Performance Evaluation of a Flat-panel Detector-based Microtomography System for Small-animal Imaging

Transcription

1 Performance Evaluation of a Flat-panel Detector-based Microtomography System for Small-animal Imaging Ho Kyung Kim, Sang Chul Lee, In Kon Chun, Myung Hye Cho, Min Hyoung Cho, Member, IEEE, Soo Yeol Lee *, Member, IEEE, Koan Sik Joo, and Gyuseong Cho, Member, IEEE Abstract We have applied a flat-panel detector to an x-ray cone-beam micro computed tomography (micro-ct) for small-animal imaging. The flat-panel detector consists of an active matrix of transistors and photodiodes with a pixel pitch of 50 µm and a thallium-doped cesium iodide (CsI:Tl) scintillator as an x-ray-to-light conversion layer. The detector was fabricated with a CMOS (complementary metal-oxide-semiconductor) technology capable of a sub-micrometer design line width, hence, it has a pixel fill-factor as high as ~80%. In addition, the detector has a very fast response characteristic with an image lag being less than 0.3% in the frame integration time of 5 s. Characterization of the CMOS flat-panel detector has been performed in terms of modulation transfer function (MTF), noise power spectrum (NPS), and detective quantum efficiency (DQE). Tomographic imaging performances of the micro-ct system, such as, voxel noise, contrast-to-noise ratio (CNR), and spatial resolution, have also been evaluated by using various quantitative phantoms. Experimental results of euthanized laboratory rat imaging suggest that the micro-ct system employing a CMOS flat-panel detector can be greatly used in small-animal imaging. I. INTRODUCTION olecular imaging is expected to be rapidly developed in M near future to screen molecular events in living organisms [1]. In molecular imaging studies, small-animal imaging is very important since screening small animals to investigate human disease models or to monitor new drug effects is very essential in developing new medicines. One of the candidates for the small-animal imaging modalities is a high-resolution x-ray micro computed tomography (CT) or simply micro-ct incorporating a microfocus x-ray source [2-4]. Owing to high Manuscript received October 23, This work was supported by a grant of KISTEP, Republic of Korea (M O3A ). H. K. Kim, S. C. Lee, I. K. Chun, M. H. Cho, Min H. Cho, and S. Y. Lee are with the Graduate School of East-West Medical Science, Kyung Hee University, Yongin, Kyungki, , Republic of Korea. ( * Corresponding author: phone: ; fax: ; sylee01@khu.ac.kr) K. S. Joo is with the Department of Physics, Myongji University, Yongin, Kyungki, , Republic of Korea. G. Cho is with the Department of Nuclear and Quantum Engineering, Korea Advanced Institute of Science and Technology, Daejeon, , Republic of Korea. spatial-resolution imaging capability of a micro-ct, combination of a micro-ct with other functional imaging modalities, particularly with a nuclear medicine scanner such as a PET (positron emission tomography) or a SPECT (single photon emission CT), is in spotlight [5]. As the x-ray detector in a micro-ct system, various digital imagers have been considered. X-ray image intensifiers (XRIIs) were frequently used in early micro-ct systems and cooled charge-coupled devices (CCDs) coupled to a phosphor screen via a fiber-optic taper are now most popular [6,7]. A flat-panel x-ray detector based on the amorphous silicon (a-si) thin-film technology has been successfully applied to a volumetric x-ray CT with various potential advantages over other digital imagers such as large-area imaging capability, compactness in thickness, and no geometrical distortions and veiling glares [8,9]. To utilize the advantages, we have applied a flat-panel detector to a micro-ct for small animal imaging. The flat-panel detector has a matrix-addressed photodiode array fabricated by CMOS (complementary metal-oxide-semiconductor) process to read optical photons emitted from the overlying CsI:Tl (thallium-doped cesium iodide) scintillator [10]. We describe the design of a cone-beam micro-ct system employing a CMOS flat-panel detector and present the experimental characterization of the CMOS flat-panel detector properties for use in the micro-ct in terms of modulation transfer function (MTF), noise power spectrum (NPS), and detective quantum efficiency (DQE). Imaging performances of the micro-ct are also evaluated in terms of voxel noise, contrast-to-noise ratio (CNR), and spatial resolution. II. MATERIALS AND METHODS A. System Overview A picture of the developed micro-ct system is shown in Fig. 1. Main components of the system are a microfocus x-ray source, a rotational subject holder, and CsI:Tl coupled CMOS flat-panel detector. The microfocus x-ray source continuously irradiates a subject and the CMOS flat-panel detector acquires 2D projection data at a given frame time (or integration time). The microfocus x-ray source (L , Hamamatsu, Japan) is a sealed tube with a fixed tungsten anode facing the electron /04/$ IEEE. 2108

