treated as an independent detector. We tested several lower energy windows in order to find the window that carries the most significant scatter information in a number of different imaging cases. The central channel of the optimal window was at 65% of the photopeak channel of each crystal element and its width was 75% of the photopeak channel. The optimal value for factor k was found to be equal to 0.5. These parameters provided improved results as compared with the standard single window method and they have been used in this work. 2.2.2. MEWFT The multi energy window acquisition method has been proposed by Todd-Pokropek et al. [2]. It relies on the assumption that the scatter component can be calculated as a weighted mean of data, recorded in more than one low energy windows. Furthermore, filter functions are used for the convolution of the Compton projection in each low energy window, as a result of the variability of the system transfer functions with regard to energy Sðx; FÞ ¼ X K T i C i ðx; FÞh i ðxþ (2) i where S(x,F) is the scatter photopeak projection, C i (x,f) the Compton projection, K T i ðx; FÞ the weighting coefficient and h i (x) the filter function at the ith low energy window, respectively. The derivation of the filter functions h i (x) was achieved by using the Line Spread Function (LSF). It is supposed that LSFs are good approximations of the system transfer functions in each low window used LSF s ¼ LSF i h i (3) where LSF s and LSF i are the LSF of the Compton component at the photopeak and the ith low energy window. The last step for the calculation of the filter function is the use of the Fourier theorem, where F and F 1 indicates the Fourier and Inverse Fourier transform, respectively, FðLSF s Þ¼FðLSF i ÞFðh i Þ (4) and the filter functions are h i ¼ F 1 FðLSF s Þ W (5) FðLSF i Þ where W is a window for filter function stabilization [12]. The corrected images T(x,F) finally calculated by subtracting the scatter component S(x,F) of Eq. (2) from the photopeak projections P(x,F) Tðx; FÞ ¼Pðx; FÞ Sðx; FÞ. (6) The photopeak window was set at 126 154 kev (20%) and the different low energy windows vary from 80 to 120 kev. 2.3. Minimization method The function that will be minimized is shown below. The free parameters are three. The weighting factor wf i for each one of the low energy windows I low. The position pc i of low windows and its width w i are calculated I min ðx; yþ ¼I ph ðx; yþ X wf i I sc ðx; yþ i I NULL ðx; yþ. ð7þ The I ph window is the photopeak energy window (20%). In I NULL window, GATE is storing all the photons that were detected without contributing energy in other interactions, such as Compton or Rayleigh scattering. The I min function is supposed to reach the minimum value, when the scatter correction method will finally create
406 ARTICLE IN PRESS N.G. Sakellios et al. / Nuclear Instruments and Methods in Physics Research A 569 (2006) 404 408 an image close to the NULL events recorded in GATE. The subtraction of these two images should give a value close to zero; depending on the step we have chosen for the scatter parameters wf i,pc i and ww i. of the pot. The size and shape of each region is shown in Fig. 1. This phantom was imaged for 50 s and 27 000 counts were detected. 3. Phantom experiments Flood source: The flood source is a cylinder, 4 cm inner diameter and 10 cm in height. The cylinder was filled with 99m Tc solution and was placed close to the head of the camera. The simulation was running for 50 s and 25 000 counts were detected. The experiment is used in order to estimate the scatter contribution and calculate its parameters; the weighting factors, the position and the width of each one of the low energy windows. The results will be applied to the rest of the phantoms. Breast phantom: The breast phantom consists of two hot quantities of a 99m Tc solution, both 0.5 ml in volume, placed inside a pot, 8 cm in diameter and 10 cm in height, containing 99m Tc solution. The activity ratios of the spot to the background were 10:1 (S1) and 5:1 (S2), respectively. Furthermore, the initial activity for the two spots was 2.5 10 8 and 1.25 10 8 Bq, respectively. The detector was placed at a 2.5 cm distance from the top of the pot. The hot phantom was imaged for 50 s and 22 000 counts were collected. myphantom: The phantom consists of a large cylinder with four cold regions. The inner diameter and the height of the cylinder are 4 and 8 cm, respectively. The depth of each cold region varies from 2, 4, 6 and 8 cm from the top Table 1 The results of the minimization method Compton windows Factor wf i Window width (kev) Win_1 Low 1of1 0.6 89 117 Win_2 Low 1of2 0.3 84 98 Low 2of2 0.6 94 112 Fig. 2. Images projections: the photopeak 20% window, the NULL window 20% and the correction image using two low energy windows. Fig. 1. myphantom geometry.
