nuclear instruments and methods in physics research...

A large area, silicon photomultiplier-based PET detector module

R.R. Raylman a,n, A. Stolin a, S. Majewski a, J. Proffitt ba Center for Advanced Imaging, Department of Radiology, One Medical Center Dr., Box 9236, West Virginia University, Morgantown,WV 26506-9236, United Statesb AiT Instruments, 11815 Fountain Way, Suite 300, Newport News, VA 23606, United States

a r t i c l e i n f o

Article history:Received 1 April 2013Received in revised form30 September 2013Accepted 3 October 2013Available online 12 October 2013

Keywords:PETSilicon photomultipliersNuclear medicine instrumentation

a b s t r a c t

The introduction of silicon photomultipliers (SiPM) has facilitated construction of compact, efficient andmagnetic field-hardened positron emission tomography (PET) scanners. To take full advantage of thesedevices, methods for using them to produce large field-of-view PET scanners are needed. In thisinvestigation, we explored techniques to combine two SiPM arrays to form the building block for a smallanimal PET scanner. The module consists of a 26!58 array of 1.5!1.5 mm2 LYSO elements (spanning41!91 mm2) coupled to two SensL SiPM arrays. The SiPMs were read out with new multiplexingelectronics developed for this project. To facilitate calculation of event position with multiple SiPM arraysit was necessary to spread scintillation light amongst a number of elements with a small light guide. Thismethod was successful in permitting identification of all detector elements, even at the seam betweentwo SiPM arrays. Since the performance of SiPMs is enhanced by cooling, the detector module was fittedwith a cooling jacket, which allowed the temperature of the device and electronics to be controlled.Testing demonstrated that the peak-to-valley contrast ratio of the light detected from the scintillationarray was increased by "45% when the temperature was reduced from 28 1C to 16 1C. Energy resolutionfor 511 keV photons improved slightly from 18.8% at 28 1C to 17.8% at 16 1C. Finally, the coincidencetiming resolution of the module was found to be insufficient for time-of-flight applications ("2100 ps at14 1C). The first use of these new modules will be in the construction of a small animal PET scanner to beintegrated with a 3 T clinical magnetic resonance imaging scanner.

& 2013 Elsevier B.V. All rights reserved.

1. Introduction

Development of a combined MRI–PET system has receivedconsiderable attention over the last two decades. Indeed, thiscombination was first proposed in the mid-1990s [3,10]. It was notuntil the early twenty-first century, however, when arrays ofavalanche photodiodes (APD) became available, that practicalsystems were constructed. In one of the first efforts, a UC Davis-University of Tübingen collaboration created an MR-compatiblePET scanner insert that utilized APDs [8]. This device was con-structed from detector modules consisting of a 10!10 array of2!2!12 mm3 Lutetium Orthosilicate (LSO) elements coupled toa 3!3 array of APDs through a 3.5 mm thick acrylic light guide.The active area of the detector module is approximately 2!2 cm2.The PET scanner was integrated with a small MRI RF coil for usewith a 7 T MRI animal scanner. Grazioso et al. from SiemensMolecular Imaging in Knoxville, TN have also utilized APDs toproduce MRI-compatible PET detector modules designed to beplaced inside the imaging region of a 1.5 T clinical MRI scanner [4].

This device consists of 8!8 arrays of 2!2!20 mm3 LSO ele-ments coupled to 2!2 arrays of APDs. The modules weresuccessfully tested inside the bore of a Siemens 1.5 T SymphonyMRI scanner. This work led to creation of the first commerciallyavailable MRI–PET scanner by Siemens. While the scanners pro-duced with these modules had relatively good characteristics,performance was ultimately limited by the relatively low signal-to-noise ratio (SNR) due to the low gain and temperature-dependent noise of the APDs.

