Table Of ContentFULLARTICLEAPPEARSINIEEETRANS.NUCL.SCI.61:52745-2752,OCTOBER2014 1
Silicon photomultiplier-based Compton Telescope
for Safety and Security (SCoTSS)
Laurel Sinclair, Patrick Saull, David Hanna, Henry Seywerd, Audrey MacLeod and Patrick Boyle
6 Abstract—A Compton gamma imager has been developed awareofit.Alternately,itwouldallowinvestigationofascene
1 for use in consequence management operations and in security without disturbing it, providing personnel safety from loose
0 investigations.Theimagerusessolidinorganicscintillator,known
sources of radioactivity, and preserving evidence for eventual
2 forrobustperformanceinfieldsurveyconditions.Thedesignwas
n constrained in overall size by the requirement that it be person prosecution.
transportableandoperablefromavarietyofplatforms.Inorder Some earlier work in developing imagers for safety and
a
J to introduce minimal dead material in the path of the incom- security made use of solid-state devices such as high purity
ing and scattered gamma rays, custom silicon photomultipliers germanium [5]. For field operation, however, it would be
8
(SiPMs), with a thin glass substrate, were used to collect the
best to avoid the use of instruments which require cryogenic
scintillation light from the scatter layers. To move them out of
] cooling. Progress has been made to produce a Compton
t the path of thegamma rays, preamplification electronics for the
e siliconphotomultiplierswerelocatedadistancefromtheimager. imager using room temperature solid-state devices [6], and
d
This imager, the Silicon photomultiplier Compton Telescope for also there are other teams developingComptonimagersusing
-
s Safety and Security (SCoTSS) is able to provide a one-degree scintillator[7],[8].These,however,donothavethesensitivity
n image resolution in a ±45◦ field of view for a 10 mCi point
werequire.Notableparalleleffortshavetakenplacetodevelop
i source 40 m distant, within about one minute, for gamma-ray
. highly sensitive fieldable gamma imagers [9]–[11]. These are
s energies ranging from 344 keV to 1274 keV. Here, we present a
c comprehensive performance study of the SCoTSS imager. not,however,suitabletoourapplicationastheirverylargesize
i and weight necessitates operation from a particular dedicated
s Index Terms—Compton imager, gamma imager, Compton
y survey platform.
telescope, gamma camera, Compton camera, SiPM, silicon pho-
h tomultiplier. The Compton gamma-imaging technique is optimal for
p
mobile methods, in that as opposed to coded-aperture or
[
pinhole imaging, it does not require the transport of a heavy
1 I. INTRODUCTION gamma-ray shield. Moreover, the recent development of very
v CANADIAN federal radiological assessment team part- small and low-mass silicon photomultipliers (SiPMs) [12]–
9
ners, as well as border security operators, have years [14] permits the collection of light from within a layered
7
9 of experience using large-volume inorganic scintillators to Compton gamma imager, with negligible scattering induced
1 detectradioactivesubstancesfroma distance via their gamma in the dead material of the light collection device [15].
0 emissions,inoutdoormobilesurveyconditions[1]–[4].These We have designed and built an imager for use by nuclear
.
1 groupswouldbenefitenormouslyfromtheuseofamobilede- emergencyresponse and security teams, using solid inorganic
0 vice capable of providingan image of a radioactivesubstance scintillator with SiPM light collection. This paper presents
6 overlaid on a photograph of the field of view. Operated in a the first performance results for this detector, the Silicon
1
non-imaging total-count mode, an imager having a sensitive photomultiplier Compton Telescope for Safety and Security
:
v volume of the order of several litres would provide the high (SCoTSS).
i sensitivity and fast detection times that mobile survey crews
X
are used to, with the option of switching to imaging mode
ar once a radiation field of interest has been detected. II. COMPTON IMAGING
The imaging modality would bring the obvious benefit of A Compton imager relies on measurement of the energy
providing a faster source localization than could be provided and position of a Compton scatter, E1 and x1, and on the
by a non-directional detector operating in a raster pattern. measurement of the energy and position of the absorption of
An imager is capable of localizing multiple or distributed the scattered gamma ray, E2 and x2, in order to reconstruct
sources, which would give omnidirectional detectors a prob- theoriginofanincominggammaray[16].AgenericCompton
lem. Additionally, it could allow for localization of a source imager,as illustratedin Fig. 1, isthuscomposedof two parts,
without a potentially hostile party under investigation being a scatter detector at the front, in which the Compton scatter
should occur, and an absorber detector behind that, to collect
L.E.Sinclair,H.C.J.SeywerdandA.M.L.MacLeodarewiththeGeological
the scattered gamma ray. For simple events consisting of a
SurveyofCanada,NaturalResourcesCanada,601BoothSt,Ottawa,Ontario,
K1A0E8,Canada. e-mail: [email protected]. single Compton scatter followed by photoabsorption of the
P.R.B.SaulliswithMeasurementScienceandStandards,NationalResearch scattered gamma ray, the scatter angle, θC, is given by the
Council, 1200Montreal Rd,Ottawa, Ontario, K1A0R6,Canada.
expression [17]
D.S.HannaiswiththePhysicsDepartment,McGillUniversity,3600Uni-
versitySt,Montreal, Quebec,H3A2T8,Canada.