2 Fig. 1. A picture of the micro-ct system. The main components are a micro focus x-ray source, a rotational subject holder, and a CMOS flat-panel detector. Image reconstructions are performed by a parallel computing system consisting of four PCs, which is not shown in this picture. beam with an angle of 25 and it is equipped with a 200-µm-thick beryllium exit window. Although the dominant L-edge lines around 10 kev emitted by the tungsten target may give a better contrast in soft tissues of small animals, an additional filtration with a 1-mm-thick aluminum plate was employed in order to reduce the radiation dose inside a small-animal subject [11]. The CMOS flat-panel detector (C7942, Hamamatsu, Japan) consists of active matrix-addressed array with CMOS thin-film transistors (TFTs) and photodiode sensors having a pixel pitch of 50 µm, and a CsI:Tl scintillator plate. The fill factor is as high as 79% in spite of the small pixel size of 50 µm, which is hardly achievable with the a-si technology [12]. The CsI:Tl has a columnar structure with a diameter of about 10 µm and the thickness of 200 µm. The packing density is ~80%. For fast volume image reconstruction with the cone-beam projection data, a parallel data processing system has been realized with four PCs, each one equipped with dual CPUs (Athlon MP 2200+, AMD, USA). A host computer, one of the four PCs, controls the whole system and evenly distributes processing tasks to all the computers including itself. For the image reconstructions, we adopted the Feldkamp's cone-beam algorithm [13] filtering the projection data with the Ram-Lak filter. The micro-ct system has been typically operated at the tube voltage of 60 kvp in the large focal spot mode (50 µm) and operated in the small focal spot mode (5 µm) when the spatial resolution was measured. System magnification ratio has been usually set to about 2 (source-to-detector distance, SDD = 492 mm). Radiation doses, addressed in this study, are based on the measurement of exposures in the air at the position of axis-of-rotation (AOR) with a calibrated ion chamber (Victoreen , Innovision, USA). The frame time of the CMOS flat-panel detector to acquire a single projection data is normally 250 ms. B. Evaluation of the CMOS Flat-panel Detector Performance The performances of the CMOS flat-panel detector considered in this study are MTF, NPS, and DQE, on which the imaging system performances are largely dependent. The MTF measurement procedures are based on the previous work of Samei and Flynn [14]. In order to get the aliasing-free MTF of the detector, we obtained a projection image of a 10-µm width slit (I.I.E. GmbH, Aachen, Germany) which lies slightly slanted with respect to the vertical line (the slanted angle: ~2.2 ). With the slit image, we have calculated the MTF following the previous work. For the NPS evaluation, the central area of the detector was divided into 100 non-overlapping sections, each of which was in matrix size. The incident fluence as an input signal-to-noise ratio (SNR) was estimated to be ~ photons/mm 2 with an x-ray spectrum simulation program (SRS-78, Spectrum Processor, IPEM) [15]. In order to analyze the imaging performance of the CMOS flat-panel detector, we performed a theoretical evaluation of the NPS and DQE using a cascaded systems model [16]. The signal and noise properties of the CsI:Tl layer was evaluated with the Monte Carlo simulation taking into account the reduced density of 3.68 g/cm 3 due to the packing density. The detailed simulation procedures are in the previous study [17]. Assuming that the electronic noise in one pixel is not correlated with ones in other pixels, the additive NPS was considered as a white 2 2 noise whose magnitude is d σ add where d is the pixel pitch and σ add is the standard