Perhaps the most important development in the creation ofpractical and high performing MR-compatible PET detectors wasthe development of arrays of silicon photomultipliers (SiPM).These devices have higher gain than APDs, comparable to photo-multiplier tubes (on the order of 1!106), and have the sameinsensitivity to magnetic fields as APDs [11]. A number of inves-tigators have created MR-compatible PET detector modules fromwhich MRI–PET scanners can be constructed ([2,6,12,13,19–22]).For example, a group from the Seoul National University con-structed a 32.4!28.7 mm2 SiPM-based PET detector module([21]). The energy resolution of the detector was reported to be13.9% for 511 keV photons. A group from the Netherlands devel-oped a PET detector module utilizing a single 13.2!13.2!10 mm3

piece of LYSO mounted on a 4!4 array of SiPMs ([12]). The use of

Contents lists available at ScienceDirect

journal homepage: www.elsevier.com/locate/nima

Nuclear Instruments and Methods inPhysics Research A

0168-9002/$ - see front matter & 2013 Elsevier B.V. All rights reserved.http://dx.doi.org/10.1016/j.nima.2013.10.008

n Corresponding author. Tel.: þ1 304 293 1973; fax: þ1 304 293 4287.E-mail address: [email protected] (R.R. Raylman).

Nuclear Instruments and Methods in Physics Research A 735 (2014) 602–609

a monolithic piece of scintillator permits the assessment of eachphoton's depth-of-interaction (DOI). Finally, Schulz et al. devel-oped a MRI-compatible PET detector module consisting of a22!22 array of 1.3!1.3!10 mm3 LYSO elements coupled to anarray of SiPMs ([13]). While each of these efforts produced goodPET detector modules, their active areas were relatively small anddid not take full advantage of the potential performance SiPMs bynot cooling them to low temperatures (below 22 1C).

2. Material and methods

The goal of this project was to create a SiPM-based PET detectormodule that will be used as a building block of a large field-of-view(FOV) PET small animal scanner for use with a 3 T clinical MRI scanner.A cooling system for the module was constructed to aid in stabilizingand enhancing the performance of the SiPMs. Finally, a multiplexingscheme was used to reduce the total number of data acquisitionchannels, facilitating construction of practical and cost effective PETscanners.

2.1. Detector design

The new detector module utilizes a 26!58 array of 1.5!1.5!10mm3 (pitch¼1.57 mm) LYSO elements separated by ESRreflector (Proteus, Inc., Knoxville, TN). Thus, the active area of thedetector is 41.2!91.5 mm2, which is larger than other SiPM-baseddetector modules reported in the literature ([2,12,7,13,19–22]). TheLYSO array was coupled to two SensL ArraySL-4p9s (SensL Technolo-gies, LTD., Cork Ireland). These devices are made-up of a matrix of3!3 ArraySL-4 SiPMs. Each ArraySL-4 has 16 (4!4) 3.05!3.05 mm2

pixels (4774 microcells). Their typical gain is 2.4!106, at roomtemperature and recommended bias. Thus, each of the ArraySL-4p9sproduces 3!3!4!4¼144 channels of data; the light-sensitive areais 48!48mm2. A 2 mm-thick piece of acrylic was used to opticallycouple the two SiPM arrays to the scintillator array. The light guide isnecessary to spread light from the LYSO elements to bridge the opticalseam between the two SiPM arrays. Hence, the scintillation lightproduced by elements located at the seamwhere the two SiPM arraysmeet was transmitted to the active areas of the arrays. The light guidealso spreads scintillation light amongst the pixels of the SiPM arrays tofacilitate calculation of event position in the scintillator array. Fig. 1shows the complete detector unit. Note that the SiPM arrays are largerthan the scintillator array. This geometry was chosen so that thescintillator elements at the edges of the array were positioned insidethe light-sensitive area of the SiPM array, facilitating identification ofall of the elements of the scintillator array. This capability will reducegaps in detector sensitivity when the modules are combined in a ringto form a scanner.