2
Manuscript received Dec20,2013. cosθC =1+mec (1/Etot 1/E2), (1)
−
FULLARTICLEAPPEARSINIEEETRANS.NUCL.SCI.61:52745-2752,OCTOBER2014 2
3) Light Collection: We chose to use SiPMs for light col-
lectionin the scatter layersbecause the SiPMs couldbe made
extremely thin and low mass such that they would present
minimal dead material to the Compton-scattered gamma ray.
The SiPMs were manufactured by SensL [20] on a glass
substrateonly0.5mmthick.AnindividualSiPMmodule,used
to read outa single CsI(Tl) crystal,is shown in Fig. 2 c). The
SiPM module consists of 16 pads, a total of 76384 Geiger-
mode avalanche photodiodes,which are summed to providea
single analogue signal from the 1.35 x 1.35 cm2 active area
of the SiPM. The bias supply and pre-amplification circuits
were located on front-end electronics boards a distance from
thegammaimagerandthereforeoutofthepathoftheincident
and scattered gamma rays (see Fig. 2 d)).
Fig. 1. Generic Compton imager showing aCompton scatter with energy
We used conventionalphotomultipliertubes(PMTs) forthe
E1 at position x1 in the scatter detector, and the subsequent absorption of
the scattered gamma ray leading to energy deposit E2 at position x2 in lightcollectionfromthepixelsoftheabsorberdetector.SiPMs
theabsorber detector. Thecone-shaped locus ofpossible sourcelocations as could have been used for the absorber detector as well, but
reconstructed fromtheenergydeposits isalsoshown.
dead material is not a problem at the back of the absorber
detector, and at the time of parts procurement for the pro-
whereEtot =E1+E2, mec2 isthe electronrestmass, andthe totype imager, PMTs were better proven and less expensive.
approximation has been made that the electron from which The individualaluminum-cladNaI(Tl) + PMT assemblies are
the incoming gamma ray scatters is unbound and at rest. clearly visible in the rear plane in Fig. 2 a).
Thus, the position of the emitter can be reconstructed up to 4) Scintillating Material: Despite the large initial-state
an arbitrary azimuthal angle, to lie on the surface of a cone electron effects on image resolution of high-Z scintillator
with axis along the line x1 x2, and opening angle θC. By materials, materials like NaI(Tl), CsI(Tl) and LaBr3(Ce) still
−
overlaying several such cones, the location or distribution of perform better than organic scintillator due to their much
the source of gamma rays may be determined. Note that the superior energy resolution [19]. The SensL SiPM which we
image resolution of a Compton imager is tied intrinsically to use has a peak light sensitivity at 480 nm, and therefore
theimager’senergyresolution–meaningthatagoodCompton CsI(Tl), with its relatively high peak wavelength of emission,
imager is also a good spectroscopicinstrumentwhich may be was chosen for the material of the scatter detector. The
usedtoidentifytheisotopiccompositionofsourceswhichare mobile survey workhorse, NaI(Tl), which is well matched in
a priori unknown. peak wavelength of light emission to standard PMTs, was
selected for the rear detector. Our aim has been to develop
III. THEIMAGER a prototype for eventual commercialization and adoption by
Anextensivedesignphasehasbeenconductedtodetermine nuclear emergency response teams and therefore, although
the optimal imager materials and geometry [18], [19]. We an earlier study had determined that its performance was
considered the following parameters: optimal [19], LaBr3(Ce) was ruled out for either component
1) TotalSize: Theimagertotalweightandexteriorvolume due to price constraints.
were constrained by the requirement that teams composed of 5) Detector Thickness: As the thickness of the scatter
one or two people should be able to lift the imager from a detector increases, the probability for a Compton scatter to
transport container into a survey vehicle or platform, without be initiated increases, but at the same time the probabilityfor
machineassistance.Theimager,picturedinFig.2a),wasthus a second scatter, or photoabsorptionevent, also increases. We
constrained to occupy a cubic volume approximately 35 cm have optimized the total thickness of the scatter detector by
on a side, with a total mass of about 25 kg. choosingthethicknesswhichmaximizestheprobabilityforan
2) Sensitive Volume: Within the constraintson the exterior incominggammarayofenergyaround1MeVtoscatteronce,
dimensions,theimagershouldhaveaslargeasensitivevolume and then exit the scatter detector, according to the approach
as possible. This increases sensitivity to weak sources. Our described in [19].
scatter detector consists of two 9 x 9 layers of cubic pixels To obtaina better measurementofthe depthofthe position
of CsI(Tl) 1.35 cm on a side. This crystal arrangement in an oftheComptonscatterwithinthetotalthicknessofthescatter
openscatterlayerisshowninFig.2b).Theabsorberdetector detector, we have segmented the scatter detector into two
consists of a 10 x 10 layer of 2.5 x 2.5 x 4.0 cm3 pixels of layers.Thesemaybe seeninFig. 2. Inthiswork,eventswere
NaI(Tl). Thus, the total scintillator volume of the imager is selected with one energy deposit in the scatter detector, in
2898 cm3. This is comparableto the total scintillator volume either layer. In the future, events featuring Compton scatters
of current standard omnidirectional mobile survey systems in both scatter layers, and also low energy events where the
(10 x 10 x 41 cm3). Thus, second-by-second the imager photoabsorption process occurs in the second scatter layer,
provides the level of minimum detectable activity to which could be utilized.
