Design and Performance of a High Spatial Resolution, Time-of-Flight PET Detector

1092 IEEE TRANSACTIONS ON NUCLEAR SCIENCE, VOL. 61, NO. 3, JUNE 2014

Design and Performance of a High Spatial Resolution,Time-of-Flight PET Detector

Srilalan Krishnamoorthy, Member, IEEE, Benjamin LeGeyt, Member, IEEE, Matthew E. Werner, Madhuri Kaul,F. M. Newcomer, Joel S. Karp, Fellow, IEEE, and Suleman Surti, Senior Member, IEEE

Abstract—This paper describes the design and performance ofa high spatial resolution positron emission tomography (PET) de-tector with time-of-flight capabilities. With an emphasis on highspatial resolution and sensitivity, we initially evaluated the perfor-mance of several and mm and 12–15 mmlong LYSO crystals read out by several appropriately sized PMTs.Experiments to evaluate the impact of reflector on detector perfor-mance were performed and the final detector consisted of aarray of mm LYSO crystals packed with a

diffuse reflector and read out by a single Hamamatsu 64 channelmulti-anode PMT. Such a design made it compact, modular andoffered a cost-effective solution to obtaining excellent energy andtiming resolution. To minimize the number of readout signals, acompact front-end readout electronics that summed anode signalsalong each of the orthogonal directions was also developed. Ex-perimental evaluation of detector performance demonstrates cleardiscrimination of the crystals within the detector. An average en-ergy resolution (FWHM) of and average coincidencetiming resolution (FWHM) of 348 ps wasmeasured, demonstratingsuitability for use in the development of a high spatial resolutiontime-of-flight scanner for dedicated breast PET imaging.

Index Terms—LYSO, multi-anode PMT, positron emission to-mography (PET), spatial resolution, time-of-flight (TOF) detector.

I. INTRODUCTION

W HILE the importance of detector spatial resolutionand sensitivity on overall positron emission tomog-

raphy (PET) performance are well understood, including thetime-of-flight (TOF) information has been shown to improvethe quantitative powers of PET [1]–[4]. With the resurfacingof TOF PET in this past decade there have been significantadvances in developing high performance detectors [5][6].In particular, there has been a push to improve the timingresolution of detectors for TOF PET. The improvements intiming resolution have usually resulted in limiting the system

Manuscript received September 20, 2013; revised January 09, 2014; acceptedJanuary 19, 2014. Date of publication April 24, 2014; date of current versionJune 12, 2014. This work was supported in part by the National Institutes ofHealth under grants R01EB009056 (National Institute of Biomedical Imagingand Bioengineering), R01CA113941 (National Cancer Institute) and a researchagreement with Saint-Gobain crystals.S. Krishnamoorthy, B. LeGeyt, M. E. Werner, M. Kaul, and S. Surti are with

the Department of Radiology, University of Pennsylvania, Philadelphia, PA19104 USA (e-mail: [email protected]).F. M. Newcomer is with the Department of Physics and Astronomy, Univer-

sity of Pennsylvania, Philadelphia, PA 19104 USA.J. S. Karp is with the Departments of Radiology and Physics & Astronomy,

University of Pennsylvania, Philadelphia, PA 19104 USA.Color versions of one or more of the figures in this paper are available online

at http://ieeexplore.ieee.org.Digital Object Identifier 10.1109/TNS.2014.2302580

spatial resolution or sensitivity, and researchers are still ex-ploring ways to develop a detector with high spatial and timingresolutions. We describe here the development of a high spatialresolution scintillation detector with TOF capabilities. Whilesuch a detector would have numerous applications, we areparticularly interested in demonstrating its relevance in apartial ring dedicated breast PET scanner where the TOF datais expected to help mitigate artifacts arising from the limitedangular sampling of the breast [7][8].Previously [9], we had reported on the design and perfor-

mance of an initial design of one such high spatial resolutionTOF detector. The detector design was based on the pixelatedAnger-logic detector [10] where an array of individual crystalsis read out by an array of larger PMTs coupled to it via a light-guide. A 7-PMT array of 13 mm diameter PMTs was used toread out an array of mm LYSO crystals. Whilethe detector produced sufficiently good flood images with goodcrystal discrimination and a coincidence timing resolution of

ps, the measured timing resolution was poorer in compar-ison with results obtained with single crystals directly coupledto a single PMT. The loss in timing resolution is mainly due tothe insensitive photocathode area surrounding the PMTs used inthat detector.The present paper describes the design and performance

of an improved, modular detector whose TOF performanceis closer to measurements with individual crystals directlycoupled to the PMT. We first assess the timing performance ofseveral individual small cross-section crystals read out usingseveral fast PMTs to demonstrate the feasibility of achievingvery good timing with such small, long crystals. Experiments toevaluate the impact of two commonly used reflector materialsfor building a larger crystal array are also performed. Dedicatedfront-end readout electronics that summed the anode signalsfrom a multi-anode PMT along the orthogonal directions toreduce the number of readout channels from the detector aredesigned and fabricated. Finally, a complete detector module isassembled and its performance evaluation is presented.

