Development of a Compton imager based on bars of scintillator
A.M.L. MacLeod, P.J. Boyle, D.S. Hanna, P.R.B. Saull, L.E. Sinclair, H.C.J. Seywerd
DDevelopment of a Compton Imager Based on Bars of Scintillator
A.M.L. MacLeod a,1, ∗ , P.J. Boyle a , D.S. Hanna a , P.R.B. Saull b , L.E. Sinclair c , H.C.J. Seywerd c a Physics Department, McGill University, 3600 rue University, Montreal, Quebec, Canada, H3A 2T8 b Measurement Science and Standards, National Research Council, 1200 Montreal Road, Ottawa, Ontario, Canada, K1A OR6 c Geological Survey of Canada, Natural Resources Canada, 601 Booth Street, Ottawa, Ontario, Canada, K1A 0E8
Abstract
We have developed a compact Compton gamma-ray imager with a large field of view and a low channel-count thatis capable of quickly localizing gamma-ray sources in the few hundred keV – several MeV range. The two detectorplanes (scatter and absorber) employ bars of NaI(Tl) read out by photomultiplier tubes (PMTs) located at each end.The long-range imaging performance has been tested from 392 keV to 1274 keV. An angular resolution measure of2 . ◦ ± . ◦ and an e ffi ciency of (1 . ± . × − at 662 keV is obtained. A Cs (662 keV) source equivalentto a 10 mCi source 40 m away can be located in 60 seconds with an uncertainty of about a degree. No significantdegradation in imaging performance is observed for source angles up to 40 ◦ o ff axis. Keywords:
Compton Imaging, Compton Telescope, NaI, Scintillator, Security, Radiation Detection, Pulse-HeightSharing
Acknowledgements
This is NRCan / ESS Contribution 20140132.
1. Introduction
A portable Compton gamma-imaging device is a helpful tool in the prevention or clean-up of a radiological ornuclear incident. First responders require portable radiation detectors operating in the 0 - 3 MeV range that can bedeployed in the field. To meet their needs, we have developed a Compton imager that is low-cost, ruggedize-able, andadaptable to suit a number of di ff erent surveying platforms (truck, helicopter, etc.). The detector is highly sensitive,(capable of localizing a 10 mCi Cs source at 40 m distance with degree-level precision in 60 s), making it suitablefor real-life situations involving radiation sources that can be weak, distant or shielded.Some highly-compact Compton-imaging designs based on scintillator [1] and semiconductor technologies (CZT [2,3])) have been developed for commercial use. In the future, development in crystal growth technology may make theseviable options for a ff ordable larger-scale detectors. However, at present, these small ultra-portable detectors have lim-ited sensitivity. In parallel, extremely large highly-sensitive imagers have been developed [4, 5, 6, 7]. However,these do not meet the requirements of agencies with limited funds, nor are they suitable for mounting onto multipleplatforms. Our detector aims to operate in the middle range, by being both sensitive and portable (can be loaded intoand out of a vehicle), while remaining a ff ordable. A more expensive design based on scintillator with SiPMs has beendeveloped [8]. The design presented in this paper employs long bars of scintillator (NaI(Tl)) with photomultipliertube (PMT) readout, a combination well-proven for application in robust low-cost detectors operating in the field. ∗ now at the Geological Survey of Canada. Email address: [email protected] (A.M.L. MacLeod)
Preprint submitted to Preprint submitted to Nuclear Instruments and Methods in Physics Research A October 4, 2018 a r X i v : . [ phy s i c s . i n s - d e t ] J un E x )( E x ) θ C Figure 1: Compton imager schematic depicting the energy and position of the scattered electron, E , x , recorded in the scatter plane, and theenergy and position of the scattered gamma ray, E and x , recorded in the absorber plane. The Compton cone, of opening angle θ C , is calculatedusing the energy deposits and localizes the possible source locations.
2. Compton Imaging
In a Compton gamma imager, an incoming gamma ray of energy E γ Compton interacts in a scatter detector,depositing energy E . The outgoing scattered gamma ray deposits its energy E in an absorber detector. NeglectingDoppler broadening, the scattering angle between the initial and final-state gamma rays, θ C , can be determined fromthe two energy deposits, via cos θ C = + m c ( 1 E γ − E ) , (1)where E γ = E + E and m c is the electron rest energy. As illustrated in Figure 1, the position of the source liessomewhere on a cone of opening angle θ C , where the cone axis is on the line between the two energy deposits. Thelocation of the source can be reconstructed from the intersection of several cones.
3. Bar Design
We have designed, constructed and tested a Compton imager made of long bars of NaI(Tl) read out by PMTs fixedto the bar ends. Pulse-height sharing between the PMT signals is used to determine the position of the interaction.The use of long bars of NaI(Tl) o ff ers an inexpensive design solution, as NaI(Tl) is relatively low cost, and the barformat requires few readout channels.Long scintillation bars have previously been employed in Compton imagers where the energy deposits are large [5],however to employ this technique at lower energies, an optimization of the attenuation length of the bars is needed.Studies have been done to look at the e ff ect of tuning the attenuation length of scintillator in order to optimize itsposition reconstruction [9, 10, 11, 12], which holds well at high incident gamma-ray energies where one is not photo-statistics limited. However, operating at low energies (less than 150 keV in the scatter layer for incoming 662 keVgamma rays), requires tuning the bar attenuation length to optimize the overall performance of the imager. To this end,a highly detailed Monte-Carlo simulation using optical transport was developed to determine the optimal attenuationlength of the scintillator bars. A prescription was then developed to manufacture these bars for the detector.