2.2. Detector cooling system

As with most solid-state devices, the performance of SiPMs isaffected by temperature. Lowering their temperature reduces

thermal (dark current) noise and increases gain. The gain increaseis due to the reduction of the breakdown voltage at lowertemperatures, while keeping the bias voltage constant. The detec-tor temperature was reduced by enclosing it in a cooling jacket.Specifically, a piece of 4.76 mm thick copper plate was formed intoa rectangle to enclose the ArraySL-4p9s and the readout electro-nics. A channel consisting of 2.38 mm!2.38 mm brass tubing wassoldered to the exterior of the enclosure. Barbed tubing connectorswere soldered to each end of the channel. A mixture of cooled 50%distilled water and 50% ethylene glycol was circulated through thebrass channels, which in turn cooled the copper enclosure. Thecopper is in contact with the edges of the SiPM array, providingconductive cooling. The liquid is cooled with a mini-chiller (PeterHuber Kältemaschinenbau GmbH, Offenburg Germany). Thisdevice can cool the circulating liquid to %10 1C and has a coolingpower of 300 W at 14 1C, which is sufficient to cool a maximum of250 of our modules simultaneously to 14 1C (each module dis-sipates "1.2 W of power). To create convective cooling of themodule, air is circulated inside the unit via of a 3.175 mm-innerdiameter tube integrated with the cooling jacket. This tubingcontains seven small holes that permit air to flow across the SiPMreadout electronics. Air is supplied by a 4.76 mm-inner diametertube connected to a medical air receptacle in the research area.The air is pre-cooled by passing cooling fluid through a heatexchanger in contact with the air supply tube. All of the materialsused for the cooler were chosen for their high heat conductivityand non-magnetic characteristics. The complete PET detectormodule is shown in Fig. 2. The cooling system is capable ofproducing uniform cooling over the surface of the ArraySL-4p9s(variation of o"0.6 1C). Fig. 3 shows an infrared image of thefront face of the cooled module without the scintillator. Note thatthe apparent hot areas in the cooling jacket are caused by thereflection of ambient light from the copper and mismatch in theemissivity of the copper and the emissivity setting of the IRcamera, which was set to the emissivity of silicon.

2.3. SiPM readout and data acquisition system

Each SiPM array contains 144 individual elements, so our dualarray detector module (two ArraySL-4p9s) produces 288 indivi-dual analog outputs. This amount of data makes it challenging toreadout one of the devices, let alone a large number of modulesthat would make up a full scanner. Thus, to make the unit moreappropriate for use in larger systems, the number of outputchannels for each ArraySL-4p9 was reduced from 144 to four,resulting in eight analog outputs per module. This task wasaccomplished with a multiplexed readout system developed incollaboration with AiT Instruments (Newport News, VA).

Output signal multiplexing is accomplished in two stages. First,symmetric charge division circuitry decomposes the 144 channelsof the 12!12 ArraySL-4p9 to 12 rows and 12 columns. Thesesignals are reduced to four channels by applying a weighted gainto each row and column proportional to its location along eachaxis. This amplitude encoding produces two signals (Xþ , X%)representing an event column location, and two signals (Yþ , Y%)representing an event row location. A new charge division tech-nique developed by AiT Instruments uses diodes instead ofresistors to process the SiPM signals prior to amplification.Specifically, a diode is placed between the photodetector and afast transimpedance amplifier. When the photodetector signalpolarity is positive then the diode's low forward resistance allowscurrent flow from the photodetector to the amplifier, while thediode's high reverse resistance blocks current flow from theamplifier to the photodetector. Fig. 4 shows a schematic of thereadout electronics. The diodes reduce undesirable leakage currentamongst the SiPMs. Reduced leakage improves spatial uniformityFig. 1. Picture of the detector unit.

R.R. Raylman et al. / Nuclear Instruments and Methods in Physics Research A 735 (2014) 602–609 603

and event positioning. The nonlinear forward resistance of thediodes attenuates low-amplitude signals, similar to amplitudethresholding, resulting in reduced noise. The diodes isolate indi-vidual SiPM capacitance in a group of SiPMs when each photo-detector is connected to a transimpedance amplifier, whichreduces the effects of cumulative capacitance and charge loss inlarge arrays of SiPMs. Additionally, the output rise time of thetransimpedance amplifier is faster than most directly coupledreadout circuits.

The outputs from the readout boards are connected to aninterface module (AiT Model# SiPMIM4!4) that accepts signalsfrom up to four ArraySL-4p9s (the equivalent of two of ourmodules). The module routes the individual analog signals fromthe readout electronics to a 34-pin connector located on the frontof the device. In addition, the module sums the outputs from eachArraySL-4p9. These signals are used to produce the trigger forinitiation of an analog-to-digital conversion of the X–Y outputs.Additionally, the interface module supplies bias voltage and powerto the readout electronics. Finally, bias voltage and current drawn

by each SiPM array, as well as the temperature of the SiPMs (viatemperature sensors mounted on the boards) can be monitoredusing a 24-pin output on the interface module.