nuclear emergency response teams are currently accustomed, For the absorberdetector,efficiency increaseswith increas-
even without using it in an imaging mode. ing thickness, but since we do not have a depth measurement
FULLARTICLEAPPEARSINIEEETRANS.NUCL.SCI.61:52745-2752,OCTOBER2014 3
a) b)
c) d)
Fig. 2. a) The imager with the two scatter layers on the right and the absorber detector behind them at the left. The 10 x 10 array of aluminum-clad
PMT+NaI(Tl)assembliesisvisibleattherearoftheimager,clampedinanaluminumframe.Thetwoscatterlayersareshownenclosedintheircarbon-fibre
andDelrinframe,withmicro-coaxcablesexitingthelayerstotheleft.TheVelmextranslatorstowhichthescatterlayersareaffixedforadjustingtheinter-layer
spacing, can be seen underneath the scatter layers. A “red dot” alignment sight is visible on top of the absorber detector. b) The inside of a scatter layer
showing the eighty-one CsI(Tl) crystals wrapped with thread-seal (“Teflon”) tape. Foam spacers have been removed from the three right-most columns of
pixelstoshowthemicrocoaxcablesandconnections totheglassoftheSiPMs.c)AsingleSiPMdetectorunitusedforreadingoutasingleCsI(Tl)crystal.
d)Thebiasdistribution andpre-amplification boards fortheSiPM+CsI(Tl)channels, connected tothemotherboard.
in the absorber detector, there is increasing uncertainty on our imager are affixed to Velmex XSlides [23] to allow for
the depth of the energy deposit with increasing crystal depth, varying the inter-layer distance according to experimental or
leading to a loss of image resolution. We used EGSnrc [21], operational need. Monte Carlo simulation studies determined
[22]simulationstudiestobalancetheconflictingrequirements thatforthelargesourcedistancesunderstudyhere,thetimeto
of high absorption probabilityand good position precision by achieveanimageofacertainprecisiondroppedasthedistance
choosing the depth of the absorber detector which gives the between the rear scatter layer and the absorber detector was
desired image precision in the shortest time [19]. increased, up to about 14 cm, beyond which no significant
improvement or worsening was seen. For the measurements
6) Pixel Size: A smaller lateral physical dimension of the
presentedhere,thetwoscatterlayersarecentredlaterallywith
pixelsinboththescatterandabsorberdetectorswouldprovide
respecttotheabsorberdetector,andthedistancesbetweenthe
a better measurement of the positions of the energy deposits
centres of the scatter layers and the front face of the absorber
and therefore improved measurement of the cone axes and,
layer are fixed at 14.6 cm and 23.9 cm.
ultimately, improved image resolution. However, making the
pixels too small would increase the cost due to additional 8) Electronics: For the purposes of detector performance
readout channels. We have optimized the performance versus optimizationandcharacterization,powersupplyandreadoutof
cost by choosing the lateral pixelization to contribute approx- theimagerhasbeenaccomplishedwithstandardmulti-purpose
imately the same image smearing as is contributed by initial- crate-based electronics. For the imager to be eventually field-
state electron effects and finite energy resolution, following able from a variety of platforms, custom electronics with a
the approach described in [19]. suitable small and rugged form will have to be developed.
7) Layer Spacing: Increasing the separation between the Inthefinalconfigurationforthisstudy,thedetectorfeatures
scatterdetectorandtheabsorberdetectorinaComptongamma two scatter layers, each with a 9 x 9 arrangement of cubic
imager will result in improvedangularresolution, at a cost of pixels of CsI(Tl), 1.35 cm on a side, as shown in Fig. 2 b).
lowered efficiency and narrowed field of view. We chose not The CsI(Tl) crystals are connected to the SiPM with NyoGel
to set the inter-layer distance at a fixed separation, optimized OC-462 optical gel [24], and then this unit is wrapped in
for a particular application. Instead, both scatter layers of thread-sealtape(commonlyreferredtoasTeflontape).Within
FULLARTICLEAPPEARSINIEEETRANS.NUCL.SCI.61:52745-2752,OCTOBER2014 4
a layer, the pixels are grouped into nine 3 x 3 arrangements, pulses from the scatter and absorber detector, as read out by
to allow for cable routing. The distance from the centre of the digitizers, where it can be seen that CsI(Tl) has a much
one pixel to the next within one 3 x 3 group is 2 cm. The longer decay time than NaI(Tl). Triggering is performed by
distancefromthecentreofone3x3grouptothenextis7cm. the CAEN digitizers in conjunction with a CAEN V1495
Mechanical support for the scatter layers is provided by low- general purpose VME logic board. We trigger solely on the
density foam on 0.8 mm-thick carbon-fibre sheets mounted rise of the pulses in the absorber detector (NaI(Tl) + PMT)
in a Delrin frame. Custom front-end boards, shown in Fig. 2 channels.Whenevera NaI(Tl)+ PMT channelsamplecrosses
d), provide both the SiPM bias voltage and pre-amplification. a user-defined ADC threshold, a local trigger is issued. An
Further details are available in [15]. OR of these local triggers is provided at the front panel of
The 2.5 x 2.5 x 4 cm3 NaI(Tl) crystals of the absorber each of the two PMT V1740 boards as a NIM signal. These
detector are read out with Hamamatsu [25] R1924A PMTs. two signals are fed to the V1495, where they are ORed and
Proteus [26] provided custom-designed assemblies with outer fanned out to the external trigger inputs of all five boards in
dimension 3.2 x 3.2 x 15.2 cm3, as small as possible in order theDAQsystem.Thus,theexternaltriggerinitiatestheglobal
to minimize the dead space between the individual NaI(Tl) storage of an event in VME memory across all five boards.