4. The Detector
Previous work has shown that for the scatter plane, the optimal thickness for one Compton scatter of 662 keVgamma rays in NaI(Tl) is ∼
20 mm [13]. With this thickness, approximately 20% of 662 keV gamma rays Compton2 igure 2: Prototype consisting of ten scatter modules (front) and seven absorber modules (back) with a scatter-absorber spacing of 425 mm.Figure 3: A single absorber detector module. The scintillation light from a bar of encapsulated NaI(Tl) is read out by PMTs placed at the ends ofthe bar. The bar and PMTs are enclosed in a thin aluminum shell and packed with foam for protection. scatter once and exit. For the absorber plane, a thickness of 40 mm is chosen, a compromise among increasing theabsorption fraction, maintaining comparable lateral and longitudinal position resolution, and matching the bar crosssection with commercially available PMTs.A photograph of the full-scale prototype is shown in Figure 2. The modular scatter and absorber layers are425 mm apart (center to center) and mounted on a frame constructed of 1 /
2” PVC tubing. The scatter plane con-sists of ten 16 × ×
200 mm and 20 × ×
200 mm bars of NaI(Tl) and the absorber plane consists of seven40 × ×
400 mm bars of NaI(Tl).An overview of the scintillator dimensions and the PMT sizes, QEs and model numbers of all the components areprovided in Table 1. In the scatterer, we used both Super bialkali (SBA) PMTs, with QE ∼ ∼ , while the PMTs were from Hamamatsu .Each NaI(Tl) bar is encapsulated in aluminum with a quartz window located at each end. The bars are read out by Plane Scintillator PMT number size (mm ) model size (mm) QE5 20 x 20 x 200 R8900U-100 23.5 SBA (35%) Scatter
Absorber
Table 1: Summary of the detector components making up the scatter and absorber plane. Only four of the seven absorber bars were included in thedetector for assessing the performance for reasons discussed in Section 6. /
16” thickness aluminum container (see Figure 3)and packed with foam to protect the assemblies from mechanical shock. The PMT high voltage (HV) is supplied bya CAEN SY 2527 Universal Multichannel Power Supply System.The data acquisition uses a combination of VME, CAMAC and NIM standards. The PMT pulses are amplifiedby a factor of ten (Phillips Scientific ff ectively requiring one or more end-to-end coincidencesin the scatter layer together with the same pattern in the absorber layer (Phillips Scientific 754 Logic Units). Thesecond set of amplified PMT signals is fed directly into a 48-channel 10-bit VME VF48 digitizer [14] and integratedon-board when a coincidence is present. The resulting charge values are then transferred to a computer via a USB-VME bridge (CAEN V1718). Immediately, a software cut is made to include only events with exactly one scatter barhit and one absorber hit, where a hit denotes that both the left and right PMTs of a given bar are above threshold. Theenergy deposit is reconstructed in a given bar by summing the signals of the left PMT (S1) and right PMT (S2). Theposition is reconstructed using the ratio of these two signals, S1 / S2.
5. Development and Performance of Scintillator Bars
The photon propagation in a scintillator is determined by its physical and optical properties as well as its surfacefinish, coating and wrapping. The key to our design is tuning the attenuation length of the scintillator bars in order toimprove the position resolution while maintaining as good an energy resolution as possible. The attenuation can betuned by modifying the surface finish, e.g. polishing or degrading the smoothness [10, 9, 12]. This section provides adetailed explanation of how the energy and position reconstruction depend on the attenuation. The optical absorptioncoe ffi cient, α , multiplied by the length of the bar, L , is used here as a measure of the attenuation. x−L/2 +L/2Position Resolution σ x collimated sourceS1PMT PMTS2 Figure 4: Schematic of the position reconstruction using a bar of length L indicating S and S , the signals from the left and right PMTs, and x ,the displacement of the interaction from the bar center. For an interaction at the longitudinal position x in a bar of length L , the signal S from the left PMT can be writtenas S = N Pe − α ( L + x ) , (2)and the signal from the right PMT as S = N Pe − α ( L − x ) , (3)where N is the number of photons generated and P is the probability of detecting each photon. The probability, P ,incorporates all sources of photon loss that are independent of the position of interaction, such as the probability of http: // L and x are the scintillator length and thedisplacement of the interaction from the bar center, respectively (see Figure 4).Propagating the photostatistical uncertainties of S and S , yields an expression for the position resolution [11]: σ X Stat = α √ N P e α L (cid:112) cosh( α x ) . (4)Equation 4 highlights what is known to be true empirically. The position resolution depends on the initial number ofphotons produced, N , the probability of detecting each photon, P , and the absorption coe ffi cient, α . The minimumof the position resolution along a bar of length L occurs when α L = α L for a polished bar of NaI is muchless ( α L ∼ × ×
