You are on page 1of 6

1066

IEEE TRANSACTIONS ON NUCLEAR SCIENCE, VOL. 60, NO. 2, APRIL 2013

A Prototype Compton Camera Array for Localization and Identication of Remote Radiation Sources
Y. Kong, H. Brands, T. Glaser, Claus-M. Herbach, L. Hoy, M. Kreuels, M. Kster, G. Pausch, J. Petzoldt, C. Plettner, J. Preston, K. Roemer, F. Scherwinski, N. Teolov, J. Verity, A. Wolf, R. Lentering, and J. Stein
AbstractA functional prototype two-plane Compton camera array for localization and identication of remote radiation sources, consisting of four PVT and four NaI(Tl) scintillation detectors with PMT readout, is presented. The large-volume, mm scintillators provide a broad eld of view for scattered photons and facilitate maximum efciency at moderate cost. Each detector is equipped with a voxelSPEC, a compact electronic module that provides high voltage for the PMT, signal processing, detector stabilization, and an Ethernet communication interface. The voxelSPEC delivers list-mode event data with nanosecond precision timing over non-proprietary Ethernet and makes a system extension very easy. A software package has been developed for real-time data processing and image reconstruction. Advantages in the hard- and software allow stable, unattended operation of the camera array for many days, and provide easy-to-read information on the radiation source in real time. Measurements with the prototype array have been performed for a few standard scenarios and geometries to verify the model predications obtained by Monte-Carlo simulations. Simulations have been further performed to explore larger camera arrays with , , , and detectors. Index TermsCompton camera, nuclide identication, radiation detection, source imaging.

Fig. 1. Principle scheme of a two-plane Compton camera [4].

I. INTRODUCTION FFICIENT and accurate -ray imaging remains an objective in many applications, particularly including Homeland Security. Generally, the sensitivity of such an imaging system for remote radioactive sources is limited by active detector size and radiation background. The Compton camera technique, which combines background reduction with

Manuscript received June 01, 2012; revised August 20, 2012; accepted September 28, 2012. Date of publication December 11, 2012; date of current version April 10, 2013. This work was supported by the U.S. Defense Threat Reduction Agency (DTRA) under contract HDTRA1-11-C-0002. Approved for public release, and distribution is unlimited. Y. Kong, C.-M. Herbach, M. Kreuels, M. Kster, G. Pausch, J. Petzoldt, C. Plettner, K. Roemer, F. Scherwinski, N. Teolov, A. Wolf, and R. Lentering are with FLIR Radiation GmbH, 42653 Solingen, Germany (e-mail: yong.kong@ir.com; claus.herbach@ir.com; martin.kreuels@ir.com; mathias.kuester@ir.com; guntram.pausch@ir.com; joe.petzoldt@gmx.de; cristina.plettner@ir.com; katja.roemer@ir.com; nikolai.teolov@ir.com; falko.scherwinski@ir.com; andreas.wolf@ir.com; ralf.lentering@ir.com). H. Brands, T. Glaser, L. Hoy, J. Preston, and J. Verity are with FLIR Radiation Inc., Oak Ridge, TN 37830 USA (e-mail: hartmut.brands@ir.com; thomas. glaser@ir.com; les.hoy@ir.com; jeff.preston@ir.com; jeffrey.verity@ir. com). J. Stein is with FLIR Radiation GmbH, 42653 Solingen, Germany, and also with FLIR Radiation Inc., Oak Ridge, TN 37830 USA (e-mail: juergen.stein@ir.com). Color versions of one or more of the gures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identier 10.1109/TNS.2012.2222665

the ability of source localization and spectroscopy, has been considered a promising technology in various elds like nuclear medicine [1], astrophysics [2] and industry [3]. The Compton camera usually consists of a scatter detector where Compton scattering occurs and an absorption detector where the scattered photon is absorbed. Position and energy measurements in both detectors are used to reconstruct the gamma energy spectrum and the location of the source. A two-plane planar Compton camera concept for location and nuclide identication of remote radiation sources has been proposed in a previous study [4]. As shown in Fig. 1, two scintillation detector arrays constitute the scatter and the absorption planes. A photon emitted from a remote source hits under the original the scatter plane with the initial energy . The incident photon is scattered in the scatter path , resulting in an energy deposit of . plane by the angle is The scattered photon with the energy assumed to be completely absorbed within the absorption layer of the by subsequent interactions. The original path incident photon is covered by the mantle of the cone as dened and the angle , which can be by the axis and , the coworked out by using the measured energies ordinates of the detector positions, and the distance between the two planes. The intersections of different cones provide the information on the location of radiation sources. For a realistic setup, reconstructed image locations are broadened due to real detector size, nite energy resolution and Doppler broadening. By means of Monte-Carlo simulations the construction of the Compton camera has been optimized to meet the requirements for Homeland Security applications.