II. MATERIALS AND METHODS

All measurements in this paper have been performed usingNIM electronics, CAMAC based data acquisition system. Mea-surements to evaluatedetectorperformancewereperformedwitha reference detector comprised of a small SaintGobainBrilliancedetector ( cm LaBr ) coupled to a Photonis XP20D0 PMT.Two such reference detectors were measured to have a coinci-dence timing of 225 ps FWHM. The PMT under test (Table I)was biased at the manufacturer recommended bias voltage. Data

0018-9499 © 2014 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

KRISHNAMOORTHY et al.: DESIGN AND PERFORMANCE OF A HIGH SPATIAL RESOLUTION PET DETECTOR 1093

TABLE ICOMPARISON OF SELECT COMMERCIALLY AVAILABLE FAST PHOTOMULTIPLIERTUBES APPROPRIATE FOR THE DEVELOPMENT OF A HIGH SPATIAL RESOLUTIONTOF PET DETECTOR. ALL THE ABOVE TUBES ARE MANUFACTURED BY

HAMAMATSU PHOTONICS [15]. EXCEPT FOR THE H8500, WHICH IS A SQUAREMULTI-ANODE TUBE, ALL OTHER PMTS LISTED ABOVE ARE CIRCULAR. THEH8500 HAS AN ARRAY OF ANODES WITH A PITCH OF 6.08 MM.

collection was performed with a 511 keV source placedbetween the test and reference detectors. All measurementresults reported in this paper are calculated and reported as thecoincidence timing resolution from two test detectors, aftercorrecting for the contribution from the reference detector.For tests with the single channel PMTs, signals from both

the reference PMT and the PMT under test were fed to a risetime compensated fast timing discriminator (LRS 825Z). Theresulting trigger signals were then passed on to a coincidencelogic unit to generate a coincidence signal that was used as theDAQ trigger as well as the start signal for a TDC with 25 psresolution. The energy information was measured from an iden-tical copy of the signal used as described above to generatethe timing information. For the multi-anode PMT, the dynodesignal was used to extract both the energy and timing informa-tion, and the energy in multiple anodes was used to calculate the511 keV photon interaction position in software using Angerlogic. Each of the 64 anodes from the multi-anode PMT wasread out by a separate Lecroy charge integrating ADC. Fig. 1shows the schematic of the experimental setup used for the mea-surements. The DAQ was controlled by a LabVIEW program,which saved the PMT and TDC information in listmode format.All the data analysis was performed offline using software de-veloped in MATLAB.

A. Timing Resolution with Small Cross-Section ScintillationCrystals

It is well known that the timing resolution of a scintillationdetector is directly related to the light yield and light collectionefficiency in the scintillator [11]. Since small cross-section andlong crystals have inherently poorer light collection efficiency[12][13], it was important to assess the feasibility to obtain goodtiming performance. Hence we evaluated the timing perfor-mance of several small and long crystals directly coupled to fastPMTs appropriate for a TOF detector design. Table I lists thePMTs and their specifications that were used in this study. TheR4998 is one of the best-in-class TOF PMTs and was used tobenchmark the timing performance that could be achieved withthe small cross-section crystals. The R3478 is identical in size tothe 19 mm PMTs used to read out an array of mmcrystals as used in the development of a small animal PETscanner [14]. The R4124 and R1635 are fast PMTs that are

Fig. 1. Schematic of the experimental setup and data acquisition system usedto perform detector measurements.

smaller in size andmay be appropriate for use in detectors wherecrystals smaller than mm in cross-section need to bediscriminated. The H8500 is a multi-anode PMT that has goodtiming characteristics and is cost-effective in comparison withthe R4124 and R1635 single channel PMTs for decoding largercrystal arrays. In addition, the H8500 allows a modular detectordesign that provides flexibility in the design of a scanner.With a target spatial resolution of 1–2 mm, we evaluated the

timing resolution achieved with (lutetiumoxyorthosilicate, LYSO) crystals of small cross-sections (

mm & mm ). To achieve good detector sensi-tivity, crystal lengths of 12–15 mm were chosen. Each of thecrystals was wrapped in several layers of Teflon on five sidesand grease coupled (BC-630, Saint-Gobain Crystals) directly toeach of the different PMTs listed in Table I. For comparison, wealso performed measurements with a mm LYSOcrystal that is similar in size to the crystals used in some com-mercial whole-body TOF PET scanners.