200 mm bar). Thus, increasing α should improve the position resolution. However,increasing α also has the undesirable e ff ect of increasing the position dependence of the position resolution. The expression √ S × S is proportional to N and independent of x and can thus be used to reconstruct the energywithout knowing the position where deposition occurred. A derivation similar to the position resolution calculation canbe used to obtain an expression for the relative uncertainty of the energy due to statistical fluctuations by propagatingthe statistical uncertainties of S and S : σ E Stat E = √ N P e α L (cid:112) cosh( α x ) . (5)Equation 5 shows that increasing the number of detected photons, either by increasing N or P , improves theenergy resolution whereas increasing α worsens the energy resolution. Increasing α also has the undesirable e ff ect ofincreasing the position dependence of the energy resolution. The situation is therefore complicated. Decreasing theattenuation length (increasing α L ) both improves the position resolution and worsens the energy resolution. Thus, todetermine the ideal attenuation length, an optimization is undertaken with simulations. The total energy resolution, σ E , is equal to a combination of σ E Stat , the energy resolution due to statistical fluctua-tions, and σ E Int , the intrinsic energy resolution, σ E = σ E Stat + σ E Int . (6)By combining Equations 4 and 5, one obtains an expression relating the statistical component of the positionresolution to the statistical component of the energy resolution, σ X Stat = α σ E Stat E . (7)Equation 7 is then combined with Equation 6 to obtain an expression for the position resolution, σ X , as a functionof the measured total energy resolution, ( σ E E ), σ X ≈ σ X Stat = α (cid:114)(cid:18) σ E E (cid:19) − (cid:18) σ E Int E (cid:19) . (8)By assuming a fixed value of 6% for the intrinsic energy uncertainty, we use Equation 8 to obtain an estimate of theposition resolution indirectly from the energy resolution measurement at energies for which the position resolutionis not directly measured. Note that the intrinsic energy resolution depends on the quality of the crystal and is notmeasured for our scintillator. Thus, the assumption of a fixed 6% intrinsic energy uncertainty contributes to a moderateerror in the estimation of the position uncertainty at higher energies ( ∼
5% error at 662 keV), with the error decreasingat low energies where the uncertainties from photostatistics dominate.5 .4. Optimal α L from Simulations
GEANT4 [15] simulations were used to determine the optimal attenuation length of the bars.First, a detailed optical-photon transport was carried out to determine the energy and position resolution forNaI(Tl) bars as a function of attenuation length. Di ff erent attenuation lengths were modeled by making part of thebar surface absorbent. The surface treatment is referred to as polished front-painted in GEANT4 and is specularlyreflective with 96% reflectivity. The bars simulated were of sizes 20 × ×
200 mm and 40 × ×
400 mm andhad 0%, 25%, 50%, 75% and 100% of one side surface-treated, exploring α L values in the range of ∼ ff erent combinations (5 ×
5) of scatterer and absorber were compared. For eachdetector combination, a simulation equivalent to a three-hour exposure to a 10 mCi
Cs source located on thedetector symmetry axis 40 m away was carried out. The energy and position of the interactions recorded in GEANT4were smeared using the energy and position resolution functions derived from the detailed optical simulation. Onlyevents with full gamma-ray energy deposit shared between exactly one scatter bar and exactly one absorber bar wereselected. Furthermore, events exhibiting the energy signature of a back-scatter event (incoming gamma scattering ∼ ◦ backwards from absorber to scatterer) were rejected.The width of the angular resolution measure (ARM) for the selected events was used as a figure of merit fordetector performance. ARM = θ C − θ geom , where θ C is the Compton angle calculated from the energy depositsand θ geom is the observed scattering angle based on the known source position and the reconstructed hit positions.Figure 5 (left) shows the standard deviation, σ ARM , as a function of scatter bar α L , with the absorber α L fixed at 1.01(25% of one side surface-treated). A significant improvement ( ∼ σ ARM is observed, decreasing from 3.7 ◦ to 2.8 ◦ with increasing surface treatment. Figure 5 (right) shows σ ARM as a function of absorber bar α L , with thescatter α L fixed at 1.14 (50% of one side surface-treated). Once again, σ ARM decreases with increasing α L , levelingo ff at α L ∼
1. The ARM for the remaining possible combinations of scatter and absorber α L demonstrated similarbehaviour. Since decreasing the attenuation length (or increasing α ) also increases the non-linearity of the energy andposition resolution along the bar, an α L of 1, where the improvement in σ ARM levels o ff , was deemed optimal. L a scatter bar ) (cid:176) ( AR M s a absorber No Surface Treatment 1 Side 25 % Diffuse 1 Side 50 % Diffuse 1 Side 75 % Diffuse 1 Side 100 % Diffuse L a absorber bar ) (cid:176) ( AR M s a scatter No Surface Treatment 1 Side 25 % Diffuse 1 Side 50 % Diffuse 1 Side 75 % Diffuse 1 Side 100 % Diffuse Figure 5: Simulation. Left: σ ARM as a function of scatter bar α L , for absorber α L fixed at 1.01 (25% of one side surface-treated). Right: σ ARM asa function of absorber bar α L , for scatter α L fixed at 1.14 (50% of one side surface-treated). Note that the y axis is zero-suppressed and the errorbars are smaller than the symbols in both the left and right figures. .5. Determination of the Surface Treatment Position (mm) -100 -50 0 50 100 æ S Æ / æ S Æ
120 grit220/120 grit220 gritPolished120 grit220/120 grit220 gritPolished120 grit220/120 grit220 gritPolished120 grit220/120 grit220 gritPolished
Figure 6: Plot of the ratio, (cid:104) S (cid:105) / (cid:104) S (cid:105) for a 20 × ×