0018-9499/$31.00 2012 IEEE

KONG et al.: PROTOTYPE COMPTON CAMERA ARRAY FOR LOCALIZATION AND IDENTIFICATION

1067

TABLE I ANGULAR RESOLUTIONS (FWHM, IN DEGREE) FOR DIFFERENT INCIDENT GAMMA ENERGIES, ESTIMATED FOR PLASTIC-NaI(Tl) AND CaF Eu NaI Tl SYSTEMS WITH DIFFERENT DETECTOR GRANULARITIES. THE AREAS OF SCATTER AND ABSORPTION PLANES ARE FIXED TO 1 m

Fig. 2. voxelSPEC mounted on a 76 mm

76 mm NaI(Tl) detector.

TABLE II HARDWARE SPECIFICATIONS OF THE VOXELSPEC

In this paper we introduce a large-size, scalable Compton camera design based on this concept, and present a functional prototype camera array consisting of four scatter and four absorber detectors, denoted as the system. Experiments with the prototype camera allowed us to develop and to test software algorithms, and to verify the model predictions of instrumental performance, which should be further applied to explore larger arrays. II. SCALABLE TWO-PLANE COMPTON CAMERA SYSTEM Advanced by a new generation of nuclear electronics called the voxelSPEC, a large-size scalable Compton camera design utilizing commercially available scintillation detectors with photomultiplier (PMT) readout has been developed by FLIR Radiation. According to detailed investigations on the system performance of the camera system under design, we decided to use plastic (PVT) or CaF (Eu) scintillator for the scatter plane, and NaI(Tl) detector for the absorption plane. For a low-threshold detector system, the camera angular resolution is basically determined by the limited energy resolution of the scatter detectors, which may lead to uncertain scattering angles that are determined from energy depositions in the scatter and the absorber planes. Moreover, the inuence of detector granularity on the achievable angular resolution has been investigated. For a xed area of scatter and absorption planes of 1 m , the angular resolutions for plastic-NaI(Tl) and CaF (Eu)-NaI(Tl) systems with different granularities were estimated for different incident gamma energies (Table I). For gamma energies lower than 500 keV, there is only little gain in angular resolution if the detector size (edge length of the detector face) is smaller than 5 cm. Finally, a detector size of 7.6 cm (3 inch) has been chosen for our purpose, considering a tradeoff between cost and performance. The large-volume detectors provide a broad eld of view for scattered photons and facilitate maximum efciency at moderate cost. Each scintillation detector is coupled with a compact voxelSPEC electronics module, distinguished by a very small form factor, which is directly plugged onto the PMT. A voxelSPEC mounted on a 76 mm 76 mm plastic detector is shown in Fig. 2. The voxelSPEC comprises a high-voltage supply, an

Fig. 3. Sample Cs-137 energy spectrum from a mm mm LaBr Ce detector collected by the voxelSPEC with energy range 03 MeV.

analog frontend and digitization stage, a digital signal processor based on an FPGA, a LED pulser and corresponding software for stabilization and temperature compensation [5], [6] of the PMT, and an Ethernet interface for communication and data transmission. Table II lists basic hardware specications of the voxelSPEC, and Fig. 3 shows a sample Cs-137 energy spectrum measured with a mm mm LaBr Ce detector coupled to a voxelSPEC with dynamic range 1:1000. With novel system-on-chip technology the voxelSPEC integrates all the components into a compact electronic module, and is able to deliver list-mode event data with nanosecond precision timing over non-proprietary Ethernet [7]. Multiple time-synchronized voxelSPEC units can serve as an enabling technology for many coincident-pulse and time-of-ight based radiation detection systems. The general block scheme of a Compton camera based on the voxelSPEC technology is shown in Fig. 4. The list-mode data streams delivered from individual voxelSPECs have to be combined and analyzed, and nally translated into pictures and messages which can be interpreted by an operator. For such a pur-