B. Detector Design and Impact of Reflector Material

In previous work we have shown that the active area of thePMT is an important criterion in designing a large area highspatial resolution detector with good performance i.e. crystaldiscrimination, energy and timing resolution [9]. The use of amulti-anode PMT such as the H8500 should improve the ac-tive area, minimize the deterioration in detector performancearising from light loss and also provide a relatively cost-effec-tive, modular design. The smaller sized anodes should increasethe detector sampling, improving crystal discrimination. With a6.08 mm anode pitch, the entrance glass window on the H8500PMT (1.5 mm) provides enough light sharing to help decodecrystals that are smaller than the anode pitch. In addition, aswill be later shown in Section III.A, the H8500 provides goodtiming resolution as well, making it appropriate for use in thedesign of a TOF detector. We therefore chose to investigateuse of the H8500 PMT for developing a multi-crystal detectorwith good timing performance and crystal discrimination. Fig. 2shows the schematic of the detector. A few groups have previ-ously designed PET detectors utilizing the H8500 PMT to readout scintillation crystals with cross-section of about mm[16]–[18].While these investigations have reported good crystaldiscrimination they have not carefully studied the timing perfor-mance or considered their use for TOF PET.


Fig. 2. Schematic of the detector design, left: side view, right: top view. Thedetector consists of an array of scintillators read out by a single multi-anodePMT. With minimal dead space, the design is compact, modular and offers agood compromise between cost and achieving very good detector performance.

Fig. 3. Left: Photo of a single array of mm LYSOcrystals assembled using a white plastic reflector incorporated in a transparentadhesive used to bond the crystals in an array. Right: Photo of a singlearray of similar sized LYSO crystals assembled using a diffuse sheet reflectorwithout any adhesive bonding.

Initial experiments (Section II.A) to assess the expectedtiming performance with individual, small crystals were per-formed by wrapping the crystal with several layers of Teflon.While Teflon is a well regarded reflector, this technique doesnot extend well to building large, close packed arrays withsmall scintillation crystals. Hence we studied the impact oftwo different reflector crystal assembly techniques used inthe manufacture of larger crystal arrays. A array ofpolished mm LYSO crystals was assembledwith a m thick white plastic reflector incorporated ina transparent adhesive typically used to bond crystals into amonolithic block. The above technique is known to be robust,allows ease of array assembly, and is hence commonly usedto manufacture scintillation arrays with high packing density[19][20]. Since reflector performance was expected to impactdetector performance with high aspect ratio crystals, in parallel,an array of same size crystals was also assembled usinga m thick layer of diffuse sheet reflector without anyadhesive bonding. Fig. 3 shows a photo of the two assembledprototype crystal arrays. Measurements to evaluate crystaldiscrimination in a flood map, as well as to obtain the energyand timing resolution of individual crystals by gating on eventswithin each crystal, were performed for both crystal arrays.

C. Reducing the Number of Readout Channels from theMulti-Anode PMT

Our previous simulation studies and experiments [9] had sug-gested 13 mm diameter PMTs to be sufficient for obtaining goodcrystal discrimination. Hence, while the 6.08 mm sized anodesprovided better sampling, we were interested in determining ifwe could reduce the number of readout channels from the PMT

Fig. 4. Schematic of the circuit used to realize the anode summing circuit inthe front-end electronics for read out of the multi-anode PMT. While a singlesumming circuit is shown here (in the dotted region), the entire PMT readoutconsists of 16 such summing circuits with each anode contributing to two sep-arate anode summing circuits.

without significantly degrading the detector energy and timingresolution performance. The reduced number of readout chan-nels would make the detector design cost-effective and practicalwhen building a scanner. To evaluate its feasibility, we used thepreviously acquired detector data with the crystal arraycoupled to the H8500 PMT (Section II.B). From the listmodedata, the 64 anode signals were reduced in software torow-column anode sums i.e. 8 each along both the orthogonaldirections. The reduction did not correct for the gain variationamong the anodes.Once again,measurements to evaluate crystaldiscrimination in a floodmap, as well as to obtain the energy andtiming resolution of individual crystals, were performed usingboth, the 64 anodes and row-column anode sums datasets.