200 mm bar at 662 keV, as a function of position for a polished bar (blue stars); four sidesdegraded with 220 grit sandpaper (green triangles); two sides degraded with 220 grit and the remaining two sides with 120 grit (red squares); andfour sides degraded with 120 grit (black circles). The data sets can be described by exponential functions. Note that the data have been scaled suchthat the middle point is located at (0,1). Surface Treatment α L ∆ EE σ x (% FWHM) (mm)Polished 0.23 7.3 18220 0.90 8.4 7220 /
120 0.92 9.8 8120 1.14 9.5 6
Table 2: Scionix data for α L measured for a 20 mm bar with successive levels of sanding and the energy resolution measured at the bar center at662 keV. The position resolution values presented here are estimated from the energy resolution and α using the technique described in Section 5.3.1. A technique to modify the bar surface was developed with the manufacturer Scionix to achieve scintillator barswith attenuation lengths of α L ∼
1. Due to the hygroscopic nature of NaI(Tl), any alteration of the scintillator surfacemust take place in a low-humidity environment. Thus, the procedure to modify the bar surface was undertaken in adry room at the manufacturer’s site and carried out by the manufacturer.To tune the attenuation length, the bar surface was degraded by sanding with various grit sizes in the longitudinaldirection. Four di ff erent surfaces were compared: polished; four sides degraded with 220 grit sandpaper; two sidesdegraded with 220 grit and the remaining two sides with 120 grit; and four sides degraded with 120 grit. Figure 6shows the ratio of the mean values of the signals at the two ends, (cid:104) S (cid:105) / (cid:104) S (cid:105) , versus the position of a collimated662 keV energy source along the bar, for these four surface treatments. As the sandpaper grit number decreases(becomes more abrasive), the attenuation of the bar increases. The attenuation length was obtained by applying anexponential fit to the data. As shown in Table 2, α L increases from 0.23 for an untreated bar to 1.14 for 120 gritapplied to all four sides.Energy resolutions measured with the 662 keV source at the center of the treated bars are also presented in Table 2,along with position resolutions calculated using the technique described in Section 5.3.1. (Note that the ratio values7sed to determine α L and the energy resolutions were provided by Scionix without uncertainties.)We find, as expected, a dramatic improvement in position resolution as α L increases to close to 1 with onlyminor improvements with further degradation of the bar surface. The accompanying worsening of energy resolutionis comparatively minor. The second most attenuating surface treatment (220 grit to two sides and 120 grit to the twoother sides) was chosen for the 20 mm bars, yielding α L = α L values ∼
1. For the 16 mmbar, the same combination of 220 grit and 120 grit was chosen, while for the 40 mm bar, a surface treatment of 120grit was chosen. To test the reproducibility of the technique, bars were then ordered from two other manufacturers,Hilger and Saint-Gobain, who employed the same prescription for the surface treatment.
In total, 18 bars were tested from three manufacturers, Scionix, Hilger, and Saint-Gobain. The bars came in threesizes: 16 mm width (five bars), 20 mm width (six bars), and 40 mm width (seven bars). Each set had the same surfacetreatment applied to it. For in-house testing, the bars were paired with PMTs having areas that matched their crosssection. The 16 mm and 20 mm bars were paired with R8900U-100 PMTs (QE = = µ Ci Cs sourcewas placed behind a cylindrical collimator having a thickness of 70 mm and an aperture of 5.5 mm. The collimatorwas attached to a motorized platform (Arrick Robotics Model XY-30). Four scans of each bar were made, two in eachdirection, with 5 mm steps. The left and right PMT outputs ( S and S ) were obtained by digitizing the PMT signalswith a 1 GS / s, 8-bit digitizer (Acqiris DC270) and computing the area of each pulse. The digitizer was triggered bycoincidence of the left and right PMT signals. At each step, 10 000 coincidence events were recorded.To obtain the position response, Gaussians were fit to the photopeaks in S and S and the photopeak pulse-heightmeans, (cid:104) S (cid:105) and (cid:104) S (cid:105) , were measured. Figure 7 shows the signal ratio, (cid:104) S (cid:105) / (cid:104) S (cid:105) , versus distance along the bar, foran individual 20 mm bar from each of the three manufacturers.To determine the position resolution, only events within 2.5 standard deviations of the mean of the photopeaksin S , S , and their sum, were considered. For these events, the ratio, (cid:104) S (cid:105) / (cid:104) S (cid:105) , was calculated and converted toa position using the position versus signal characterization. The width of the resulting distribution of positions wasadopted as the position resolution at 662 keV.Calibration of the bars at other energies was performed di ff erently. Sources were placed only on the bar center. AGaussian was fit to the summed signal to obtain a parametrization of the energy resolution and energy versus signal.To obtain a parametrization of the position resolutions for these energies, the approach described in Section 5.3.1 wasused.The average α L , and the energy and position resolutions obtained at 122 keV and 662 keV for the 16 mm and20 mm scatter bars are shown in Tables 3 and 4, respectively. The uncertainties on the attenuation measurements(when N , the number of bars measured, was greater than one) were estimated from the spread between the di ff erentbars and their relatively small values ( ∼ ff erent manufacturers have similar attenuation. However, the measuredenergy and position resolutions di ff er significantly from one manufacturer to another, with the Hilger bars exhibitingthe worst performance.The performance of the bars was also compared with the performance of a simulated bar of equivalent α L . Theseresults are shown in Table 3 for the 16 mm bar, and similar results were observed for the other bar sizes. The measuredenergy and position resolution values at 662 keV were found to be in good agreement with the simulations. However,those for 122 keV were found to be slightly worse than predicted ( < ∼