deviation of the additive electronic noise. This assumption provides a conservative estimation of the DQE. The image lag of the flat-panel detector, a carryover of the image signal at a frame to the next frame, is a limiting factor in developing of a micro-ct. The image lag is mainly due to incomplete charge collections by the readout electronics in a pixel. The remnant charges are read as image lag signals in the successive readout periods. The image lag is commonly observed in most digital flat-panel detectors. After a single-shot x-ray exposure, we have measured the image-lag characteristics of the detector at various frame times. The n-th image-lag signal is defined as the ratio of the residual image signal magnitude at the n-th frame after the single-shot exposure to the one during the exposure. Every averaged pixel signal value was corrected by the detector offset signal. C. Evaluation of Tomographic Imaging Performance Tomographic imaging performances of the developed micro-ct system have been evaluated with various quantitative phantoms as shown schematically in Fig. 2. A water-filled cylindrical acrylic vessel was used in evaluating the image uniformity and voxel noise characteristics. The CNR was analyzed with the contrast phantom, which consists of six low-contrast inserts with the diameter of 5 mm immersed in a water bath. The inserts are made of commercial electronic /04/$ IEEE. 2109

3 Fig. 2. Quantitative phantoms used in the tomographic performance evaluation of the micro-ct system. (a) The uniformity phantom: a water-filled cylindrical acrylic vessel for evaluating uniformity as well as noise characteristics. (b) The contrast phantom: six low-contrast inserts are immersed in water inside a cylindrical acrylic vessel. (c) The resolution phantom: an 18-µm-thick aluminum foil overlaid on an acrylic plate. The line spread response of the micro-ct system is obtained from the cross-sectional images of the foil at the slit region. density phantoms (Model , Nuclear Associates, NY, USA) such as plastic water (1.03 g/cm 3 ), nylon (1.15 g/cm 3 ), polyethylene (0.95 g/cm 3 ), acryl (1.18 g/cm 3 ), polystyrene (1.11 g/cm 3 ), and polycarbonate (1.18 g/cm 3 ). The CNR, in this study, is defined as Si S w CNR i = (1) 2 2 σi + σw where S and σ are the mean and standard deviation of the pixel values, respectively, in a given area of the images. The subscripts i and w represent the inserts and the background water region in the contrast phantom, respectively. The spatial resolving power of the micro-ct was evaluated with the phantom made of an 18-µm-thick aluminum foil attached on an acrylic plate. The evaluation method is the same as the work by Boone [18]. III. RESULTS A. Performances of the CMOS Flat-panel Detector Figure 3 summarizes the Fourier analyses of the CMOS flat-panel detector performances. All the measurements were performed with the x rays of 60 kvp and the exposure of ~17 mr. The MTF obtained without any geometric magnification gives the spatial resolution of about 7 lp/mm (or ~71 µm) at 10% as shown in Fig. 3(a). Considering the detector's inherent resolution limit, i.e., the pixel pitch (50 µm), we can notice that the spatial resolution is mainly governed by the optical photon Fig. 3. Fourier analyses of the imaging characteristics of the detector. (a) MTF, (b) NPS, and (c) DQE. Solid lines in the figures are the theoretical calculations based on the cascaded linear-systems theory. scattering within the CsI:Tl layer. As shown in Fig. 3(b), the frequency response of the normalized NPS shows the behavior of a typical indirect detector, convergence after rapid decrease of the noise power in the high frequency region. According to the cascaded linear-system theory [16], the NPS consists of two terms; the quantum-correlated noise and the additive electronic noise. The quantum-correlated noise term is dependent on the square of the system MTF. The cascaded linear-system theory assuming the white additive electronic noise power well agrees with the measurement result. It is expected that the additive electronic noise term can be neglected at higher exposures. The discrepancy between the theory and measurement in the low frequency region is probably due to the enforced reduction process of the low frequency trends in the measurement [14]. Based on the MTF and NPS measurement results, we calculated the DQE as shown in Fig. 3(c). For a comparison, the calculation based on the theory is also plotted, which agrees reasonably with the measurements. According to the Monte Carlo analysis of the CsI:Tl layer with an x-ray spectrum from a tungsten target operated at 60 kvp, the quantum absorption efficiency, A Q, is 0.47 and the Swank factor, A S, is 0.84 when the energy absorption fluctuation is only taken into account. Since the DQE at zero frequency is calculated by A Q A S = 0.40 [17], /04/$ IEEE. 2110