D. Hardware Design of Row-Column Sum Readout Electronics

In conjunction with the feasibility studies described inSection II.C, we also built dedicated front-end electronicsto implement the row-column anode sums in hard-ware. Other groups have previously built readout electronicsfor multi-channel PMTs [21][22], and have typically madeuse of a resistive network and charge division to reduce thenumber of PMT outputs and provide a position sensitivedetector. Others [23][24] have included a more sophisti-cated approach to incorporate a method of correcting forthe anode gain non-uniformities. While these techniqueshave been shown to work well, we observed that the anodegain non-uniformities (

, ,; as provided by the manufacturer’s

specification sheet [15]) did not have a significant effect onthe performance of our detector. Hence we adopted a simplertechnique to realize the electronic circuit as shown in Fig. 4.The current output from each anode is first converted into a

voltage and then sent to two anode summing circuits, one eachfor a row sum and a column sum (dotted region in Fig. 4 showsa single summing circuit). Each summing circuit uses a wide-band, high speed current feedback amplifier (Texas InstrumentsTHS3201). A resistive network (R1-R16) is also included to ad-just for the anode gain non-uniformities, though we did not uti-lize this part in current work. The entire readout board consists


TABLE IICOINCIDENCE TIMING MEASUREMENTS FOR DIFFERENT (SIZE) INDIVIDUALLYSO CRYSTALS PLACED DIRECTLY ON SEVERAL TEST PMTS OUTLINEDIN TABLE I. THE TIMING MEASUREMENTS INDICATE THAT COMPARABLEOR BETTER TIMING CAN BE ACHIEVED EVEN WITH SMALL AND LONG

SCINTILLATION CRYSTALS. MEASUREMENTS WITH A mm LYSOCRYSTAL ARE ALSO SHOWN FOR COMPARISON.

of 16 such summing circuits. Each of the row/column sum sig-nals is buffered using an additional op-amp (not shown in Fig. 4)and routed to a micro miniature coaxial connector on the I/Oboard. The dynode signal is buffered with a variable gain am-plifier using the same fast op-amp as mentioned above. Separatecopies of the dynode signal are made available for energy andtiming measurement. The variable gain amplifier provides theflexibility to adjust the dynode gain when using several detec-tors in a scanner. The entire readout was designed to be mechan-ically compact and allow packing of multiple readout modulesin the dedicated breast scanner that is being developed. In thiswork the row-column anode sums are integrated with theLeCroy charge integrating ADCs (see Fig. 1), followed by po-sition calculation using Anger-logic in software.

III. RESULTS

A. Timing Resolution with Small Cross-Section ScintillationCrystals

Our timing measurements (Table II) indicate that directlycoupling the small cross-sectional crystals with suitable fastPMTs yields comparable (or better) timing resolution to thatpresently achieved with crystals used in clinical whole-bodyTOF PET scanners [25]–[27]. While some systematic effectscan be observed, these results nevertheless demonstrate theirsuitability for building a high spatial resolution TOF PETdetector. While care was taken to collect data with enoughstatistics and test for repeatability, an average of the measure-ments across both, different crystals sizes (Table II, last row)and across the different individual PMTs (Table II, rightmostcolumn) better represents some of the trends we observe.The faster rise time of the R1635 PMT results in better timing

resolution in comparison with the R4124. The marginally betterrise time, but slightly worse transit time spread of the R4124 incomparison with the R3478 results in comparable timing resolu-tion for these two PMTs. The R4998 PMT has the best rise timeand transit time spread, but the R1635 offers the best overalltiming resolution due to its higher quantum efficiency. Overall,the H8500 provides good timing resolution that is similar to ourbenchmark R4998 PMT. For our detector we chose the H8500

Fig. 5. Top row: Flood images acquired with the , adhesive bonded,white plastic reflector array and the diffuse sheet reflector array placed onthe H8500 PMT. Middle row: Profiles drawn through a single row of crystals(indicated in red) from the flood images acquired above. Bottom row: Repre-sentative energy spectra from the above described arrays. The array assembledwith the adhesive bonded, white plastic reflector shows a DOI dependent vari-ation in light collection that results in a decrease in overall collected light. Leftand Right columns show results for the , adhesive bonded, white plasticreflector array and the diffuse sheet reflector array, respectively.

over the R1635 (which gives the best timing resolution) dueto its potential for achieving good crystal discrimination of the

mm cross-section crystals in a cost-effective manner.

B. Detector Design and Impact of Reflector Material

Flood images acquired with the two crystal arrays assembledwith the two different reflector materials placed at the centerof the H8500 PMT are shown in Fig. 5. While all the crystalscan be clearly separated in both the arrays, the flood quality isbetter with the adhesive bonded, white plastic reflector arrayhaving improved peak-to-valley ratio. However, the average en-ergy and coincidence timing resolution for all crystals within thediffuse sheet reflector array is improved in comparison to the ad-hesive bonded white plastic reflector array (Table III).With the plastic reflector there is a depth-of-interaction (DOI)

dependent variation in the light collection efficiency and overalldecrease in the collected light that explains the degradation inthe energy and timing resolutions (see Fig. 5 Bottom). It is alsoimportant to note that the average coincidence timing resolutionwith the diffuse sheet reflector is very close to measurementswhen using a single mm crystal (albeit shorter,12 mm as opposed to 15 mm long in the final detector design)directly coupled to the H8500 PMT (Table II).