crystals. An aluminum frame provides compression to pack The board clocks are carefully synchronizedso that the event
the NaI(Tl) + PMT assemblies tightly in a 10 x 10 array.The timestamps for coincident data agree. These timestamps are
high voltage of the PMTs is supplied using a CAEN [27] HV monitoredduring data taking to ensure that the boards do not
1535SNunitandthePMThighvoltagesaretrimmedsuchthat fall out of synchronization (not observed during any of the
each channel reconstructs the 662 keV photopeak of 137Cs at runs described here). Note that the local trigger threshold can
the same pulse height. only be set for groupsof eightchannelson the V1740which,
in conjunction with a residual channel-to-channel baseline
IV. DATA ACQUISITION variation, leads to an effective spread in the trigger threshold
across all channels of about 20 keV. The trigger threshold
We investigated the performance of the imager by placing
∼
a 137Cs point source of known strength at various angles θ fordatatakingcorrespondstoapulseenergyofapproximately
200 keV for all runs except 152Eu, where it was lowered to
with respect to the symmetry axis of the detector. We used
several different isotopes, 152Eu, 22Na, 137Cs, and 54Mn, and about 50 keV.
To keep the data rate within the VME system limit (40
were thus able to investigate the detector performance as
MB/s), thepulsesampleswereintegratedonboardtheV1740
a function of gamma-ray energy as well. The detector has
digitizersusingcustomfirmwaresuppliedbyCAEN,andonly
been configured for optimal performance at long range, and
the pulse areas read out. For each channel on a board, the
thereforewherepossiblethesourceswereplacedatadistance
firmware provides a simple sum of the digitized samples
of 10 m from the centre of the front face of the detector. The
152Eu and 54Mn sources were weaker so they were placed at within a desired integration window. The user defines this
window by specifying the offset time with respect to the
5 m from the centre of the front face. (In the investigation
trigger and the number of samples over which to integrate.
in Section VI-B of how long it takes to achieve an image
These parameters are common to all channels on a board,
of a desired precision, we scale the results such that all of
so it was necessary to split the readout into two boards of
the sources are effectively of the same strength at the same
NaI(Tl) + PMT and three boardsof CsI(Tl) + SiPM channels
distance.) The run configurations for the data sets presented
to account for the different decay times. The firmware also
in this study are shown in Table I.
provides a method for channel-by-channel online baseline
subtraction,sothattheresultingsamplesumsareproportional
to pulse area and therefore energy.
CsI(Tl) with SiPM ThepulseintegrationwindowfortheCsI(Tl)+SiPM chan-
2200
NaI(Tl) with PMT nelsischosentocoincidewithwhereacoincidentpulsewould
s
nt fall with respect to the PMT pulse (see Fig. 3). Consequently,
u
o although we do not trigger directly on the CsI(Tl) + SiPM
C2100 channels,iftherearepulsesfromtheseboardscoincidentwith
C
the PMT trigger their energies are correctly reconstructed.
D
A For each triggered event, sample-sums from all five boards
2000 arereadouttoLinuxworkstationmemory.Tolimittheamount
of data written to disk, the sample-sums from only those
0 100 200 300 400 channelswithenergygreaterthanthreestandard-deviationsof
Sample Number
the pedestal (typically 10 keV for PMT energies and 25 keV
for the SiPM energies) are written to disk for further offline
Fig. 3. Coincident traces, digitized with a 16 ns sampling rate, from a
scatter andanabsorberpixel.Vertical lines showtheintegration windows. processing.Ourtriggerratefornaturallyoccuringbackground
radiation,forthe200keVtriggerthreshold,isabout1100Hz.
The signals from all channels are digitized every 16 ns This increased to at most 2200 Hz for runs with a source
usingCAENV174064-channeldigitizers,withtwoboardsfor present,correspondingtoatmost1%deadtime.Forthe152Eu
the PMTs and three for the SiPMs. Fig. 3 shows coincident run at lower threshold, the trigger rate increased to 6700 Hz
FULLARTICLEAPPEARSINIEEETRANS.NUCL.SCI.61:52745-2752,OCTOBER2014 5
TABLEI
DATASETS
Source
Run Distance θ Livetime Numberoftriggered events
Isotope Energy Emissionprobability Rate
(keV) (%) (s−1) (m) (◦) (s)
1 137Cs 662 85.0 2.1×107 10.0 0.0 11236 1.8×107
2 137Cs 662 85.0 2.1×107 10.0 10.0 11218 1.8×107
3 137Cs 662 85.0 2.1×107 10.0 20.0 11119 1.8×107
4 137Cs 662 85.0 2.1×107 10.0 30.0 10968 1.8×107
5 137Cs 662 85.0 2.1×107 10.0 40.0 11067 1.8×107
6 152Eu 344 27.7 1.6×106 5.0 20.0 17817 1.2×108
7 22Na 511 180.8 2.6×107 10.0 20.0 10524 2.3×107
8 54Mn 835 100.0 4.3×106 5.0 20.0 40667 6.0×107
7 22Na 1274 100.0 1.6×107 10.0 20.0 10524 2.3×107
(3%dead-time).Noeffectsduetopulsepile-upwereobserved layer), is apparentin both plots, tightly clustered at particular
in the data. values of E and E . For example, for the 137Cs run, the
scat abs
Note thatourtriggerdoesnotrequirea coincidencein time full energy deposition back-scatter events all occur at E
scat
∼
between the rising edges of the scatter and absorber pulses. 200 keV and E 460keV. This kinematic tendencymakes
abs
∼
Instead, whatever energy is reconstructed in the scatter layer the rejection of back-scatter events straightforward. Isotope-
integration window is considered to be due to a coincident dependent energy requirements were applied, as shown in
pulse, unless the energy deposit fails the threshold cut, or the Table II.