8f NaI(Tl) accurately, where the properties of the scintillator, scintillator surface, and PMT photocathode are di ffi cultto model and also vary from one manufacturer to another.The average α L , and energy and position resolutions obtained at 662 keV for the 40 mm absorber bars are shownin Table 5. The Scionix bars are significantly more attenuating ( α L = α L ∼ Position (mm) -100 -50 0 50 100 æ S Æ / æ S Æ Saint Gobain 0.001 – L = 0.863 a Position (mm) -100 -50 0 50 100 æ S Æ / æ S Æ Hilger 0.0004 – L = 1.0027 a Position (mm) -100 -50 0 50 100 æ S Æ / æ S Æ Scionix 0.001 – L = 1.032 a Figure 7: Plot of the ratio, (cid:104) S (cid:105) / (cid:104) S (cid:105) , for an individual 20 mm bar from three manufacturers (Saint-Gobain (left), and Hilger (middle) and Scionix(right)). To obtain α L , the central region of each set of points is fit to an exponential function extending from - 60 mm to +
60 mm.
Manufacturer N BAR α L ∆ EE
122 keV ∆ EE
662 keV σ x
122 keV σ x
662 keV% (FWHM) % (FWHM) (mm) (mm)Hilger 2 1.02 ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± Table 3: α L , and the energy resolution and position resolution measured at both 662 keV and 122 keV for the 16 mm bars provided by Hilger andScionix. N BAR represents the number of bars tested. The position resolution values followed by * are estimated from the energy resolution usingthe technique described in Section 5.3.1. The performance of a simulated bar of equivalent α L is presented in parentheses. Manufacturer N BAR α L ∆ EE
122 keV ∆ EE
662 keV σ x
122 keV σ x
662 keV% (FWHM) % (FWHM) (mm) (mm)Saint-Gobain 1 0.863 ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± Table 4: α L , and the energy resolution and position resolution measured at 662 keV and 122 keV for the 20 mm bars. The position resolutionvalues followed by * are estimated from the energy resolution using the technique described in Section 5.3.1. N BAR α L ∆ EE
662 keV σ x
662 keV% (FWHM) (mm)Hilger 2 0.76 ± ± ± ± ± ± ± ± ± Table 5: α L , the energy resolution and position resolution measured at 662 keV for the 40 mm bars.
6. Bar Failures
Three of the seven absorber bars experienced a sudden change in performance. The attenuation length decreased,accompanied by an increase in the number of photoelectrons detected at one end of the bar and a decrease at theother end. Both the Scionix and Saint Gobain bars had this issue, but not the Hilger bars. This is currently underinvestigation and there are several possible hypotheses, such as contamination of the NaI crystal, problems with theglue on the glass-scintillator interface, failure of the plastic seal that surrounds the quartz windows at the bar ends, andcracking due to the thermal expansion of the crystal, glass and aluminum. Integrating the PMT into the encapsulationis a possible development that would improve ruggedness and may mitigate this problem.
7. Detector Performance
Energy Source Source Strength(keV) s − ( × )392 Sn 3.2 ± Cs 22.5 ± Na 22.3 ± Table 6: List of the radioactive sources imaged along with the gamma-ray energy and source strength. While Na provides two peaks, 511 keVand 1274 keV, only the imaging performance of the 1274 keV peak was assessed.
Due to the failure of three absorber bars, only four absorber bars, located centrally in the absorber plane, wereincluded in the detector performance studies.Long-range imaging tests were performed in a large room (16 m x 17 m) with a false floor at the National ResearchCouncil (NRC) in Ottawa, Canada. A list of the sources used is provided in Table 6. The gamma-ray emitting sourceswere placed 9 m away from the detector, at di ff erent angles [( φ = ◦ , θ = , , , ◦ ), ( φ = ◦ , θ = , ◦ )and ( φ = ◦ , θ = , ◦ )]. The origin of the coordinate system used lies at the center of the scatter plane. Thex-axis lies along the length of the bars, the y-axis points upwards, and the z-axis points out of the front-face of thedetector away from the absorber. The azimuthal angle, φ , lies in the x-y plane and θ is the polar angle with respect tothe z-axis. A description of the trigger selection is provided in Section 4. Several criteria were used to select events o ff -line.First, events with exactly one scatter bar and exactly one absorber bar above threshold were selected. Second, a ∼ σ energy cut around the photopeak in the total energy distribution was implemented. Third, to reject back-scatter events,where a gamma ray is scattered in the absorber first, and then absorbed in the scatterer second, a ∼ σ exclusion cutwas applied to the scatter-energy back-scatter peak. Fourth, a minimum threshold of 30 keV was required in the scatter10ource Energy Scatterer Sum(keV) (keV) (keV) Sn 392 30 - 100 340 - 440
Cs 662 30 - 150 600 - 720 Na 1272 30 - 180 or 280 - 650 1150 - 1350
Table 7: The event selection used includes a back-scatter rejection cut and a photopeak selection cut.