TABLE IIIAVERAGE ENERGY AND TIMING PERFORMANCE MEASURED FOR ALLINDIVIDUAL CRYSTALS FROM THE AND LYSO ARRAY

DESCRIBED ABOVE. ALL MEASUREMENTS WERE PERFORMED WITH A 511 KEVSOURCE AND SUFFICIENT STATISTICS. WHILE WE PRESENT AVERAGE ENERGYAND TIMING RESOLUTION, MEASUREMENT FOR EACH INDIVIDUAL CRYSTAL

HAD 5% PRECISION.

Fig. 6. Top row, left column: Flood image from the diffuse sheet reflectorarray measured using information from all 64 anodes of the H8500 PMT. Toprow, right column: Flood image from the same array derived by performing asoftware based reduction of the 64 anodes to anodes. Bottom row: Profiledrawn through a single row of crystals (indicated in red).

C. Effect of Reducing the Number of Readout Channels fromthe Multi-Anode PMT

Fig. 6 shows the flood image generated from the 64 anodesand the reduced row-column anode sums performed insoftware. As demonstrated there is only a marginal deteriorationin the flood image with all crystals well discriminated.While thedynode signal is used to measure the energy and timing infor-mation, any event mispositioning arising from the row-columnsum procedure can also affect the measured energy and timingresolution for each crystal in the array. Hence to assess its im-pact, we also assessed the energy and timing resolution for allindividual crystals within the array by gating on eventswithin each crystal region in the floodmap. Table IV reports theaverage energy and timing resolution for all 56 crystals fromthe crystal array that was used for this measurement. As shown,the row-column sum reduction has almost no impact on the re-ported energy and timing resolution. The measurements shownin Fig. 6 were performed with the array placed at the centerof the PMT, and a similar trend was also observed when thecrystal array was shifted towards the corner of the PMT (datanot shown).

TABLE IVAVERAGE ENERGY AND TIMING RESOLUTION FOR THE 56 CRYSTALS UNDERTHE TWO SCENARIOS DISCUSSED ABOVE. ALL MEASUREMENTS WERE

PERFORMED WITH A 511 KEV SOURCE AND SUFFICIENT STATISTICS. WHILE WEPRESENT AVERAGE ENERGY AND TIMING RESOLUTION, MEASUREMENT FOR

EACH INDIVIDUAL CRYSTAL HAD 5% PRECISION.

Fig. 7. Left: Photo of a single Hamamatsu H11951 multi-anode PMT and afully assembled, packaged detector module. The detector module consists of a

array of mm LYSO crystals coupled directly to themulti-anode PMT. The scintillator PMT assembly is enclosed within a very thincarbon fiber housing that provides minimal gamma ray attenuation. Right: Photoof the compact front-end readout electronics attached to the multi-anode PMT.The electronics reduces the 64 anodes to row-column anode sums, 8 eachalong both the in-plane directions.

D. Final Design and Experimental Performance of FinalizedDetector Module

Since the detector is being designed for use in a dedicatedbreast scanner, the final design comprised of a array ofpolished mm LYSO crystals assembled with a

m layer of the diffuse sheet reflector. The number of crystalsand reflector thickness were chosen to maximize the packingfraction of the detector without significantly degrading overalldetector performance. The scintillation crystals are coupled tothe Hamamatsu H8500 multi-anode PMT with its high voltagedivider circuit modified to ensure sufficient divider current foroperation at kcps, rates that we expect to encounter inour dedicated breast scanner (Hamamatsu H11951, [28]). Thescintillator-PMT assembly is enclosed in a very thin carbon fiberhousing that provides minimal gamma-ray attenuation (Fig. 7,left). The detector design is compact and modular, providing theability to close pack themwhen building a complete scanner [8].Performance evaluation of a fully assembled detector module

housing a array of crystalswas performed in conjunctionwith a hardware design of the row-column anode sum front-endreadout electronics described earlier (Section II.D) and shownin Fig. 7, right. The experimental setup was very similar to theone outlined in Fig. 1. Instead of using all 64 anodes from thePMT, the row-column anode sums generated by the ded-icated readout electronics were each read out with a charge inte-grating ADC. As explained earlier, energy and timing informa-tion was generated from the dynode signal, while position wascomputed in software using Anger-logic with the row-


Fig. 8. Left: Flood histogram acquired from the final detector assembled asshown in Fig. 7. The detector was read out by the compact readout electronicsalso shown in Fig. 7. Right: A profile (indicated in red) drawn through a singlerow of acquired flood histogram. As can be seen, 30–31 crystals can be clearlyresolved in the flood map. The poorer discrimination over the edge crystal is onaccount of the crystal extending beyond the active area of the PMT.

