Signal Analysis for Improved Timing Resolution with Scintillation Detectors for TOF PET Imaging

Signal Analysis for Improved Timing Resolution with Scintillation Detectors for TOF PET Imaging R.I. Wiener, Student Member, IEEE, M. Kaul, Student Member, IEEE, S. Surti, Senior Member, IEEE and J.S. Karp, Senior Member, IEEE Abstract-Clinical TOF PET systems offer significantly degraded timing performance in comparison to that measured with small crystal geometries of the same scintillators. The usage of long, narrow crystal geometries is a major contributor to the degradation in system timing performance. We explore the effect of increased pixel length on timing performance of LaBr3[Ce], CeBr3 and LYSO detectors. Using fast signal digitization, we characterize the response of the detector to SUkeV photons interacting at different depths. Systematic shift in time pickoff with interaction depth is shown to degrade the timing performance of long crystals. Collected light, signal shape and signal arrival time are shown to correlate with interaction depth. The depth dependence of detector response shows strong dependence on surface treatment, with finer surface treatment corresponding to reduced depth sensitivity. The correlated dependencies of time pickoff, energy and signal shape offer a method for recovering timing information lost due to interaction depth dispersion. An energy-based correction to the time pickoff is shown to improve detector timing resolution. The improvement in timing performance of thick detectors with single sided readout demonstrates the benefit of inclusion of additional signal information attainable by fast signal digitization. 1. INTRODUCTION he availability of scintillator materials exhibiting high light Toutput, short decay time and fast risetime[i,2] has allowed the detection of coincidence annihilation photons with <loops precision (Table I). Such precise timing measurements have been achieved using detectors comprising of small crystals coupled directly to a fast photodetector and readout path. Maintaining the excellent timing performance of these fast scintillators in a TOF PET detector poses a challenge due to the additional performance requirements of clinical PET imaging. In a whole-body system, detector sensitivity is achieved using crystals of thickness 2-3cm. Spatial resolution and low readout channel count may be achieved using pixelated crystal arrays read by arrays of large PMTs in a light sharing configuration. The usage of long, narrow scintillator pixels results in degraded timing performance as compared to that achieved with short crystal samples of the same material. Light losses in the detector and uncorrected inhomogeneities in detector response result in additional loss of timing information[3,4]. These losses are delineated for two whole-body TOF PET systems in table 1. The higher light output and shorter decay constant of LaBr3[Ce] and CeBr3 result in detectors with improved timing performance in comparison to L YSO based detector configurations[5,6]. The deleterious effect on timing performance of pixel length, choice of PMT and detector configuration is evident for both halides and L YSO. LaBr3 [5%Cel LaBr3 [30%Cel CeBr3 LYSO 4x4x5rnrn 31R4998(ps) 128 98 83 228 Pixel1R4998 (ps) 212 212 245 313 PixelIXP20DO(ps) 276 296 297 453 Detector(p s) 325 Systern(ps) 375 540 (LaPET) (GerniniTF) Table I: Losses in timing performance from small detector to full scanner for selection of halides and L YSO. Halide pixels and L YSO pixels were 4x4x30mm3 and 4x4x25mm3 in size, respectively. Pixel timing resolution is reported for two identical detectors in coincidence. Detector and system timing resolution represent an average over all crystals. The degradation in timing performance resulting from light losses in a light sharing detector may be reduced by improving light collection as well as photodetector response uniformity. The losses due to limited photodetector detection efficiency and high timing jitter may also be reduced in future photodetectors, such as semiconductor-based photodetectors. Reducing the deleterious effect of crystal thickness on detector timing resolution would thus reduce a fundamental limit on the timing resolution of the most common clinical TOF PET detector configuration. In this work we investigate the effect of crystal length on timing performance. We focus on evaluating and correcting for the contribution of 511 kev interaction depth dispersion on timing performance. Section II describes the experimental setup, data acquisition and signal processing techniques used. Section III explores the effect of crystal length on timing performance for three fast scintillators: LaBr3[30%Ce], CeBr3 and LYSO. Section IV describes the depth dependence of detector response for a 4x4x30mm 3 LaBr3[30%Ce] crystal. Section V describes how the depthdependence of detector response can be exploited in order to correct for DOl-related time pickoff dispersion. Section VI explores the effect of surface treatment on DOl dependence of detector response, as well as pulse-shape metrics of DOL II. METHODS Manuscript received November 21, 2010. This work was supported in part by NlH Grant No. ROICAI13941. R. 1. Wiener, M. Kaul and J. S. Karp are with the Department of Physics and Astronomy at the University of Pennsylvania, Philadelphia, PA 19104 USA (corresponding author's e-mail: riwiener@hep.upenn.edu). S. Surti and J. S. Karp are with the Department of Radiology at the University of Pennsylvania, Philadelphia, PA 19104 USA. A. Scintillators The effect of crystal length on timing performance was evaluated for four TOF scintillators: LaBr3[5%Ce], 978-1-4244-9105-6/10/$26.00 2010 IEEE 1991