1068

IEEE TRANSACTIONS ON NUCLEAR SCIENCE, VOL. 60, NO. 2, APRIL 2013

Fig. 4. Block scheme of the scalable Compton camera systems based on the voxelSPEC technology. Scatter (S) and absorber (A) detectors are coupled to voxelSPEC modules (vS), which deliver time-stamped list-mode data to a PC hosting Event Builder software. Ethernet switches featuring Power over Ethernet (PoE) and PTPv2 cross-link and synchronize the voxelSPECs (red arrows). Ethernet connects the switches with the PCs analyzing the data streams (black arrows).

pose a software package has been developed providing real-time list-mode data processing and image reconstruction. The time stamps of raw list-mode data are rstly considered to identify coincident detector hits, which are further analyzed to select valid event data for image reconstruction. In the case of current camera construction the valid events are distinguished by a single hit in the scatter detectors in coincidence with a single hit in the absorber detectors, and by energy depositions and which are compatible with Compton scattering in the scatter plane. The selected data stream is then used for monitoring, alarming, imaging, identication, and localization. Depending on additional lters implemented in the software, the incoming coincident events are evaluated for real-time image reconstruction by Compton back-projection, and by an advanced fast-imaging algorithm that is capable of providing a reasonable estimate of the source position. As shown in Fig. 4, the use of the voxelSPEC technology replaces the signal and HV cables of usual systems and makes a system extension very easy. The camera system can be extended by simply connecting additional detector-voxelSPEC modules to free channels in the Ethernet switch, and editing the list of declared detectors in the system software. No adjustment of timing conditions or calibration factors is necessary. III. THE PROTOTYPE SYSTEM

Fig. 5. The functional prototype Compton camera system. Four plastic scatter detectors and four NaI(Tl) absorber detectors equipped with voxelSPEC prototypes (housed in the black metal tubes coupled with the PMT) are mounted in a mechanical frame. Overall view (top) and view of the voxelSPEC and the detector planes (bottom).

A. Experimental Results The prototype camera array has been used for many measurements altogether representing weeks of real data acquisition. The detector hard- and software allow stable, unattended operation of the camera array for many days, and provide easy-to-read information on the radiation source in real time. Benchmark measurements have demonstrated a system performance very close to the predictions based on detector characterization and modeling. Fig. 6 shows a screen shot of the system software showing exemplary results for a 10 Ci Cs-137 source, positioned in 50 cm distance from the scatter plane at elevation and azimuth with respect to the detector axis, in real time. In fact, a Cs-137 alarm showed up. Both the back-projection (right) and the fast-localization (middle) image indicate the source position correctly, but the advantage of the fast-localization algorithm is evident. B. Comparison With Model Predictions Measurements with the prototype camera array for common radiation sources have been performed to verify the model predictions made by Monte-Carlo simulations. Two complementary simulation frameworks, the SiST package developed by Herbach et al. [4], and Geant4, a standard simulation tool developed at CERN [8], have been applied for modeling the prototype array. Geant4 is generally appropriate for modeling of detailed geometries. All the construction elements could be considered in their real geometries, provided that the CAD-based data are available. Nevertheless, the most complex components like electronics boards or PMT dynodes were modeled as volumes homogeneously lled with an arbitrary medium reecting the average elemental composition (pseudo medium)

The prototype camera array used for performance tests and model verications is shown in Fig. 5. The camera consists of four mm plastic (PVT) detectors forming the scatter plane, and four mm NaI(Tl) detectors forming the absorption plane. Both the PVT and the NaI(Tl) detectors have squared cross-sections and are equipped with prototypes of the voxelSPEC electronics. All detectors are mounted in a mechanical frame designed for a larger array, so that the scatter-absorber inter-planar distance as well as the detector positions in the planes could be easily rearranged for a variety of test geometries.

KONG et al.: PROTOTYPE COMPTON CAMERA ARRAY FOR LOCALIZATION AND IDENTIFICATION

1069

Fig. 6. Screen shot of the prototype camera software showing live plots representing an estimate of the source position (middle: fast localization image; right: back-projection image) and the corresponding sum energy spectrum with the Cs-137 ROI highlighted (left). Data were obtained with a Cs-137 source located top right in front of the prototype system. The image axes are and in arbitrary units, where and denote polar and azimuth angle, respectively.