event fails the selection cuts. This is done intentionally, to
maintain the sensitivity of our detector when it is functioning VI. RESULTS
inagross-count(non-imaging)modeasasurveyspectrometer.
A. Efficiency and Angular Resolution
Energy calibration of the detector is carried out using a
varietyofgammasourceswithknownenergies.Forthescatter We define efficiency as the number of events which pass
detector, the energy scale is parametrized linearly using the the event selection minus the number of naturally occuring
662 keV peak of 137Cs and the zero-energy pedestal. For the background events, divided by the total number of gamma
absorberdetector,theenergyscaleisparametrizedwithathird- rays crossing the 20 x 20 cm2 area of the first scatter layer
degree polynomial using emissions over the range 40 keV to during the run. The efficiency measurements are presented
2614 keV from 241Am, 113Sn, 152Eu, 22Na, 137Cs, 40K and in Table III. The dominant uncertainties in the measurement
208Tl. We achieve mean energy resolutions of 7.9% (29%) of efficiency are the uncertainty on the measurement of the
and7.5%(23%)at662keV(60keV)forCsI(Tl)+SiPM and source emission rate and the uncertainty on the background
NaI(Tl) + PMT channels, respectively. subtraction. The efficiency does not vary strongly with angle,
and varies slightly with energy, with the detector being more
Datatakingwascarriedoutafterthesetuphadbeenwarmed
up over several days in a lab having stable temperature. We efficient at lower energies.
found that drifts in the response of the CsI(Tl) + SiPM The Angular Resolution Measure (ARM) is defined as
channels could be limited to about 1% through daily re- θCR − θgeom, where θCR is the Compton scatter angle as it is
calibrations with a 137Cs source. For the NaI(Tl) + PMT reconstructed from the measured energy deposits and θgeom
channels,ahighvoltagetrimonamonthlybasiswassufficient is the angle between the line connecting the two energy
to accomplish the same. deposits, and the line between the first energy deposit and
the source [28]. The ARM distribution, for Run 3 with the
137Cs source at 20◦, is shown in Fig. 5. The distribution
V. EVENTSELECTION
showstwocomponents.Thewell-reconstructedforward-going
Only events with exactly one energy deposit in each of scatters fall in the Gaussian peaked part of the distribution.
the scatter and absorber detectors were considered for further There is also a broad pedestal extending to negative values
analysis. The measured scatter energy, E , is shown versus due to contributionsfrom incompletelycontained background
scat
the measured absorber energy, E , for two-hit events for radiation, random overlap between different events, poorly-
abs
Run 3, 137Cs at 20◦, in Fig. 4a) and for Run 7, 22Na at reconsructed events, and back-scatter events that pass the
20◦, in Fig. 4b). Full-energy deposition events with E and selection cuts.
scat
E summing to the photopeak energies are clearly seen as The ARM distributions for all runs have been fit to the
abs
diagonal lines with intercepts at 662 keV for 137Cs and at sumof a Gaussian distributionanda third-degreepolynomial,
511 keV and 1274 keV for 22Na. For image reconstruction, as indicated for Run 3 in Fig. 5. The sigma values from
we select events where the energy sum of the two hits is the Gaussian fits are presented in Table III. We find that
within abouttwo standard deviationsof the photopeakenergy the angularresolution of the imager improveswith increasing
of interest; see Table II. energy, ranging from about 4.7◦ to 2.8◦.
A class of full energy deposition events, the back-scatter Efficiency and angular resolution are quality measures
events (scatter in the absorber layer, absorption in the scatter which can be used to compare the performance of different
FULLARTICLEAPPEARSINIEEETRANS.NUCL.SCI.61:52745-2752,OCTOBER2014 6
a) b)
Fig.4. a)ScatterenergyversusabsorberenergyforRun3,137Cssourceat20◦.b)ScatterenergyversusabsorberenergyforRun7,22Nasourceat20◦.