Total Energy (MeV) S ca tt e r E n e r g y ( M e V ) Cs-137
Figure 8: Scatter energy plotted against summed energy for
Cs (662 keV). The vertical lines indicate the cuts performed on the summed energy.The horizontal lines indicate the selection performed on the scatter energy. The events that fall into the rectangular region formed by the overlapof the four lines were selected. Note that events from the back-scatter peak, which is visible at a total energy of 0.66 MeV and a scatter energy of0.19 MeV, were thereby excluded. detector. A summary of the energy selection used for the three sources is provided in Table 7. A visual representationof the cuts used for a
Cs source is shown in Figure 8.Figure 9 shows the summed energy distributions (scatter + absorber) for a Cs source before and after theselection criteria are applied, overlaid with the estimated contribution from naturally occurring radioactive material(NORM). The NORM contribution was determined by taking a run of equal duration with no source present. Energydistributions for the Na and
Sn sources are shown in Figure 10. The varying fractional contribution from NORMto the total energy distribution is due to the di ff erent source strengths. There also exists a significant number ofaccidental coincidence events with NORM.We define the detector e ffi ciency as N / N Total , where N is the number of events that satisfy the event selectioncriteria and N Total is the total number of gamma rays passing through the active area of the front face of the detector(active area =
40 000 mm ). To determine the e ffi ciency, we relaxed the photopeak cut and in its place used a fitto isolate events in the photopeak. All of the other selection criteria were applied. To account for the number ofbackground events, a Gaussian and a third order polynomial is fit to the summed energy distribution, see Figure 11.The correctly reconstructed forward-going scatters fall in the Gaussian part of the distribution. The e ffi ciency of thedetector for a 662 keV source placed o ff -axis ( θ = ◦ ) was measured to be (1 . ± . × − . The e ffi cienciesat 392 keV and 1274 keV are provided in Table 8. 11 otal Energy (MeV) N AllNORMSelected
Cs-137
Figure 9:
Cs (662 keV) energy distribution. White - all data. Grey Shaded - estimated contribution from NORM. Red Lined - selected events.
Total Energy (MeV) N AllNORMSelected
Na-22 Total Energy (MeV) N AllNORMSelected
Sn-113
Figure 10: Energy distributions for Na (left) and
Sn (right) from 2 hours of exposure. White - all data. Shaded Grey - estimated contributionfrom NORM. otal Energy (MeV) N Cs-137
Figure 11: The summed energy distribution for
Cs (662 keV) fit to a Gaussian and a third order polynomial.
Energy E ffi ciency(keV) ( × − )392 2.00 ± ± ± Table 8: E ffi ciency for three sources positioned at θ = ◦ , φ = ◦ . The uncertainties reflect both the statistical fluctuations and the uncertainty inthe source emission rate. .1.1. Angular Resolution Measure ARM (degrees) -30 -20 -10 0 10 20 30 N Experimental data o – o = 2.72 s ARM (degrees) -30 -20 -10 0 10 20 30 N Monte-Carlo data o – o = 2.66 s Figure 12: ARM distribution for a
Cs source located on-axis at θ = ◦ for selected events comparing measurement (left) and Monte Carlo (right).The distribution is fit to the sum of a Gaussian and a third-order polynomial. The widths of the Gaussian fits are shown. Figure 12 (left) shows the ARM distribution for the selected events of a
Cs source located on-axis. The ARMdistribution has been fit to the sum of a Gaussian distribution and a third-order polynomial. The width of the Gaussianfit yields σ ARM of 2 . ◦ ± . ◦ .The experimental ARM was compared with GEANT4 simulation. The detector configuration matched the imagerthat was tested, with ten scatter bars and four absorber bars. The simulation made use of realistic energy and positionresolutions based on experimental data (see Section 5.6). The recorded energies and longitudinal coordinates (x)of the interactions were smeared by Gaussian probability distributions with widths determined by the energy andposition resolutions measured for each bar. Both the z and y-coordinates of the interactions were assumed to be atthe bar centers. The simulated ARM distribution is shown in Figure 12 (right), and the width of its Gaussian fit is σ ARM = . ◦ ± . ◦ . The good agreement with experimental data confirms our understanding of the detectorperformance. A back-projection image is a 2-D distribution formed by back-projecting Compton cones into angle space. Fig-ure 13 shows a back-projection image from ∼
20 000 Compton cones (two hours of data) collected with the
Cs sourcelocated 9 m away and 20 ◦ o ff axis. The back-projection image correctly indicates the position of the source.14 (degrees)-40 -30 -20 -10 0 10 20 30 40 Y ( d e g r ees ) -40-30-20-10010203040 010002000300040005000 Cs-137 at 9 m
Figure 13: A back-projection image of selected events for a
Cs source. The x and y-axis are presented in vertical and horizontal degrees. Thetrue source position is indicated by cross-hairs. The broad excess at the origin is caused by NORM. χ Minimization Algorithm