The analog voltage signals from the interface module arerouted through a 34-pin cable to an FPGA-based multi-channel,analog-to-digital converter unit (ADC) (AiT Instruments, NewportNews, VA). Each channel uses an analog delay to compensate fortrigger latency, followed by a gated integrator and an ADC todigitize the integrated signal. These units can process a maximumof 64 channels (each ADC box can thus accommodate up to eightof our modules). Digitization of the analog signals is triggered by alogic pulse. The logic signals are produced by passing the sumoutputs from the interface module (one from each ArraySL-4p9)through a constant fraction discriminator (Philips Scientific Model715) and then combining the two pulse height-discriminatedoutputs with a logic OR operation in a Philips Scientific Model755 logic unit. The digitized outputs are transmitted to thecontrolling computer via a USB2 connection. Fig. 5 shows aschematic of the data acquisition system used to test the module.Determination of the position of a photon interaction point in thescintillator array was performed by calculation of the center-of-mass of the digitized signals from the readout electronics inconjunction with a pre-calculated look-up table (often referredto as a crystal map). This map correlates event position with

Fig. 2. Picture showing the complete PET module.

Fig. 3. Infrared image of the front face of the cooled PET detector module.

Fig. 4. Schematic of the SiPM readout electronics.

R.R. Raylman et al. / Nuclear Instruments and Methods in Physics Research A 735 (2014) 602–609604

scintillator pixel element number. Additionally, a pre-measuredenergy calibration table was applied to the data to calculate theenergy deposited in the scintillator. This table contains a factor foreach scintillator pixel element that converts ADC channel numberto energy. Data acquisition was controlled using the Kmax pro-gramming environment (Sparrow Corp., Port Orange, FL).

2.4. Detector module testing

Testing was performed to assess the capabilities and limitationsof the detector module. First, the dark current as a function oftemperature was determined by measuring the current drawn bythe SiPMs (without the scintillator array attached) as the tem-perature was reduced. To gauge the ability of the system toidentify scintillator elements as the temperature was reduced,the module was irradiated with six, 4.5 μCi 22Na disk sources(Eckert & Ziegler, Valencia CA) evenly distributed across the face ofthe module. Data were acquired for 300 s. The applied bias voltagewas set to 30.4 V, which is the manufacturer-recommendedvoltage at room temperature. Thus, we were able to assess theeffects that changing temperature has on module performance at aconstant bias voltage. Note that the gain of a SiPM is directlyrelated to the difference between the bias voltage and breakdown

voltage, also known as the overvoltage. As the device is cooled, thebreakdown voltage is reduced. Since the bias voltage is constant,the result is an increased overvoltage and consequently increasedgain as temperature decreases. It should be noted that SiPM gaincould be increased by increasing the bias voltage, but thisprocedure has the effect of increasing dark current. The resultingdata was used to produce a two-dimensional plot of detectedevent positions. The pixel map is created from single rather thancoincidence events. These data were used to calculate the peak-to-valley contrast ratios (PVCR) for each detector element, which iscalculated by subtracting the counts present in the region betweenthe positions of adjacent detector elements (valley) from thecounts present in the region of a detector element (peak) dividedby the counts in the peak. This parameter is a measure of theprecision of the center-of-mass calculation used to identify theposition of a photon interaction in the detector. To avoid theinherent variability of using intensity profiles to determine theposition of peaks and valleys, PVCRs were calculated using soft-ware specially created to search for a peak and adjacent valley inthe pixel map. The PVCRs for each scintillator element werecombined to produce a mean value for the module at a giventemperature. Measurements were performed with no cooling ofthe unit (unit temperature¼28 1C) and with the cooling system

Fig. 5. Schematic drawing of the detector module data acquisition system.

Fig. 6. Results from measurement of dark current (a) and calculation of gain (b) as a function of temperature.


turned on (temperature ranged from 22 1C to 16 1C in incrementsof 2 1C). This range was judged reasonable for standard operation.