Fig. 9. Representative energy (left) and coincidence timing (right) spectra forinteractions within a single mm LYSO crystal from the finalizeddetector module.

column anode sums. The readout electronics did not correct forthe individual anode gain non-uniformities, but during data anal-ysis the row-column anode sum signals were normalized usingthe PMT anode gain information provided by the manufacturer.Fig. 8 (left) shows the flood map generated from the measure-

ment with the finalized detector module using events with de-posited energy keV (photopeak events). As can be seen,most of the 1024 crystals can be clearly resolved, demonstratingthe potential for achieving good spatial resolution with this de-tector. The somewhat degraded discrimination of the edge crys-tals (last row and columns) can be explained by the poorer lightcollection on account of them extending slightly beyond thePMT active photocathode footprint ( mm ). Also shownin Fig. 8 (right) is a profile drawn through a single row of crys-tals (highlighted in red). The flood quality from the final detectormodule is very similar to results obtained from the prototypearray assembled with a diffuse sheet reflector (Fig. 5, middlerow, right). We also report the energy and timing resolution forthe detector. An automated algorithm was used to segment the

crystal array flood map using Voronoi decomposition[29] into a lookup table that maps each measured interaction po-sition in the flood map to a physical crystal in the detector array.As shown in Fig. 8, the grid of yellow lines shows the result ofthis segmentation algorithm.The crystal look-up table was then used to assign each event

to a given crystal. In this manner, energy (all events) and timing(events depositing keV in energy) spectra were generatedfor each crystal. A Gaussian fit was performed to the energy andtiming spectra peaks to estimate the respective FWHM values.Fig. 9 shows representative energyand timing spectra for interac-tions from a single crystal within the detector. The measured en-ergy and timing resolution is very uniform over most of the de-

Fig. 10. 2D map (left) and histogram (right) of the measured energy resolutionfor all 1024 crystals from the finalized detector module. An average energy res-olution of FWHM was measured.

Fig. 11. 2D map (left) and histogram (right) of the measured coincidencetiming resolution for all 1024 crystals from the finalized detector module. Anaverage timing resolution of ps was measured. After accountingfor the contributions from the reference detector (225 ps for two referencedetectors in coincidence), an average coincidence timing resolution (FWHM)of 348 ps is calculated for two such detectors in coincidence.

tector, except the edge crystals (Figs. 10 and 11). As explainedabove, this arises from the poorer light collection which some-what degrades both the energy and timing resolution.An averageenergy resolutionof FWHM(Fig.10)andanaveragecoincidence timing resolution of ps FWHM (Fig. 11)was measured for all 1024 crystals within the detector module.

IV. DISCUSSION & CONCLUSION

This paper describes the design and performance of a highspatial and timing resolution PET detector. With a target spatialresolution of 1–2 mm we initially studied the timing perfor-mance of several single, small cross-sectional crystals read outby fast PMTs. While our measurements show some systematiceffects based on the PMT specifications, our results demonstratethat very good timing resolution can be obtained with small andlong scintillation crystals. In particular, our results demonstratethat the crystal cross-section can be reduced from mmto mm , and crystal length increased from 12 mmto 15 mm (35% increase in coincidence photopeak efficiency),without sacrificing timing resolution. Overall, timing perfor-mance of the H8500 PMT is comparable to the R4998 PMT,which is one the best-in-class TOF PMTs. Additionally, theH8500 multi-anode PMT is compact in size, and offers a mod-ular design at moderate cost for building a scanner. Hence it wasour PMT of choice for further evaluation of a multi-crystal PETdetector. Experiments with small detector arrays to assess theimpact of reflector choice for building larger close packed arrayswere performed and it was observed that the diffuse reflectorcrystal assembly without adhesive bonding was better suited for


such a design due to excellent energy and timing resolutionwhilealso maintaining good crystal separation. Use of a white plasticreflector incorporated in a transparent adhesive used for bondingthe crystals resulted in poorer light collection and also a DOIdependent variation in light collection efficiency. Based on theseset of results our choice for thefinal detector designwas aarray of mm LYSO crystals assembled with a