4 Fig. 4. The image lags (see the text for the definition) with respect to the elapsed time for three different frame times. Fig. 6. A transaxial image of the contrast phantom obtained with exposures of 33 R at the AOR. The voxel size is µm 3 and the image matrix size is Fig. 5. Voxel noise characteristics and CNR with respect to the exposure. we can notice that the overall SNR of the CMOS flat-panel detector is governed by the CsI:Tl layer. We think that the somewhat lower value of A Q in this study is due to the 1 mm thick aluminum enclosure of the detector. Figure 4 shows the frame retention property of the flat-panel detector. The first frame image lag is about 0.25% even in the very long frame time, T frame = 5 s. This is less than one-tenth of the reported data of 2-10% in the a-si flat-panel detector, which tends to have large image lags because of a considerable density of charge traps (e.g., dangling bonds in amorphous structure) [19, 20]. It can be noticed that the image lag is increased as the frame time is increased. B. Tomographic Imaging Performances In Fig. 5, the measured standard deviation in CT numbers in the uniformity phantom image are plotted as a function of exposure. The standard deviation is reciprocally proportional to the square root of exposure, which agrees well with the noise model of CT image reconstructions [21]. Voxel noise of the micro-ct system is about 50 CT numbers. In order to reduce voxel noise further without increasing radiation dose, enhancement of the detector quantum efficiency is very essential. A transaxial image of the contrast phantom obtained at ~33 R Fig. 7. The MTF curve of the micro-ct system measured with the thin aluminum foil phantom (circle). The predicted MTF curve (solid line) is also shown to account for the physical parameters affecting the system MTF such as the focal spot size, the magnification ratio, the detector MTF, and the reconstruction algorithm. The most significantly affecting parameter is the detector MTF (dotted line). is shown in Fig. 6. It is noted that the contrasts of the inserts are not solely dependent upon their physical densities. The contrasts are also dependent on the insert material compositions. For example, the plastic water contains chlorine and calcium unlike other phantom materials which basically consists of hydrogen, carbon, nitrogen or oxygen. Chlorine and calcium have much higher interaction probabilities than the others at around ~30 kev. From the acquired images we have calculated the CNR as a function of exposure. The CNR has been found to be proportional to the square root of exposure as shown in Fig. 5. The CNR at the exposure of 33 R is about 1.7 times (or ~ 3) as high as that at the exposure of 11 R. From the measurement results of the contrast phantom, we can notice that the micro-ct system can differentiate less than 2.8% contrast at the exposure of 11 R. If we assume that 1 R is equivalent to 1 rad in tissue, the exposure 11 R is approximately 1.2% of LD 50/30 (lethal dose to 50% of population after 30 days) for a mouse (900 rad) /04/$ IEEE. 2111

5 Fig. 8. Micro-CT images of a euthanized rat. (a) A coronal image of the torso and (b) a transaxial image of the abdomen region. Figure 7 shows the measured total system MTF of the micro-ct system at the magnification ratio of 2. The total system MTF is affected by various physical parameters such as the focal spot size, the magnification ratio, the reconstruction algorithm, and the detector resolving power [6]. Since these parameters are independent from each other, the total system MTF can be expressed as multiplications of the MTF of each parameter. As derived by Holdsworth et al [6], each MTF component was calculated and the total MTF was estimated as shown in Fig. 7, which agrees well with the measurement. From the results, we can notice that the system MTF is almost determined by the detector