2.5. Coincidence timing resolution

The coincidence timing resolution of the detector module wasmeasured by placing it in coincidence with a fast detector comprisedof a 1 cm3 piece of LSO coupled to a Hamamatsu R2496 PMT(Hamamatsu Photonics, Hamamatsu City, Japan). The spread ofelectron transit time of this device is 600 ps full-width-at-half-max-imum (FWHM). Three 4.5 μCi 22Na disk sources were placed betweenthe detector module and the fast detector (separation¼3 cm). Theinput signals from the SiPM and PMT were connected to a constantfraction discriminator (Philips Scientific 6915). The resulting pulseswere input to a time-to-digital converter (TDC) (ATMD-GPX, ACAMElectronics, Stutensee-Blankenloch, Germany). The fast detector signalsupplied the start signal and the signal from the detector module wasthe stop signal. Coincidence timing resolutionwas calculated by fittingthe resulting timing distribution to a Gaussian function. The FWHM ofthe fit (corrected for the transit time spread of the PMT used toproduce the start signal) is the coincidence timing resolution of themodule.

3. Results

The plot shown in Fig. 6(a) demonstrates the reduction in darkcurrent with reduced temperature. The relationship between darkcurrent and temperature is linear over this temperature range, asdemonstrated by the fit of the curve to a line. Fig. 6(b) shows a plot ofArraySL-4 gain calculated from SensL data [15] as a function oftemperature. The gain increases as the temperature is reduced becauseit is directly related to overvoltage, which increases with decreasing

temperature (assuming a constant bias voltage). Fig. 7(a) shows arepresentative pixel map. Importantly, all of the 26!58 LYSO detectorelements of the scintillator array were discernable, even at the edgesand seam where the two arrays meet. This finding is important sinceidentification of the elements in these areas can be challenging. Fig. 7(b) shows an intensity profile measured from one of the rows of thepixel map shown in Fig. 7(a). PVCR and energy resolution weremeasured from maps similar to the one shown in Fig. 7. Fig. 8(a) shows a plot of PVCR versus temperature. Energy resolution versustemperature is shown in Fig. 8(b). Finally, Fig. 9(a) shows representa-tive coincidence timing resolution curves. Fig. 9(b) displays the resultsof the coincidence timing resolution measurements. The curve was fitto the function TRes¼M1þM2e(M3T), which was chosen because it bestfit the data. M1 is the temperature-independent component ofcoincidence timing resolution for the module (TRes), M2 is theamplitude of the exponential component of the timing resolution,M3 is the temperature coefficient of the timing response and T is thetemperature of the SiPMs. The best timing resolution achievable withthis detector is "2112 ps (the value of the temperature independentcomponent of the curve), which is just slightly below our measuredvalue of 2115 ps at 16 1C. It is challenging to put these results incontext with the results from other groups because of the largevariation in devices and testing methods. Groups measuring coin-cidence timing resolution for LYSO-based SiPM detectors reportresults ranging from 120 ps to 1500 ps ([9,6,14,1,5], Shimizu et al.,[16,18]). It should be noted that none of these results are for the SensLArraySL-4p9.

4. Discussion

Construction of PET detectors that are not susceptible to theeffect of strong magnetic fields has been significantly advanced by

Fig. 7. (a) Pixel map from the detector at a temperature of 16 1C and (b) representative intensity profile from a central row (denoted by the white line) of the map shownin (a).


the introduction of SiPMs. Commercially available SiPMs arraysare, however, not of sufficient size to be used to create large-FOVPET scanners. Therefore, they must be tiled to produce larger units.In this investigation, we developed and tested a SiPM-baseddetector module that will be used as a building block for a plannedsmall animal PET scanner.

Fig. 1 shows the detector/electronics components of the unit.The complete PET detector module, detector components andcooling system, is shown in Fig. 2. This system was capable ofproducing relatively uniform cooling across the SiPM arrays of themodule, as demonstrated in the infrared image shown in Fig. 3,which is important to maintain uniform performance of the SiPM'sacross the active area of the detector. As with the detectorcomponents, including the electronics shown in Figs. 4 and 5,the cooling jacket concept can be to scaled up for construction oflarger detector modules.