LaBr3[30%Ce], CeBr31 and LYS02. The scintillation properties of these materials are reported in table II: LaBr3 LaBr3 CeBr3 [5%Ce] [30%Ce] LYSO Relative Light Output 1.00 0.82 0.71 0.32 Energy Resolution(t.EIE) 3.3% 4.2% 4.7% 12.0% The test detector consisted of the wrapped crystal, greasecoupled to a R4998 PMT. An encapsulated 19mm X 19mm LaBr3[5%Ce] crystal2, coupled to a Photonis XP20DO PMT was used as a reference detector for all measurements. The test detector was irradiated in one of two configurations: head-on and from the side. The two configurations are depicted respectively in the top and bottom of figure 1: Decay(ns) 21 21 21 43 10-50% Signal Rise(ps) 993 717 709 808 J.1(cm-1) 0.47 0.47 0.47 0.86 Table II: Scintillator properties for teflon-wrapped 4x4x5mm 3 crystal samples. Relative light output was measured with respect to LaBr3[5%Ce], for which absolute light output of 61,000 PhotonslMeV is reported. Energy resolution was measured at 511 key. Light output and energy resolution were measured using a Photonis XP20DO PMT. Signal rise and decay times were measured using a Harnamatsu R4998 PMT. The LaBr3[5%Ce], LaBr3[30%Ce] and CeBr3 crystals were polished using a 1500 grit micro-mesh polishing kit, corresponding to 30!J1Il abrasive size. Due to the hygroscopic nature of these scintillators, all measurements were taken in a dry environment, maintained in a glove box. The L YSO crystals exhibited a finer polish than the halides as received from the manufacturer. The L YSO crystals were tested with the manufacturer supplied surface finish, though were systematically cut to shorter lengths in our lab. The crystals were wrapped in teflon tape, and were directly coupled to the photodetector using optical grease. B. Photodetectors Two photodetectors were used for tumng measurements: Hamamatsu R4998 and Photonis XP20DO. The performance characteristics of these PMTs are summarized in table III: Hamamatsu R4998 Photonis XP20DO Diameter I" 2" Gain 5.7x106 2.4x105 QEma 0.18 0.25 Risetime 0.7ns 2.5ns TTS 160ps 600ps Table III: Performance characteristics of Harnamatsu R4998 Photonis XP20DO PMTs. The R4998 and XP20DO were operated at a bias voltage of -2100V and -1200V, respectively. The Hamamatsu R4998 offers superior timing performance to the Photonis XP20DO. Its fast risetime allows for more accurate characterization of signal pulse shape. The Photonis XP20DO offers improved energy resolution, and is a more cost-effective PMT for a multi-channel detector. C. Detector Configuration A 1mm wide lead slit collimator was used to produce a collimated beam of annihilation photons from a 22Na source. l LaBr3[5%Ce], LaBr3[30%Ce] and CeBr3 are courtesy of Radiation Monitoring Devices. 2BrilLanCe 380 is a registered trademark for LaBr3[5%Ce] by Saint-Gobain Crystals. The LYSO and encapsulated LaBr3[5%Ce] crystals are courtesy of Saint-Gobain Crystals. Figure la: Experimental set-up for measuring detector response in a typical detector configuration (top). Figure Ib: Set-up for measuring detector response to fixed irradiation depth (bottom). For the fixed irradiation depth measurements, interaction depth is measured as the distance from the edge of the crystal coupled to the PMT. D. Data Acquisition and Analysis In order to maximize the flexibility of our data analysis, we digitized the output signals of our coincidence detectors and analyzed the waveforms offline. We used an 8-bit Agi1ent Acquiris DC27l cpci digitizer, with a maximal sampling rate of 4Gs/s. To reconstruct the signal without aliasing, the signal must be sampled at the Nyquist rate[7], i.e. twice the fastest frequency component of the photodetector signal:!sampling = 2vmax For a signal with risetime of TlOr 90%, the bandwidth[8] will be: vmar=0.349 I TlOr 90% For a Hamamatsu R4998 PMT, with a risetime of 70Ops, the signal bandwidth will be 500MHz and the required sampling rate will be 1 Gs/s. Oversampling allows for a reduction in the approximation error due to the fmite resolution of each sample and is thus of practical benefit[9]. A sampling rate of 2Gs/s was used for all measurements. Event energy was determined by summing the baselinesubtracted pulse samples over a >3'1" time window. Photo-peak energy depositions were well separated in all detector configurations. An energy gate was applied to both reference and test detector, selecting only photo-peak energy depositions. Event arrival time was determined as the crossing time of a constant fraction of pulse amplitude. Timing threshold was optimized for each photodetector, and was set at 8% Vpp for the R4998 PMT and 12% Vpp for the Photonis XP20DO PMT. The timing performance measured using digital constant fraction discrimination was comparable to that measured with a LeCroy 825Z rise time-compensated discriminator, an analog NIM discriminator. 1992