TABLEII
EVENTSELECTION
Run Id Etot (keV) Photopeak selection Back-scatter rejection
1–5 137Cs 662 |Escat +Eabs -Etot|<40keV Escat<170keV
6 152Eu 344 |Escat +Eabs -Etot|<33keV Escat<110keV
7 22Na 511 |Escat +Eabs -Etot|<40keV Escat<160keV
8 54Mn 835 |Escat +Eabs -Etot|<45keV Escat<170keV
7 22Na 1274 |Escat +Eabs -Etot|<80keV Escat<170keV or280keV<Escat<635keV
TABLEIII
ANGULARRESOLUTIONMEASUREANDEFFICIENCY
6000
Run Id Energy[keV] θ (◦) ARM(◦) Efficiency (%)
1 137Cs 662 0.0 3.5±0.1 0.59±0.11 5000
32 113377CCss 666622 2100..00 33..56±±00..11 00..6652±±00..1121 bin4000
r
54 113377CCss 666622 4300..00 33..65±±00..11 00..6612±±00..1111 pe3000
N
6 152Eu 344 20.0 4.7±0.1 0.78±0.21 2000
7 22Na 511 20.0 4.0±0.1 0.70±0.13
8 54Mn 835 20.0 3.1±0.1 0.41±0.06 1000
7 22Na 1274 20.0 2.8±0.1 0.40±0.07
0
-60 -40 -20 0 20 40 60
degrees
designs. However, it often happens that varying a certain
design parameter can be beneficial to efficiency, and harmful Fig.5. AngularresolutionmeasureforRun3,137Cssourceat20◦,shown
as the solid black histogram. A fit to the data of the sum of a Gaussian
toangularresolution,orviceversa.Therefore,tooptimizeour
andathird-degree polynomial is shownas thesolid curve. Thethird-degree
design, and to compare it with other designs, we make use of polynomialcomponentisshownasthedashedcurve.TheGaussianfithasa
a quantitycalled “time to image”, defined in the nextsection. sigmaof3.5◦.
B. Time to image
A simple back-projection of the Compton cones into space algorithm used for the current analysis, which differs slightly
can provide an image of the emitter with angular precision fromtheversionin thatreference.We notethatthisalgorithm
commensurate with the angular resolution. However, this functions to locate a point source, providing a quantitative
method fails to make the best use of the available statistics measure of the detector’s performance, which we have used
andofthedifferinguncertaintiesaffectingeachback-projected to optimise our design. For operational use, we intend to
cone. We have developeda χ2 minimization algorithm,based implement other imaging modalities which will be applicable
on MINUIT [29], for finding the source direction which best for multiple or extended sources.
satisfiesalltheComptonconesinasampleofN events,taking For the first iteration, a starting seed direction for the fit
into account the uncertainties on cone opening angle and is found by back-projecting all cones onto a 2-D histogram
axis. The algorithm is iterative and allows for the rejection of with4◦ 4◦ binningandselectingthedirectioncorresponding
×
eventswithpoorlyfitting cones,suchasthoseassociated with to the most populated bin. All events in the sample which
naturally occuring background events, back-scatter events, have cone surfaces coming within five standard deviations of
eventswhichcontaina noise hit, oreventswith partialenergy angular separation from this seed direction are accepted in
deposit. Details can be foundin [18]. Here we summarise the the χ2 minimizationstep, whichgeneratesa newdirection.In
FULLARTICLEAPPEARSINIEEETRANS.NUCL.SCI.61:52745-2752,OCTOBER2014 7
the second iteration, all cones in the full sample of N events which has proven performance in outdoor conditions, and
coming within two standard deviations of the fitted direction readoutwithcustom-madelow-masssiliconphotomultipliers.
are consideredin the fitand a new directionis generated.The The imager has an intrinsic angular resolution of about four
third iteration is a repeat of the second, to allow more cones degrees. Image precisions of one degree in a 45◦ field of
±
to be included in this final fit using the updated direction. view can be obtained for a 10 mCi point source at 40 m
Fig. 6 a) shows an image reconstructed in this way, for within about one minute, for gamma-ray energies ranging
four seconds of data from Run 3 with the 137Cs source at from 344 keV to 1274 keV. Two-degree image precisions are
20◦. The one-, two- and three-sigma confidence intervals are obtainable within 20 seconds.
indicatedbythe colouredcontours.Fig. 6 b)showsthe image This achievement has been obtained with early genera-
reconstructed from the same run, using 60 seconds of data. tion SiPMs. Further improvements in efficiency are to be
The localization of the gamma emitter is much more precise. expected with future improvements in SiPM technology as
To quantify the improvement in precision with time, we noise levels are reduced and it becomes possible to lower
look at the root mean square spread of reconstructed image energythresholds.Afuture,fieldableversionofthisinstrument
directions among datasets of a certain acquisition time, for would have to be environmentally isolated and make use of
different acquisition times. This image precision is shown the individual bias settings featured on the SiPMs to control
versus acquisition time in Fig. 7, where the results have been gain fluctuations due to temperature. Warm up time for the
scaled to reflect a 10 mCi source with 100% branching ratio, instrument would likely be around 15 minutes and a price
at 40 m. We find that angular precision improves with time, point in the low 100 k US$’s should be feasible. The total
going approximately as 1/√t where t is the aquisition time, volume of scintillator and corresponding second-by-second
asexpectedfromcountingstatistics.Theangularprecisionfor isotope identification ability will allow the fieldable version
thisconfigurationreachesonedegreeinjustunderoneminute of this instrument to function as a drop-in replacement for
– indicated in the graph by a filled dot with horizontal error the current ubiquitous non-directional NaI(Tl) mobile survey
bars. spectrometers.