The data shown in the plots of Fig. 6 demonstrate the effectthat lowering the temperature of the module has on gain and dark

current. Dark current has a linear relationship with temperatureover the range tested (as demonstrated by the plot in Fig. 6a), witha slope of "2 μA/1C. Cooling the SiPMs removes energy from thedevices, so the probability of spontaneous electron–hole pairs islowered, reducing the dark current. Gain increase as a function oftemperature (Fig. 6(b)) is caused by the decrease of breakdownvoltage, which is related to the reduced probability of phonon–electron interactions at lower temperatures. Since the bias voltagewas fixed, the overvoltage, and hence gain of the device was increased.It should be noted that the reduction of dark current is slightly offsetby the increase in gain (the dark current is amplified by same gainfactor as the signal). Instead of cooling the SiPMs, it is possible tocompensate for the loss in gain caused by increase in devicetemperature (produced by the power dissipated by the SiPMs andreadout electronics) by increasing the bias voltage. This method,however, does not address the increase in dark current due to highertemperatures. Large arrays of SiPMs (288 in the case of our module)read out with multiplexing electronics are especially susceptible to the

Fig. 8. (a) Mean peak-to-valley contrast ratio (PVCR) versus temperature (the error bars show the standard deviation of the mean PVCR value) and (b) energy resolutionversus temperature.

Fig. 9. (a) Representative coincidence timing resolution curves and (b) plot of the module coincidence timing resolution as a function of temperature.


effects of increased dark current since the signal from each SiPMcontributes to detector noise. Thus, cooling was deemed the bestmethod to improve and stabilize our detector's performance.

The pixel map shown in Fig. 7 illustrates that all of the detectorelements are identifiable, even at the edges of the scintillatorarray. The inability to detect all of the elements of a scintillatorarray is often caused by reduction in detection efficiency at theedges of the light detection device. In our detector unit, the twoArraySL-4p9s are approximately "2.5 mm wider and longer thanthe scintillator arrays. Thus, it is possible to efficiently detect lightfrom the edges of the scintillator array since all of the scintillatorelements are located within the light sensitive area of the SiPMarrays. This lack of dead areas around the detector reducesdegradation in detection sensitivity at the seams of two adjacentmodules when they are arranged in a ring scanner, reducingartifacts caused by incomplete angular sampling. Another impor-tant finding from the pixel map is the fact that the scintillatorelements located at the seam where the two ArraySL-4p9s meetare identifiable. Hence, there are no gaps in the active area of themodule. This capability is achieved by the use of a 2 mm-thickacrylic light guide, which allows the light from scintillator ele-ments to be spread from an area where there are no SiPMelements to active areas of the device. The increase in gain andreduction in noise achieved by cooling also contributes to theability to identify events that occur in detector elements located atthe seam between the two SiPMs. The success of this conceptmeans that it is theoretically possible to tile a number of SiPMarrays to create larger detector modules.

Perhaps the most relevant way to gauge how increased signaland reduced noise affects detector performance is by assessing theprecision with which events are localized in the scintillator array.To characterize this quantity, we calculated the mean peak-to-valley contrast ratio (PVCR) for the module. The results shown inFig. 8(a) demonstrate that the PVCR for the detector moduleincreases by "45% as the temperature is reduced from 28 1C to16 1C. As described above, cooling the detector increases the gainand reduces dark current, which results in increased signalamplitude and reduced noise (mitigated somewhat by theincreased amplification of noise). Since these signals are used tocalculate event position in the scintillator array with a center-of-mass algorithm, higher amplitude signals in the presence of lowernoise makes these calculations more precise. While an increase inPVCR should result in improved spatial resolution, it is not a directmeasure of this parameter. Ultimately, improvement in spatialresolution resulting from cooling the SiPMs will have to beassessed once a scanner is constructed.