Fig. 7. Drawing of the prototype system model as implemented in Geant4. The voxelSPEC is modeled separately and not shown here.

and the total mass (pseudo density). This simplication signicantly reduces the expense of model construction as well as the computing time. Note that the simulations discussed here neither consider the real radiation source construction (encapsulation), nor the mechanical frame and other supporting materials as shown in Fig. 5. Fig. 7 shows the Geant4 model of the prototype camera array. Simulations were performed for a few standard scenarios and geometries which have been realized in measurements. The comparison with measured data has been focused on absolute count rates reecting detection efciencies, and selected differential data, such as energy spectra and scattering angle distributions, which are sensitive to energy resolution and detection thresholds. These parameters are critical for the instrumental performance of the prototype camera array. As examples, the sum energy spectra simulated for Co-57 and Cs-137 sources placed in the front center of the prototype system are shown in Fig. 8, together with the measured spectra for the same source geometry. The comparison demonstrates good agreement between the simulated spectra and the background-corrected measurements. Absolute valid total and full-peak event rates have been derived from the measured background-corrected sum energy spectra, and compared with the rates from simulations. Both simulation tools provide

Fig. 8. Sum energy spectra for Co-57 (top) and Cs-137 sources (bottom) positioned in the front center of the prototype system at a distance of 50 cm from the scatter plane. Measured spectra without background correction (black), measured background spectra (green), measured spectra with background correction (red), and the simulated spectra (blue) are plotted.

reliable event rate estimates. Fig. 9 presents exemplary Geant4 results in form of ratios representing the simulated event rates, divided by the measured event rates, for all valid events as well as for full-peak events generated by a couple of common radioactive sources. The simulation results are conrmed within 10%, except for the full-peak rate of Na-22 exhibiting a larger deviation. The normalization of simulated data is based on the source activities calibrated with an identiFINDER instrument of FLIR, thus reducing the activity uncertainties from 30% according to the source provider to about 15%. On the other hand, both Fig. 9 and Fig. 10 indicate a slight overestimation of full-peak response by simulation. This is not surprising since not all the construction details have been considered in the model. Those details basically represent additional absorbing or scattering materials which can only reduce the photodetection efciency but might feed the energy region of scattered gammas in the measured sum energy spectrum. Exemplary differential data are given in Fig. 10, showing the scattering angle distributions for Co-57, Ba-133 and Cs-137 sources in front center position. The scattering angles have been

1070

IEEE TRANSACTIONS ON NUCLEAR SCIENCE, VOL. 60, NO. 2, APRIL 2013

Fig. 9. Ratio of simulated to measured total (blue) and full-peak rates (red) of valid events for different sources placed in the front center of the prototype array at a distance of 50 cm from the scatter plane.

Fig. 11. Background contributions to sum energy spectra simulated for model camera arrays, corresponding to normal background level as measured in Solingen. The inter-planar distance is set to 30 cm for all the camera arrays.

Fig. 10. Distributions of scattering angles derived for Co-57, Ba-133, and Cs-137 sources in central position of the prototype system: Comparison of SiST simulations with measurements.

calculated event-by-event from the linearization-corrected energy depositions in corresponding scatter and absorber detectors. It is observed that the scattering-angle distributions measured extend to lower angles in the case of lowest gamma energies (Co-57) when compared with the curves derived from the simulated data with SiST. The reason is that the real detection threshold in the PVT scatter detectors is slightly lower than the threshold of 10 keV assumed in the simulations. IV. TOWARDS LARGER CAMERA ARRAYS As discussed above, the model predictions on the prototype camera array have been veried by a variety of experimental results. The good agreement between simulation and measurements provides condence that both simulation tools are appropriate for modeling of larger, advanced camera arrays.

Geant4 simulations have been performed for larger camera arrays consisting of 4 4, 8 8, 10 10, 13 13 and 16 16 detectors in scatter and absorption planes, respectively. Using the terrestrial background model introduced by Mitchell et al. [9], the background contributions to the sum energy spectra have been simulated for the model camera arrays and are shown in Fig. 11. The background modeling assumes an outdoor background radiation level of about 90 nSv/h as measured at FLIR Radiation in Solingen, Germany. The simulated backgrounds are of great importance for performance predictions of the camera arrays. The valid total and full-peak event rates simulated for the model camera arrays are presented in Fig. 12 as a function of the array size, translated into the number of detectors per plane. The lowest data points correspond to the prototype array. The overall efciency of a two-plane Compton camera is expected to be proportional to the scatter plane area (or the number of scatter detectors), times the absorption probability of scattered gammas in the absorber plane which scales with the solid angle of the absorber plane as seen by an average scatter detector. For large plane distances or small arrays, is approximately proportional to the number of absorber detectors, which leads to a square rule for the given construction with parallel, congruent planes . For small plane distances or large arrays, absorber detectors far off a given scatter detector contribute less to than the absorber detectors close by. Consequently, evolves slower, and the square rule is invalidated. This effect is enforced by the angular dependence of the Compton cross section preferring forward scattering. This general trend is nicely reected in the simulation results (Fig. 12): The simulated event rates increase with the array size and scale approximately with square of the detector number per plane from 4 4 to 13 13 array but show a noticeable fall-off for 16 16 array compared to the square rule. Because of shielding effect of outer detectors the inner detectors in the array planes experience a reduced level of natural background, and therefore the background count rates of the camera array increase slower