resolving power. If we assume that the limiting spatial resolution corresponds to the point when the MTF drops to 10% then we can infer that the spatial resolution limit of the micro-ct system is about 36 µm at a magnification ratio of 2, which is almost half the detector resolving power (71 µm). The overall performance of the micro-ct system is demonstrated with images of a euthanized laboratory rat as shown in Fig. 8. No contrast agent was used during imaging. Figure 8(a) and (b) show a coronal image of the torso and a transaxial image of the abdomen region, respectively. IV. DISCUSSIONS We have successfully applied the CMOS flat-panel detector to the micro-ct for small-animal imaging. The CMOS flat-panel detector has some advantages over other type flat-panel detectors, small image lag and large pixel fill factor. Especially, the pixel fill factor is an important parameter related to the sensitivity. The flat-panel detector utilizing the present a-si technology has poor fill factor because of the limited design rules. Assuming the current design rules with the TFT size of 200 µm 2, the metal line width of 10 µm, and the gap between pixels of 30 µm, the pixel fill factor is estimated as a function of the pixel pitch and plotted in Fig. 9 with the measurement data reported after 1995 [10,12,22]. For a comparison, the design results from the CMOS technology are also shown. However, there are still some barriers for the micro-ct to be used as a routine screening device in small-animal studies. First, the structural non-uniform pixel sensitivity of the detector should be overcome. In the contrast phantom image shown in Fig. 9. Estimation of the pixel fill factor as a function of the size of pixel pitch based on the conventional design rules of the a-si technology. The prediction curve well fits the published data. Data from the CMOS process are also plotted. Fig. 6, ring artifacts are apparent even though we had applied the flat field correction before the image reconstruction. In the CMOS flat-panel detector in this study, the signal charges accumulated in each pixel photodiode are read out in the column direction and they are converted into voltage signals by the eight columns of charge-sensitive amplifiers (CSAs) [10]. A columnar non-uniform sensitivity pattern has always appeared in the projection image due to the non-identical sensitivities of the CSAs. It is also observed that the non-uniformity pattern drifts to some extent as time goes. Due to the time variant nature of the non-uniformity pattern, it has been impossible to correct the non-uniformity pattern completely by the flat field correction. The remnant non-uniformity pattern after the flat field correction has made the ring artifacts in the images. Even though the ring artifacts are usually smaller than the random noise in the reconstructed images, they sometimes hamper proper reading of low contrast images. Second, the detector sensitivity has to be further improved to reduce the scan time. Typical scan times of the present study range from several minutes to tens of minutes depending on the required spatial resolution. Especially in imaging in vivo, reducing the scan time is very essential since anesthetization for a long time period is difficult for a small mouse. In the present study, we have used a detector with a protection enclosure made of 1-mm thick aluminum plate. From the Monte Carlo analysis, we have found that the aluminum enclosure gives rise to reduction of the x-ray absorption quantum efficiency by 30% in the energy band of interest. In the future studies, we plan to replace the aluminum enclosure with one made of radiation-transparent material to get better SNR performance. When a spatial resolution larger than 50 µm is desired in small animal imaging, a CMOS flat-panel detector with larger pixel pitch may be used to improve the sensitivity. We have a plan to apply a CMOS flat-panel detector with an x-ray transparent enclosure, 100-µm-pitch, and 87%-fill-factor to the micro-ct system. In this new design, sensitivity enhancement of about 60% is expected /04/$ IEEE. 2112