Increased signal amplitude and reduced noise produced bycooling also improves the energy resolution of the detector byincreasing gain. The results in Fig. 8(b) show that energy resolu-tion improved from 18.8% at 28 1C to 17.8% at 16 1C for 511 keVphotons. This small changes over the temperature range tested islikely due to the fact that the energy resolution for the module isdominated by differences in light yield from the 1508 individualscintillator elements and slight differences in gain amongst the288 individual SiPMs. Neither of these effects has a strongtemperature dependence. The results shown in Fig. 9 demonstratethat cooling of the SiPMs improves the coincidence timing resolutionof our detector module. The coincidence timing resolution is poorcompared to that reported by a number of other groups for SiPMs anddSiPMs [9,6,14,1,5,18]. For example, Seifert et al. [14] reported anextremely good coincidence timing resolution of 171 ps. Unlike ourmeasurements, this group used a single 3!3mm2 piece of LYSOcoupled to a single SiPM (Hamamatsu MPPC) (instead of an array ofSiPMs and scintillator elements). Indeed, almost all of the reportedcoincidence timing resolution results reported in the literature are forsingle or small SiPM-scintillator arrays, in contrast to our results for a

26!58 array of 1.5!1.5 mm2 LYSO elements (spanning 41!91mm2)coupled to 288 SiPMs. This effect is demonstrated by the work ofShimizu, et al. who compared coincidence timing resolution measuredfrom one SiPM to that for an array of SiPMs [16]. Specifically, thecoincidence timing resolution of the single SiPM (MPPC)-scintillator(5!5!20mm3 LYSO crystal) combinationwas reported to be 354 ps,for an 8!8 array of SiPMs the coincidence timing resolution increasedto 1208 ps. This increase in coincidence timing resolution for the largerarray is caused, at least, in part, by choices of readout scheme and thelack of module wide timing calibration.

Our results indicate that coincidence timing resolution has atemperature independent component, in addition to an exponen-tial, temperature dependent component. The most likely explana-tion for the temperature dependence is linked to the fact that astemperature is decreased, device gain increases, which increasesthe photon detection efficiency (PDE). PDE is major factor influen-cing coincidence timing resolution. Another effect of reducedtemperature is reduced dark current, which will also affects timingresolution. The temperature independent component is the majorcontributor to the timing resolution measured for the ArraySL-4p9. While the coincidence timing resolution is affected by thelength of the scintillator elements, their surface treatment and thepresence of a 2 mm-thick light guide, much of the temperatureindependent component is due to the interaction of the SiPM'sintrinsic capacitance with the multiplexed readout. Specifically,this interaction affects the rise time of the output pulses, whichdegrades timing resolution. Thus, as we have shown, the coin-cidence timing resolution capabilities of our detector module can beimproved modestly by reducing temperature. It will require changesin the structure of the ArraySL-4p9, readout electronics (possiblyutilizing specialized ASICs [17]) and timing calibration of the moduleto significantly improve coincidence-timing resolution.

In summary, we have constructed a large area SiPM-based PETdetector module using the SensL ArraySL-4p9 devices. The modulehas the capability to cool the SiPMs, as well as special electronicsfor acquiring and processing signals from the unit. Performance asassessed by the measurement of some relevant metrics waspromising. The next step is to incorporate 12 of the new modulesinto a ring to form a small animal PET scanner to be used inconjunction with a 3 T MRI scanner.

Acknowledgments

The authors thank Carl Jackson and Wade Appelman fromSensL Technologies Ltd. for their assistance with the siliconphotomultipliers. This work was supported by Funds from theNIBIB (R01 EB007349).

References

[1] M. Ahmed, B. Camanzi, J. Matheson, Nucl. Instrum. Methods A 695 (2012) 252.[2] H. Chagani, R. Dolenec, S. Korpar, P. Krizan, R. Pestotnik, A. Stanovnik,

R. Verheyden, IEEE Nucl. Sci. Symp. Med. Imaging Conf. Rec. 1 (2009) 1518.[3] N.L. Christensen, B.E. Hammer, B.G. Heil, K. Fetterly, Phys. Med. Biol. 40 (1995)

691.[4] R. Gazioso, N. Hang, J. Corbeil, M. Schmand, R. Ladebeck, M. Vester, G. Schnur,