m thick layer of the diffuse sheet reflector. The choice ofcrystal lengthwas a compromise between obtaining good systemcoincidence sensitivity and limiting the degradation in imagingperformance arising due to parallax error [30]. To minimize thenumber of readout channels, compact front-end electronics thatsummed anode signals along each of the orthogonal directionswas also developed. Experimental performance evaluation of acomplete detectormodule demonstrates excellent energy, timingresolution and clear discrimination ofmost crystals. The reduceddiscrimination of edge crystals suggests that while almost 90%of the detectorwill have an intrinsic spatial resolution of 1.5mm,near the detector edges this resolution will be close to 3 mm.Characterizing the impact on the reconstructed system spatialresolution is, however, beyond the scope of our current work. Anaverage energy resolution (FWHM) of and averagecoincidence timing resolution (FWHM) of ps for twosuch detector modules was measured for all 1024 crystals in thedetector. In comparison, current state-of-the-art whole-bodyclinical TOF PET scanners use crystals with mmcross-section and achieve a coincidence timing resolution of500–600 ps [25]–[27]. The energy and timing measurementsof the final detector module are consistent with our initial char-acterization using single crystals directly coupled to a singlePMT, and also the prototype array that was assembledwith a diffuse sheet reflector. It also satisfies our design goals ofbuilding a high spatial resolution TOF detector suitable for usein a dedicated breast PET scanner. Design optimization studiesof partial ring TOF breast PET scanner has suggested that a highspatial resolution detector with 300–600 ps timing resolutioncould aid in providing tomographic images with relatively fewerartifacts and good lesion uptake estimation relative to othernon-TOF partial ring scanner designs [7].

ACKNOWLEDGMENT

The authors wish to thankM. Mayhugh and L. Perna at Saint-Gobain Crystals for their collaboration and help in assemblingthe scintillation crystals and detector.

REFERENCES[1] T. F. Budinger, “Time-of-Flight positron emission tomography: status

relative to conventional PET,” J. Nucl. Med., vol. 24, pp. 73–76, 1983.[2] S. Surti, J. S. Karp, L. M. Popescu, M. E. Daube-Witherspoon, and

M. Werner, “Investigation of time-of-flight benefit for fully 3-D PET,”IEEE Trans. Nucl. Sci., vol. 25, no. 5, pp. 529–538, 2006.

[3] J. S. Karp, S. Surti, M. E. Daube-Witherspoon, and G. Muehllehner,“Benefits of Time-of-Flight: experimental and clinical results,” J. Nucl.Med., vol. 49, no. 3, pp. 462–470, 2008.

[4] C. Lois, B. W. Jakoby, M. J. Long, K. F. Hubner, D. W. Barker, M. E.Casey, M. Conti, V. Y. Panin, D. J. Kadrmas, and D. W. Townsend,“An assessment of the impact of incorporating Time-of-flight informa-tion into clinical PET/CT Imaging,” J. Nucl. Med., vol. 51, no. 2, pp.237–245, 2010.

[5] W. W. Moses, “Time of flight in PET revisited,” IEEE Trans. Nucl.Sci., vol. 50, no. 5, pp. 1325–1330, 2003.

[6] S. Surti, J. S. Karp, G. Muehllehner, and P. S. Raby, “Investigation ofLanthanum scintillators for 3-D PET,” IEEE Trans. Nucl. Sci., vol. 50,no. 3, pp. 348–354, 2003.

[7] S. Surti and J. S. Karp, “Design considerations for a limited angle,dedicated breast, TOF PET scanner,” Phys. Med. Biol., vol. 53, pp.2911–2921, 2008.

[8] E. Lee, M. E. Werner, J. S. Karp, and S. Surti, “Design optimization ofa time of flight, breast PET scanner,” IEEE Trans. Nucl. Sci., vol. 60,no. 3, pp. 1645–1652, 2013.

[9] S. Krishnamoorthy, R. I. Wiener, M. Kaul, J. Panetta, J. S. Karp, andS. Surti, “Development of a high-resolution and depth-of-interactioncapable detector for time-of-flight PET,” in Proc. 2011 IEEE NSS/MIC,pp. 2457–2461.

[10] S. Surti, J. S. Karp, R. Freifelder, and F. Liu, “Optimizing the perfor-mance of a PET detector using discrete GSO crystals on a continuouslightguide,” IEEE Trans. Nucl. Sci., vol. 47, pp. 1030–1036, 2000.

[11] P. Lecoq, E. Auffray, S. Brunner, H. Hillemanns, P. Jarron, A.Knapitsch, T. Meyer, and F. Powolny, “Factors influencing timeresolution of scintillators and ways to improve them,” IEEE Trans.Nucl. Sci., vol. 57, no. 5, pp. 2411–2416, 2010.

[12] C. S. Levin, “Design of a high-resolution and high-sensitivity scintilla-tion crystal array for PET with nearly complete light collection,” IEEETrans. Nucl. Sci., vol. 49, no. 5, pp. 2236–2243, 2002.

[13] J. R. Stickel, J. Qi, and S. R. Cherry, “Fabrication and characterizationof a 0.5 mm lutetium oxyorthosilicate detector array for high-resolutionPET applications,” J. Nucl. Med., vol. 48, no. 1, pp. 115–121, 2007.

[14] S. Surti, J. S. Karp, A. E. Perkins, R. Freifelder, and G. Muehllehner,“Design evaluation of A-PET: A high sensitivity animal PET camera,”IEEE Trans. Nucl. Sci., vol. 50, no. 5, pp. 1357–1363, 2003.