6 V. CONCLUSIONS A CsI:Tl coupled CMOS flat-panel detector has been successfully applied to the prototype micro-ct system pursuing small-animal imaging. From the experimental measurements and theoretical calculations, we have found that the overall performances of the detector such as MTF, NPS, and DQE are mainly determined by the first x-ray detection stage, the CsI:Tl converting layer. The effect of the additive electronic noise on the detector performance could be neglected in typical applications. In the micro-ct system, the image non-uniformity mainly comes from the structural non-uniform sensitivity pattern of the detector and the voxel noise is governed by the quantum efficiency of the detector. The spatial resolution is also governed by the detector resolving power. The minimum resolvable contrast of the micro-ct is less than 2.8% at 1.2% of the LD 50/30 level for a mouse. Further improvements of the micro-ct with optimized flat-panel detectors will make it more useful in small-animal studies. REFERENCES [1] R. Weissleder and U. Mahmood, "Molecular imaging," Radiology, vol. 219, no. 2, 2001, pp [2] M. J. Paulus, S. S. Gleason, S. J. Kennel, P. R. Hunsicker, and D. K. Johnson, "High resolution x-ray computed tomography: an emerging tool for small animal cancer research," Neoplasia, vol. 2, 2000, pp [3] E. L. Ritman, "Molecular imaging in small animals roles for micro-ct," J. Cell. Biochem. Supp. vol. 39, 2002, pp [4] D. W. Holdsworth and M. M. Thornton, "Micro-CT in small animal and specimen imaging," Trends in Biotechnology, vol. 20, no. 8 (Suppl.), 2002, pp. S34-S39. [5] A. L. Goertzen, A. K. Meadors, R. W. Silverman, and S. R. Cherry, "Simultaneous molecular and anatomical imaging of the mouse in vivo," Phys. Med. Biol., vol. 47, 2002, pp [6] D. W. Holdsworth, M. Drangova, and A. Fenster, "A high-resolution XRII-based quantitative volume CT scanner," Med. Phys., vol. 20, no. 2, 1993, pp [7] G. Wang and M. Vannier, "Micro-CT scanners for biomedical applications: an overview," Advanced Imaging, vol. 16, 2001, pp [8] D. A. Jaffray and J. H. Siewerdsen, "Cone-beam computed tomography with a flat-panel imager: initial performance characterization," Med. Phys., vol. 27, no. 6, 2000, pp [9] R. Ning, B. Chen, R. Yu, et al., "Flat panel detector-based cone-beam volume CT angiography imaging: system evaluation," IEEE Trans. Med. Imaging, vol. 19, 2000, pp [10] H. Mori, R. Kyuushima, K. Fujita, and M. Honda, "High resolution and high sensitivity CMOS panel sensors for x-ray," in Conf. Rec. CD-ROM 2001 IEEE Nucl., Sci. Symp. and Med. Imag. Conf. [11] M. J. Paulus, S. S. Gleason, H. Sari-Sarraf, et al., "High-resolution x-ray CT screening of mutant mouse models," Proc. SPIE, vol. 3921, 2000, pp [12] L. E. Antonuk, Y. El-Mohri, A. Hall, K-W. Jee, M. Maolinbay, S. C. Nassif, X. Rong, J. H. Siewerdsen, Q. Zhao, and R. L. Weisfield, "A large-area, 97 µm pitch, indirect-detection active matrix, flat-panel imager (AMFPI)," Proc. SPIE, vol. 3336, 1998, pp [13] L. A. Feldkamp, L. C. Davis, and J. W. Kress, "Practical cone-beam algorithm," J. Opt. Soc. Am. A, vol. 1, no. 6, 1984, pp [14] E. Samei and M. J. Flynn, "An experimental comparison of detector performance for direct and indirect digital radiography systemms," Med. Phys., vol. 30, no. 4, 2003, pp [15] K. Cranley, B. J. Gilmore, G. W. A. Fogarty, and L. Desponds, "Catalogue of diagnostic x-ray spectra and other data," The Institute of Physics and Engineering in Medicine, Rep. No. 78, [16] I. A. Cunningham, "Applied linear-systems theory," in Handbook of Medical Physics, vol. 1, J. Beutel, H. Kundel, and R. V. Metter, Eds. Bellingham, Washington: SPIE, 2000, pp [17] G. Cho, H. K. Kim, Y. H. Chung, D. K. Kim, H. K. Lee, T. S. Suh, and K. S. Joo, "Monte Carlo analyses of x-ray absorption, noise, and detective quantum efficiency considering therapeutic x-ray spectrum in portal imaging detector," IEEE Trans. Nucl. Sci., vol. 48, 2001, pp [18] J. M. Boone, "Determination of the presampled MTF in computed tomography," Med. Phys., vol. 28, no. 3, 2001, pp [19] J. H. Siewerdsen and D. A. Jaffray, "A ghost story: spatio-temporal response characteristics of an indirect-detection flat-panel imager," Med. Phys., vol. 26, no. 8, 1999, pp [20] J. H. Siewerdsen and D. A. Jaffray, "Cone-beam computed tomography with a flat-panel imager: effects of image lag," Med. Phys., vol. 26, no. 8, 1999, pp [21] H. H. Barrett, S. K. Gordon, and R. S. Hershel, "Statistical limitations in transaxial tomography," Comput. Biol. Med., vol. 6, 1976, pp [22] Refer to the corresponding and relevant articles in Proceedings of SPIE: Physics of Medical Imaging published after /04/$ IEEE. 2113