Nucl. Instrum. Methods A 569 (2006) 301.[5] C. Lee, Y.S. Kim, W.S. Sul, H. Kim, S.H. Shin, G. Choo, Nucl. Instrum. Methods A

633 (Suppl 1) (2011) S163.[6] G. Llosa, N. Belcari, M. Bisogni, G. Collazul, S. Marcatili, S. Moehrs, F. Morsani,

C. Piemonte, A. Del Guerra, IEEE Trans. Nucl. Sci. 56 (2009) 543.[7] G. Llosa, J. Barrio, J. Cabello, C. Lacasta, J.F. Oliver, M. Rafecas, C. Solaz,

P. Barrillon, C. de La Taille, M.G. Bisogni, A. Del Guerra, C. Piemonte, IEEENucl. Sci. Symp. Med. Imaging Conf. Rec. 1 (2011) 3631.

[8] B.J. Pichler, M.S. Judendofer, C. Catana, J.H. Walton, M. Kneiling, R.E. Nutt, S.M. Siegel, C.D. Claussen, S.R. Cherry, J. Nucl. Med. 47 (2006) 639.

[9] C. Piemonte, A. Gola, E. Mazzuca, A. Piazza, A. Tarolli, V. Zorzi, V. Schulz, T. Solf,IEEE Nucl. Sci. Symp. Med. Imaging Conf. Rec. 1 (2010) 1804.


[10] R.R. Raylman, B.E. Hammer, N.L. Christensen, IEEE Trans. Nucl. Sci. NS-43(1996) 2406.

[11] E. Roncali, S.R. Cherry, Ann. Biomed. Eng. 39 (2011) 1358.[12] D.R. Schaart, H.T. van Dam, S. Seifert, R. Vinke, P. Dendooven, H. Lohner, F.

J. Beekman, Phys. Med. Biol. 54 (2009) 3501.[13] V. Schulz, B. Weissler, P. Gebhardt, T. Solf, C.W. Lerche, P. Fischer, M. Ritzert,

V. Mlotok, C. Piemonte, B. Golschmidt, S. Vandenberghe, A. Salomon, P.K. Schaeffer, IEEE Nucl. Sci. Symp. Med. Imaging Conf. Rec. 1 (2011) 4467.

[14] S. Seifert, R. Vinke, H.T. van Dam, H. Löhner, P. Dendooven, F.J. Beekman,D. Schaart, IEEE Nucl. Sci. Symp. Med. Imaging Conf. Rec. 1 (2009) 2329.

[15] SensL Technologies, LTD., Cork Ireland 2011, ArraySL-4 Scalable Silicon Photo-multiplier Array Data Sheet.

[16] K. Shimizu, H. Uchida, K. Sakai, M. Hirayanagi, S. Nakamura, T. Omura, IEEETrans. Nucl. Sci. NS-60 (2013) 1512.

[17] T. Solf, V. Schulz, B. Weissler, A. Thon, P. Fischer, M. Ritzert, V. Mltotok,C. Piemonte, N. Zorzi, IEEE Nucl. Sci. Symp. Med. Imaging Conf. Rec. 1 (2009)2798.

[18] H.T. Van Dam, G. Borghi, S. Seifert, D.R. Schaart, Phys. Med. Biol. 54 (2013)3243.

[19] Y. Wang, Z. Zhang, Wang B. Li, L. Shuai, B. Feng, P. Chai, S. Liu, H. Tang, T. Li,Y. Liao, X. Huang, Y. Chen, Y. Liu, Y. Zhang, Nucl. Instrum. Methods A 670 (2012)49.

[20] S. Yamamoto, M. Imaizumi, T. Watabe, H. Watabe, Y. Kanai, E. Shimosegawa,J. Hatazawa, Phys. Med. Biol. 55 (2010) 5817.

[21] H.S. Yoon, G.B. Ko, S.I. Kwon, C.M. Lee, M. Ito, I. Chan Song, D.S. Lee, S.J. Hong, J.S. Lee, J. Nucl. Med. 53 (2012) 608.

[22] N. Zorzi, M. Melchiorri, A. Piazza, C. Piemonte, A. Tarolli, Nucl. Instrum.Methods A 636 (2011) S208.


nuclear instruments and methods in physics research...

Documents