[15] Hamamatsu Photonics K.K. [Online]. Available: http://www.hama-matsu.com

[16] R. R. Raylman, S. Majewski, B. Kross, V. Popov, J. Proffitt, M.F. Smith, A. G. Weisenberger, and R. Wojcik, “Development of adedicated positron emission tomography system for the detection andbiopsy of breast cancer,” Nucl. Instrum. Methods Phys. Section A, vol.569, no. 2, pp. 291–295, 2006.

[17] H. Alva-Sanchez, T. Murrieta, E. Morena-Barbosa, M. E. Brandan, C.Ruiz-Trejo, A. Martinez-Davalos, and M. Rodriguez-Villafuerte, “Asmall-animal PET system based on LYSO crystal arrays, PS-PMTs anda PCI DAQ board,” IEEE Trans. Nucl. Sci., vol. 57, no. 1, pp. 85–93,2010.

[18] T. Yanagida, A. Yoshikawa, Y. Yokota, K. Kamada, Y. Usuki, S. Ya-mamoto, M. Miyake, M. Baba, K. Kumagai, K. Sasaki, M. Ito, N.Abe, Y. Fujimoto, S. Maeo, Y. Furuya, H. Tanaka, A. Fukabori, T. R.Santos, M. Takeda, and N. Ohuchi, “Development of Pr:LuAG scintil-lator array and assembly for positron emission mammography,” IEEETrans. Nucl. Sci., vol. 57, no. 3, pp. 1492–1495, 2010.

[19] M. S. Andreaco, C. W. Williams, J. C. Moyers, and K. Vaigneur,“Method for producing a high resolution detector array,” U.S. Patent6749761 B1, Oct 5, 2001.

[20] J. Uribe, W.-H. Wong, H. Baghaei, R. Farrell, H. Li, M. Aykac, D.Bilgen, Y. Liu, Y.Wang, and T. Xing, “An efficient detector productionmethod for position-sensitive scintillation detection arrays with 98%detector packing fraction,” IEEE Trans. Nucl. Sci., vol. 50, no. 5, pp.1469–1476, 2003.

[21] S. Siegel, R. W. Silverman, Y. Shao, and S. R. Cherry, “Simple chargedivision readouts for imaging scintillator arrays using a Multi-ChannelPMT,” IEEE Trans. Nucl. Sci., vol. 43, pp. 1634–1641, 1996.

[22] P. D. Olcott, J. A. Talcott, C. S. Levin, F. Habte, and A. M. K. Foudray,“Compact readout electronics for position sensitive photomultipliertubes,” IEEE Trans. Nucl. Sci., vol. 52, pp. 21–27, 2005.

[23] V. Popov, “Matrix output device readout system,” US #6747263B1.[24] C. M. Lee, S. Kwon, G. B. Ko, M. Ito, H. S. Yoon, D. S. Lee, S. J.

Hong, and J. S. Lee, “A novel compensation method for the anodegain non-uniformity of multi-anode photomultiplier tubes,” Phys. Med.Biol., vol. 57, pp. 191–207, 2012.

[25] S. Surti, A. Kuhn, M. E. Werner, A. E. Perkins, J. Kolthammer, andJ. S. Karp, “Performance of philips gemini TF PET/CT scanner withspecial consideration for its time-of-flight capabilities,” J. Nucl. Med.,vol. 48, pp. 471–480, 2007.

[26] B. W. Jakoby, Y. Bercier, M. Conti, M. E. Casey, B. Bendriem, and D.W. Townsend, “Physical and clinical performance of the mCT time-of-flight PET/CT scanner,” Phys. Med. Biol., vol. 56, no. 8, pp. 2375–89,2011.

[27] V. Bettinardi, L. Presotto, E. Rapisarda, M. Picchio, L. Gianolli, andM. C. Ginardi, “Physical performance of the new hybrid PET/CT Dis-covery-690,” Med. Phys. , vol. 38, no. 10, pp. 5394–5411, 2011.

[28] Hamamatsu Photonics K.K., private communication, 2012.[29] F. Aurenhammer, “Voronoi diagrams–A survey of a fundamental

geometric data structure,” ACM comput. surveys, vol. 23, no. 3, pp.345–405, 1991.

[30] E. Lee, M. E. Werner, J. S. Karp, and S. Surti, “Design optimizationof a dedicated breast PET scanner using TOF imaging in a partial ringgeometry,” in Proc. 2011 IEEE NSS/MIC, pp. 4276–4280.

Design and Performance of a High Spatial Resolution, Time-of-Flight PET Detector

Documents

Transcript of Design and Performance of a High Spatial Resolution, Time-of-Flight PET Detector