An iterative χ -minimization procedure using the MINUIT package [16] was employed to determine the vectorˆ s ( θ , φ ) which best represents the direction from the origin to the source. For N events, a χ function is constructedwith two fit parameters, θ and φ , representing the polar and azimuthal angles of the source direction vector which areto be determined: χ = N (cid:88) i = (cid:34) ACA( θ, φ ) i σ ACA i (cid:35) , (9)where ACA( θ , φ ) is the angle of closest approach of the direction vector ˆ s to cone i , and σ ACA i is the uncertainty ofthis angle. The expression for ACA is given byACA = arccos( ˆ s ( θ, φ ) · ˆ r i ) − θ Ci , (10)where ˆ r i is the unit-vector axis for cone i , and θ Ci is the Compton scattering angle of the i th cone. The uncertaintyof ACA is calculated on an event-by-event basis from the cone-axis uncertainty and the Compton opening angleuncertainty (using Equation 1), taking into account the uncertainties on the energy deposits and their positions.To extract the source direction from a given sample of N events, three iterations are performed, with the directiondetermined in each iteration passed on as a starting seed direction for the subsequent iteration. The highest bin of thecoarse-binned back-projection image is used as a seed in the first iteration. For a detailed explanation of the algorithmsee ref [17]. 15 .4. Imaging Performance at 662 keV To characterize the imaging performance of the detector, the
Cs source rates at a 9 m distance were convertedinto an equivalent 10 mCi source at 40 m with 100% branching ratio . To quantify the ability to reconstruct θ and φ ,the events were divided into groups or trials of 60 s of data, and the χ -minimization algorithm was applied to eachgroup. To determine the accuracy, the means of the resulting distributions of fitted θ and φ were used as a measureof the average reconstructed source location. To determine the angular precision, the 2-D standard deviation of thereconstructed direction was used. Table 9 summarizes (cid:104) θ (cid:105) , (cid:104) φ (cid:105) and the angular precision obtained from sixty 60-secondtrials. The sources are all reconstructed to within about a degree of the known source position. The small discrepancyobserved in the reconstructed angles versus the known source position can be explained by the fact that the quoteduncertainties on the reconstructed angles are statistical in nature and do not include a number of factors such as thee ff ect of NORM, trigger threshold variations, and the depth of interaction in the bars. The angular precision obtainedis on the order of a degree. Integration time (s) ) (cid:176) A ngu l a r P r ec i s i on (
662 keV
Figure 14: Angular precision for an equivalent 10 mCi 662 keV source located 40 m away as a function of integration time for five source locations( φ = ◦ ; θ = ◦ , 10 ◦ , 20 ◦ , 30 ◦ , and 40 ◦ ). At all angles, the angular precision reaches one degree in about a minute. Note that the data points arecorrelated and each point includes the entire data set. The relative source strengths are taken into account, as is the r dependence and the attenuation by the air. θ o φ o (cid:104) θ (cid:105) (cid:104) φ (cid:105) Angular Precision( ◦ ) ( ◦ ) ( ◦ ) ( ◦ ) ( ◦ )1 0.1 95.2 0.3 ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± Table 9: Summary of the fit values obtained at 662 keV: (cid:104) θ (cid:105) , (cid:104) φ (cid:105) and the angular precision for 60-second trials. Note that the source position forRun φ is undefined, thus the φ value obtained for this run is not meaningful. Figure 14 shows the angular precision at 662 keV as a function of time for several source positions. The detectordemonstrates a wide field-of-view with good performance for sources located up to 40 ◦ o ff -axis. ff erent Energies The 1274 keV and 392 keV peaks of a Na and
Sn source located at θ = ◦ , φ = ◦ were also successfullyimaged. The results were again scaled such that the rates of gamma rays incident on the detector are equivalent to thatfrom a 10 mCi source at 40 m with 100% branching ratio, allowing for direct comparison with the 662 keV data. Dueto the significant contribution from NORM to the Sn data, it was necessary to set the starting seed position used bythe imaging algorithm to the known location of the source. For 60-second runs, we obtained an angular precision of1.0 ◦ ± ◦ and 1.6 ◦ ± . ◦ for Na and
Sn, respectively.
8. Conclusion
We have developed a Compton-scatter gamma-ray imager composed of bars of NaI(Tl). The key to the low-costdesign is the use of pulse-height sharing to reconstruct events, made possible by the successful development of asurface treatment that was applied to the bars. The technique was shown to be robust and the bars were incorporatedinto a fully-functional imager. The detector demonstrates good imaging performance over a wide range of energies(392 keV, 662 keV and 1274 keV) and over a wide field of view. An angular precision of about one degree has beenachieved in one minute for a 10 mCi 662 keV gamma-ray source with 100% branching ratio, located 40 m away.We are currently investigating the use of di ff erent scintillator materials (CsI) and more compact readout devices,leveraging the experience our group has in constructing pixellated Compton imagers based on CsI(Tl) read out withsilicon photomultipliers [8, 18]. References [1] J. Kataoka, A. Kishimoto, T. Nishiyama, T. Fujita, K. Takeuchi, T. Kato, T. Nakamori, S. Ohsuka, S. Nakamura, M. Hirayanagi, S. Adachi,T. Uchiyama, and K. Yamamoto. Handy Compton camera using 3D position-sensitive scintillators coupled with large-area monolithic { MPPC } arrays. Nuclear Instruments and Methods in Physics Research Section A: Accelerators, Spectrometers, Detectors and Associated Equipment ,732(0):403 – 407, 2013. Vienna Conference on Instrumentation 2013.[2] Z. He and F. Zhang. Three-dimensional, position-sensitive radiation detection, August 12 2008. US Patent 7,411,197.[3] H. Chen, S. A. Awadalla, P. Marthandam, K. Iniewski, P. H. Lu, and G. Bindley. CZT device with improved sensitivity for medical imagingand homeland security applications. In
Society of Photo-Optical Instrumentation Engineers (SPIE) Conference Series , volume 7449, pages744902–744902, 2009.