KONG et al.: PROTOTYPE COMPTON CAMERA ARRAY FOR LOCALIZATION AND IDENTIFICATION

1071

handled by (temporarily) removing part of the detectors from the software-controlled coincidence scheme. V. CONCLUSION Based on commercially available scintillation detectors and the voxelSPEC technology, a large-size, scalable two-plane Compton camera design dedicated to localization and identication of standoff radiation sources is currently being developed at FLIR Radiation. The use of large-volume scintillation detectors provides a broad eld of view for scattered photons and facilitates a maximum efciency at moderate cost. With the advanced voxelSPEC technology it is possible to construct customized, scalable camera arrays at minimum expense. A functional prototype camera array has been used for performance tests and for verication of model predictions obtained by Monte-Carlo simulations. Benchmark tests have demonstrated a system performance very close to the predictions based on previous detector characterization and modeling. Model predictions for the prototype camera array have been veried by measurements performed for a few standard scenarios and geometries. Good agreements between simulation results and measurements suggest that the modeling framework including the detector and material data used are well suited for modeling of larger, advanced camera arrays. Preliminary modeling results for camera arrays with , , , and detectors were presented. REFERENCES
[1] W. L. Rogers, N. H. Clinthorne, and A. Bokozdynia, Compton cameras for nuclear medical imaging, in Emission Tomography: the Fundamentals of PET and SPECT. Amsterdam, The Netherlands: Elsevier, 2004, pp. 213228. [2] V. Schnfelder, Lessons learnt from COMPTEL for future telescopes, New Astronomy Rev., vol. 48, pp. 193198, 2004. [3] J. B. Martin, N. Dogan, J. E. Gormley, G. F. Knoll, M. ODonnell, and D. K. Wehe, Imaging multi-energy gamma-ray elds with a Compton scatter camera, IEEE Trans. Nucl. Sci, vol. 41, no. 4, pp. 10191025, Aug. 1994. [4] C.-M. Herbach et al., Concept study of a two-plane Compton camera designed for location and nuclide identication of remote radiation sources, IEEE NSS Conf. Rec. N13-224, pp. 909911, 2009. [5] G. Pausch, J. Stein, and N. Teolov, Stabilizing scintillation detector systems by exploiting the temperature dependence of the light pulse decay time, IEEE Trans. Nucl. Sci, vol. 52, pp. 18491855, 2005. [6] J. Stein and G. Pausch, Detector for the measurement of ionization radiation, US patent, US8115176B2. [7] J. Preston et al., Novel timing method using IEEE 1588 and synchronous Ethernet for Compton telescope, IEEE NSS Conf. Rec. N47161, pp. 14041407, 2010. [8] S. Agostinelli et al., GEANT4-A simulation toolkit, Nucl. Instrum. Meth. in Phys. Res. A, vol. 506, pp. 250303, 2003. [9] D. J. Mitchell and G. P. Lasche, Calculation of background radiation for gamma-ray and neutron detectors, Sandia Rep. SAND2008-4478, 2008.

Fig. 12. Simulated valid total (top) and full-peak (bottom) event rates for different sources as a function of the number of detectors per plane. In the simulations a 1 mCi point-like source was placed in the front center of the array at a distance of 10 m from the scatter plane. The inter-plane distance is set to 30 array (15 cm distance). The total event rates arising cm except for the from simulated backgrounds are plotted in the top panel, too.

than the square rule. Consequently, larger arrays are also distinguished by an improved effect-to-background ratio. It is worth noting that even in the largest array considered (16 16 detectors) the detection limits are not noticeably affected by random coincidences. With a coincidence window of 50 ns as applied in the prototype array, and a typical background count rate of 200 to 400 cps per detector, the expected random coincidence rate of 0.5 kcps is small compared with true coincidence rate of 2 kcps caused by the natural background. Random coincidences in the presence of strong radiation sources causing a high load per detector could be easily

You might also like