By taking the intercept of the precision versus acquisition
ACKNOWLEDGMENTS
time plot with horizontal lines of various desired precision
values, we can look at how time to image varies for different This work was supported by Canada’s Centre for Secu-
experimental configurations. In Fig. 8 a) we show time to rity Science, project CRTI 07-0193RD. The authors thank
image versus the angle between the line joining the centre A.W. Saull for assistance with graphics. This is NRCan/ESS
of the detector’s front face and the source location, and the Contribution number 20130419.
symmetry axis of the detector. We find that generally the
REFERENCES
imagerisabletoachieveonedegreeofimageprecisionwithin
aboutaminute,acrossa 45◦ fieldofview.Manyfeaturesof [1] L. E. Sinclair, H. C. J. Seywerd, R. Fortin, J. M. Carson, P. R. B.
± Saull, M.J.Coyle, R.A.VanBrabant, J.L.Buckle, S.M.Desjardins,
interestat40m willbe separatedbytwo degreesor more,for
and R. M. Hall, “Aerial measurement of radioxenon concentration off
example the distance between two windows of a building, as the west coast of Vancouver Island following the Fukushima reactor
viewedfromaroad.Ifthisisthedesiredprecision,theimager accident,” J. Environ. Radioactiv., vol. 102, no. 11, pp. 1018 – 1023,
2011.
isabletoproduceausefulresultinlessthan20seconds,across
[2] R.L.Grasty,“Environmental monitoring byairborne gammarayspec-
the entire field of view. trometry, experience atthe Geological SurveyofCanada,” Application
Fig.8b)showsthetimestoimageversusphotopeakenergy, of uranium exploration data and techniques in environmental studies,
IAEA-TECDOC-827,pp.93–101,1995.
for a source 20◦ off-axis, where the measurements have been
[3] “Airbornegammarayspectrometersurveying,”IAEATechnicalReports
scaled to correspond to an isotope of 10 mCi strength, with Series 323,1991.
100% branching ratio. We find that the imager is able to [4] Q.Bristow,“Theapplicationofairbornegamma-rayspectrometryinthe
search for radioactive debris from the Russian satellite COSMOS 954
reach one degree of precision within about a minute to a
(Operation ‘Morning Light’),” Geol. Surv. Can. Paper, no. 78-1B, pp.
minute and a half, depending on the isotope energy. The 151–162,1978.
[5] K. Vetter, M. Burks, C. Cork, M. Cunningham, D. Chivers, E. Hull,
imager geometry is optimized for higher-energy sources, but
T. Krings, H. Manini, L. Mihailescu, K. Nelson, D. Protic, J. Valen-
nevertheless performs adequately down to 344 keV. Future
tine,andD.Wright,“High-sensitivity Comptonimagingwithposition-
workcouldinvestigatethe imagerresponseto lower orhigher sensitive Si and Ge detectors,” Nucl. Instr. and Meth. A, vol. 64, pp.
energies of photopeak emission – or to a highly shielded 1648–1654,2006.
[6] C. Lehner, Z. He, and F. Zhang, “4 π Compton imaging using a 3-
source producing an indistinct photopeak. Times to image
Dposition-sensitive CdZnTedetectorviaweightedlist-modemaximum
of better than 20 seconds are achieved across the energy likelihood,” IEEE Trans. Nucl. Sci., vol. 51, no. 4, pp. 1618 – 1624,
range from 344 keV to 1274 keV for an image precision 2004.
[7] W.LeeandT.Lee,“AcompactComptoncamerausingscintillatorsfor
of two degrees which will be adequate for many operational
the investigation of nuclear materials,” Nucl. Instr. and Meth. A, vol.
circumstances. 624,no.1,pp.118–124,2010.
[8] J. Y. Yeom, S. Yamamoto, S. Derenzo, V. Spanoudaki, K. Kamada,
T. Endo, and C. Levin, “First Performance Results of Ce:GAGG
VII. CONCLUSIONS ScintillationCrystalsWithSiliconPhotomultipliers,”IEEETrans.Nucl.
Sci.,vol.60,no.2,pp.988–992,2013.
A Compton gamma imager for use in nuclear security and [9] K. P. Ziock, L. Fabris, D. Carr, J. Collins, M. Cunningham, F. Habte,
T. Karnowski, and W. Marchant, “A fieldable-prototype, large-area,
emergency response, has been designed, built and charac-
gamma-ray imager for orphan source search,” IEEE Trans. Nucl. Sci.,
terized. The imager is based on solid inorganic scintillator, vol.55,no.6,pp.3643–3653,2008.
FULLARTICLEAPPEARSINIEEETRANS.NUCL.SCI.61:52745-2752,OCTOBER2014 8
a) b)
Fig.6. a)Imagereconstructed usingfoursecondsofdatafromRun3,137Cssourceat20◦.b)Imagereconstructed using60secondsofdatafromRun3,
137Cssourceat20◦.Thelinesindicatetheone-,two-andthree-sigmaconfidenceintervals.Thetruesourcelocationisindicatedintheleftfigurebyablack
cross.
security,” IEEE Trans. Nucl. Sci., vol. 56 Issue: 3, pp. 1262 – 1268,
2009.