N.G. Sakellios et al. / Nuclear Instruments and Methods in Physics Research A 569 (2006) 404 408 407 4. Results Minimization with flood source: Using the flood source, and Eq. (7) we apply the minimization method, both for Fig. 3. myphantom Study: The Compton Low Energy Windows. 84 98 kev and 94 112 kev Low Energy Windows. one (1) and two (2) low energy windows. For both cases, the scatter parameters were calculated. In addition, the Line Spread Functions, in air and in water environment have been used. The results of the minimization method are shown in Table 1. It is clearly seen in Fig. 2 that the resulting image of the minimization method has approximated the NULL photon image, which is supposed to be the ideal one. Scatter component: The scatter distribution within the photopeak window was calculated in both phantoms using one or two low energy windows, according to the results presented in Table 1. In the following figure, the scatter component is presented for the cold phantom study (myphantom). Fig. 3 shows that the scatter contribution is preserving the geometry of the phantom. This property was observed in all different low-energy windows. Image contrast: Image projections have been drawn. It can be seen in the plots of Fig. 4 that the scatter correction method provides an improvement in contrast, in both phantoms and in both cases, using one or two low energy windows. Signal/noise ratio: Signal-to-Noise Ratio (SNR) was calculated for both studies with and without the scatter correction methods. It was defined as the net signal divided by the noise level. Count density was calculated from ROIs, 3 3 pixels large. Background count density was calculated from ROIs, 2 5 pixels large, drawn around the hot and cold lesions, respectively. The values of the SNR Fig. 4. Breast phantom: (A) photopeak image 20%, (B) corrected image. myphantom: (C) photopeak image 20%, (D) corrected image.
408 ARTICLE IN PRESS N.G. Sakellios et al. / Nuclear Instruments and Methods in Physics Research A 569 (2006) 404 408 Table 2 Signal/noise ratio Total 20% One low Two low Breast S1(10:1) 7.35 8.31 8.43 8.27 S2(5:1) 5.05 6.11 6.23 6.12 myphantom R1(2 cm) 0.66 1.10 1.32 1.35 R2(4 cm) 1.18 1.24 1.37 1.52 5. Discussion The use of one or more low-energy windows has improved image contrast. The increase in SNR and the reduction of noise followed by the rejection of scattered events can be clearly observed. Comparison between the total images and the corrected ones can be observed in Fig. 5. The scatter events subtraction in both studies provides better quality in the reconstructed images. MEWST seems to provide more robust results and this technique will be further investigated as well as the Dual Photopeak [13]. References Fig. 5. Breast Phantom: (A) photopeak image 20%, (B) corrected image. myphantom: (C) photopeak image 20%, (D) corrected image. are reported in Table 2, for the breast phantom study and for only two out of four cold regions of myphantom study. An increase in SNR was observed after the correction with both techniques. Using one or more Compton energy windows, the gain in the contrast of the image is greater than the increase of noise due to scatter rejection. [1] R.J. Jaszczak, K.L. Greer, C.E. Floyd, C.C. Harris, R.E. Coleman, J. Nucl. Med. 29 (1984) 893. [2] A.E. Todd-Pokropek, G. Clarke, R. Marsh, Preprocessing of SPECT data as a precursor for attenuation correction, in: F. Deckonic (Ed.), Information Processing in Medical Imaging, Martinus Nijhoff Publishers, Bruxelles, 1983, pp. 130 150. [3] P. Msaki, B. Axelsson, A. Israelsson, S.A. Larson, An improved scatter correction technique in SPECT using point spread scatter function, Presented at the International Conference on Medical Biology and Engineering and International Conference on Medical Physics, Espoo, Finland, 1985. [4] C.A. Lowry, D.N. Taylor, Br. J. Radiol. 59 (1986) 728. [5] C.A. Lowry, D.N. Taylor, R.S. Holt, M.J. Cooper, J.A.R. McIntosh, Single photon emission computed tomography: correction for Compton scatterd photons, in: Conference Abstracts, 43rd Annual Conference on HPA, Coventry, England, 1986. [6] C.E. Floyd Jr., R.J. Jaszczack, K.L. Greer, R.E. Coleman, J. Nucl. Med. 26 (1985) 403. [7] S.D. Egbert, R.S. May, IEEE Trans. Nucl. Sci. NS-27 (1980) 543. [8] D.J. Wagett, B.C. Wilson, Br. J. Radiol. 51 (1977) 1004. [9] G. Santin, et al., IEEE Trans. Nucl. Sci. NS-50 (2003) 1516. [10] G.K. Loudos, et al., Appl. Radiat. Isot. 58 (4) (2003) 501. [11] G. Loudos, et al., A modification of the dual energy window subtraction method for scatter compensation in pixelized scintillators for SPECT, in: IInd International Conference on Imaging Techniques in Biomedical Sciences ITBS2003, 26 30 May 2003, Athens and Milos Island, Greece. [12] R. Marsh, Scatter correction for single photon emission computed tomography images, Thesis, University College London, London, UK, 1983. [13] M.A. King, G.J. Hademenos, S.J. Glick, J. Nucl. Med. 33 (4) (1992) 605.