4] L.J. Mitchell, BF Phlips, W.N. Johnson, E.A. Wulf, A.L. Hutcheson, CJ Lister, K.D. Bynum, B.E. Leas, and G. Guadagno. Mobile Imagingand Spectroscopic Threat Identification (MISTI): System overview. In
IEEE Nuclear Science Symposium Conference Record (NSS / MIC) ,pages 110–118. IEEE, 2009.[5] Daniel Wakeford, H.R. Andrews, E.T.H. Cli ff ord, Liqian Li, Nick Bray, Darren Locklin, Michael V. Hynes, Maurice Toolin, Bernard Harris,John McElroy, Mark Wallace, and Richard Lanza. The SORDS trimodal imager detector arrays. In Society of Photo-Optical InstrumentationEngineers (SPIE) Conference Series , volume 7310, pages 73100D–9–73100D–9, 2009.[6] M.V. Hynes, B. Harris, E.E. Lednum, M.S. Wallace, L.J. Schultz, D.M. Palmer, D.T. Wakeford, H.R. Andrews, R.C. Lanza, E.T.H. Cli ff ord,et al. Multimodal radiation imager, January 4 2011. US Patent 7,863,567.[7] K.P. Ziock, L. Fabris, D. Carr, J. Collins, M. Cunningham, F. Habte, T. Karnowski, and W. Marchant. A fieldable-prototype, large-area,gamma-ray imager for orphan source search. IEEE Transactions on Nuclear Science , 55(6):3643–3653, 2008.[8] P.R.B. Saull et al. First demonstration of a Compton gamma imager based on silicon photomultipliers.
Nuclear Instruments and Methods A ,679:89–96, 2012.[9] Y. Shao, K. Meadors, R.W. Silverman, R. Farrell, L. Cirignano, R. Grazioso, K.S. Shah, and S.R. Cherry. Dual APD array readout of LSOcrystals: Optimization of crystal surface treatment.
Nuclear Science, IEEE Transactions on , 49(3):649–654, 2002.[10] J.S. Huber, W.W. Moses, M.S. Andreaco, M. Loope, C.L. Melcher, and R. Nutt. Geometry and surface treatment dependence of the lightcollection from LSO crystals.
Nuclear Instruments and Methods A , 437(2):374–380, 1999.[11] J. N. Carter, P. Charalambous, A. J. Dean, J. B. Stephen, R. C. Butler, G. Di Cocco, E. Morelli, G. Spada, A. Spizzichino, L. Barbareschi,G. Boella, F. Perotti, G. Villa, M. Badiali, C. La Padula, F. V. Polcaro, and P. Ubertini. A position sensitive detector for a gamma-ray imagingtelescope.
Nuclear Instruments and Methods A , 196:477–482, 1982.[12] C. Labanti, E. Caroli, E. Rossi, and A. Spizzichino. Methods for optimizing the performances of position-sensitive CsI (Tl) scintillation barswith photodiode readout.
Nuclear Instruments and Methods A , 310(1):327–331, 1991.[13] L.E. Sinclair, D.S. Hanna, A.M.L. MacLeod, and P.R.B. Saull. Simulations of a scintillator Compton gamma imager for safety and security.
IEEE Transactions on Nuclear Science , 56(3):1262–1268, June 2009.[14] J.P. Martin and P.A. Amaudruz. A 48 channel pulse shape digitizer with DSP.
IEEE Transactions on Nuclear Science , 53(3):715 – 719, June2006.[15] S. Agostinelli, J. Allison, K. Amako, J. Apostolakis, H. Araujo, P. Arce, M. Asai, D. Axen, S. Banerjee, G. Barrand, et al. GEANT4 - asimulation toolkit.
Nuclear Instruments and Methods A , 506(3):250–303, 2003.[16] F. James and M. Roos. MINUIT - a system for function minimization and analysis of the parameter errors and correlations.
Computer PhysicsCommunications , 10(6):343–367, December 1975.[17] P. R. B. Saull, L. E. Sinclair, H. C. J. Seywerd, P. J. Boyle, A. M. L. MacLeod, and D. S. Hanna. A Two-Pixel Compton Imager. In
Societyof Photo-Optical Instrumentation Engineers (SPIE) Conference Series , volume 7665, April 2010.[18] L.E. Sinclair et al. Silicon photomultiplier-based Compton telescope for safety and security.
IEEE Nuclear Science Symposium ConferenceRecord (2013), Seoul, Korea; submitted to IEEE Transactions on Nuclear Science, January ..