) 3.5
°( [20] Silicon photomultipliers are available from SensL Technologies Ltd,
n 3 Cork,Ireland.
o [21] I. Kawrakow, E. Mainegra-Hing, D. W. O. Rogers, F. Tessier, and
cisi 2.5 B.R.B.Walters,“TheEGSnrcCodeSystem,NRCReportPIRS-701,”
e 2 2011.
pr [22] I.Kawrakow,E.Mainegra-Hing, F.Tessier,andB.R.B.Walters,“The
r 1.5 EGSnrcC++classlibrary, NRCReportPIRS-898,”2011.
a [23] Velmexslides areavailable fromVelmexInc.Bloomfield, NY,USA.
ul 1 [24] Optical gelisavailable fromNyeLubricants, Fairhaven, MA,USA.
g
n [25] Photomultipler tubes are available from Hamamatsu Photonics K.K.,
A 0.5 HamamatsuCity,Japan.
0 [26] Customscintillator+photomultipliertubeassemblieswereprovidedby
0 20 40 60 80 100
ProteusInc.,ChagrinFalls,OH,USA.
Data acquisition time (s)
[27] Digitizers and high voltage supplies are available from CAEN S.p.A.,
Viareggio, Italy.
Fig.7. PrecisionversusacquisitiontimeforRun3,137Cssourceat20◦.The [28] V. Schoenfelder, H. Aarts, K. Bennett, H. De Boer, J. Clear, W. Coll-
acquisition timerequiredfortheprecisiontodroptoonedegreeisindicated mar, A. Connors, A. Deerenberg, R. Diehl, A. Von Dordrecht et al.,
bythefilleddotwithhorizontal errorbars. “Instrument description and performance of the imaging gamma-ray
telescope COMPTEL aboard the Compton Gamma-Ray Observatory,”
TheAstrophys.J.Suppl.Ser.,vol.86,pp.657–692,1993.
[29] F. James and M. Roos, “Minuit: A system for function minimization
[10] D. Wakeford, H. Andrews, E. Clifford, L. Li, N. Bray, and D. Lock- and analysis of the parameter errors and correlations,” Comput. Phys.
lin, “The SORDS trimodal imager detector arrays,” Proc. SPIE 7310 Commun.,vol.10,pp.343–367,1975.
73100D,2009.
[11] L. Mitchell, B. F. Phlips, W. Johnson, E. A. Wulf, A. Hutcheson,
C. Lister, K. Bynum, B. Leas, and G. Guadagno, “Mobile Imaging
and Spectroscopic Threat Identification (MISTI): system overview,” in
NuclearScienceSymposiumConferenceRecord,Oct2009,pp.110–118.
[12] P. Eckert, H.-C. Shultz-Coulon, W. Shen, R. Stamen, and A. Tadday,
“Characterisation studies of silicon photomultipliers,” Nucl. Instr. and
Meth.A,vol.620,no.2-3,pp.217–226,2010.
[13] N. Otte, B. Dolgoshein, J. Hose, S. Klemin, E. Lorenz, R. Mirzoyan,
E.Popova, andM.Teshima, “TheSiPM–Anew PhotonDetector for
PET,”Nucl.Phys.B–Proc.Supp.,vol.150,pp.417–420,2006.
[14] V. D. Kovaltchouk, G. J. Lolos, Z. Papandreou, and K. Wolbaum,
“Comparison of asilicon photomultiplier to a traditional vacuum pho-
tomultiplier,” Nucl. Instr. andMeth. A,vol. 538, Issues 1-3, pp. 408 –
415,2005.
[15] P.R.B.Saull,L.E.Sinclair,H.C.J.Seywerd,D.S.Hanna,P.J.Boyle,
and A. M. L. MacLeod, “First demonstration of a Compton gamma
imagerbasedonsiliconphotomultipliers,”Nucl.Instr.andMeth.A,vol.
679,pp.89–96,2012.
[16] R.W.Todd,J.M.Nightingale,andD.B.Everett,“Aproposedgamma
camera,” Nature,vol.251,pp.132–134,1974.
[17] A.H.Compton,“AquantumtheoryofthexcatteringofX-raysbylight
elements,” Phys.Rev.,vol.21,pp.483–502,1923.
[18] P. R. B. Saull, L. E.Sinclair, H. C. J. Seywerd, P. J. Boyle, A. M. L.
MacLeod,andD.S.Hanna,“Atwo-pixelComptonimager,”Proc.SPIE,
vol.7665,pp.76651E–76651E–11,2010.
[19] 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
FULLARTICLEAPPEARSINIEEETRANS.NUCL.SCI.61:52745-2752,OCTOBER2014 9
120 Time to reach 1.0° 120 Time to reach 1.0°
a) b)
s) Time to reach 1.5° s) Time to reach 1.5°
e ( 100 Time to reach 2.0° e ( 100 Time to reach 2.0°
g g
a 80 a 80
m m
i 60 i 60
o o
t t
e 40 e 40
m m
Ti 20 Ti 20
0 0
-10 0 10 20 30 40 50 0.2 0.4 0.6 0.8 1 1.2 1.4
θ (°) Energy (MeV)
Fig. 8. a) Timeto image versus the angle between the line joining the centre ofthe front face of the detector and the source, and the symmetry axis of
thedetector. Thisisforthe662keVlineofthe137Cssourceandresultshavebeenscaledtorepresenta10mCisourcewith100%branchingratioat40m.
b)Timetoimageversusenergyforasourceat20◦ fromthesymmetryaxisofthedetector. Results havebeenscaledtorepresent 10mCisourcesof100%
branching ratioat40m.