Digital signal processing allows for the evaluation of constant fraction time pickoff with no dependence on signal shape or baseline. The complexity of implementing constant fraction discrimination in analog electronics disfavors the usage of constant fraction discrimination in clinical TOF PET systems, despite the demonstrated superior timing performance it offers[10]. Coincidence measurement of the reference detector with an identical detector resulted in timing resolution of 21Ops±5ps. We subtract the contribution of the reference detector in quadrature from the measured timing resolution, and report the expected timing resolution for two detectors in coincidence. III. EFFECT OF CRYSTAL LENGTH ON TIMING PERFORMANCE The degradation in timing resolution with increased crystal length can be measured for a variety of scintillators. Figure 2 shows the coincidence timing resolution measured with crystals of constant cross-section cut to different lengths. The excellent intrinsic timing properties of LaBr3[30%Ce] and CeBr3 result in coincidence timing measurements of 98ps and 129ps, respectively, measured with a Hamamatsu R4998 PMT. The reduced light output and longer decay time of L YSO allows for coincidence timing resolution of 228ps for a short pixel coupled to a R4998 PMT. The degradation in timing resolution with increasing crystal length is affected by the choice of time pickoff method. Applying a leading edge time pickoff to the L YSO signals results in consistently worse timing resolution than that achieved with a constant fraction time pickoff. The loss of light in longer pixels increases the threshold level with respect to signal amplitude. The resulting increase in amplitudedependent ("walk") time pickoff dispersion is expressed in the measured leading edge timing resolution. IV. DEPTI! DEPENDENCE OF DETECTOR RESPONSE Increased crystal length results in increased range of 511 key DOl dispersion as well as in increased reflections and longer path length of the scintillation photons before detection. In order to measure these two effects experimentally, we controlled DOl by irradiating the pixel from the side with a collimated beam of 511 key annihilation photons (figure Ib). We used a 4x4x30mm 3 LaBr3[30%Ce] crystal as our test detector. The crystal was polished using a 1500 grit micromesh sanding kit, and was wrapped in teflon tape. Detector response was measured using a R4998 PMT coupled to one of the 4x4mm2 faces (section II). figure 3 shows the averaged detector response to 511 ke V photon beams incident from the side at 7.5mm intervals: Figure 3: Averaged detector signals to a collimated beam of 511 key photons interacting at fixed depths. Note the decrease in signal amplitude with increasing interaction distance from PMT. A. Light Collection Figure 2: Coincidence timing resolution vs. crystal length for CeBr 3, LaBr 3 [30%Ce] and L YSO. The solid lines represent timing measurement achieved with a digital constant fraction time pickoff. The dashed line (L YSO LE) represents the timing resolution achieved with a digital leading edge time pickoff for the same L YSO data. For each crystal length, the timing performance of LaBr3[30%Ce] is superior, followed closely by CeBr3 and then by L YSO. For all three scintillators, coincidence timing resolution degrades linearly with crystal length, exhibiting a loss of 4-5ps/mm. The magnitude of the degradation in timing performance is quite significant, and is comparable in magnitude to the timing resolution measured with the shortest crystal. Figure 3 shows the decrease in signal amplitude with increasing interaction distance from PMT. The underlying decrease in light collection results in a monotonic decrease in photo-peak energy centroid (figure 4). The high light output and energy response uniformity of LaBr3[30%Ce] result in excellent energy resolution (table II). The ability to resolve the change in energy with interaction depth allows good depth separation of events interacting in the 15mm proximal to the PMT: Figure 4: Photo-peak energy centrol<l vs. mteractlon distance from PMT for a 4x4x30mm3 LaBr 3 [30%Ce]. The error bars show the FWHM of the photo-peak. Note the monotonic decrease in energy with increasing interaction distance from PMT. Events interacting Omm and 7.5mm from the PMT show well separated energy peaks. 1993

B. Timing Figure 5 shows the amplitude-nonnalized detector response to 511 key photon beams incident from the side at 7.5mm intervals: resolution of 212ps. These results indicate that measurement of event DOl could be used to correct for DOl-dependent offset in time pickoff and thus improve pixelated detector timing perfonnance. V. DEPTH COMPENSATION OF TIME PICKOFF The monotonic change in energy with DOl (figure 4) suggests energy may be used as a measure of DOl, and hence may be used to correct for the DOl related change in time pickoff (figure 6). The correlated changes in time pickoff and energy are manifested in the response of the detector when irradiated head-on (figure 7). Figure 5: Averaged detector signals to a collimated beam of 511 ke V photons interacting at fixed depths. Signal amplitude is normalized unity, and the horizontal axis is scaled to the rise of the signal. Note the delay in signal arrival time with increasing interaction distance from the PMT. The averaged signals shown in figure 5 were measured with respect to a start trigger defined by the reference detector. The increasing delay in signal arrival time with increasing interaction distance from the PMT is consistent with the additional propagation distance transversed by the scintillation photons. Measuring event detection time with respect to the reference detector shows the magnitude of the shift in pulse detection time (figure 6). An increase in interaction distance from the PMT of 30mm results in a shift in mean time pickoff of 390ps. Note that the shift in time pickoff with interaction distance from PMT significantly exceeds 180ps, the delay in first photon arrival time expected for light traveling 3cm in a medium of n3ronm=1.8. Figure 7: Constant fraction time pickotf vs. energy for a 4x4x30mm' LaBr,[30%Ce] crystal irradiated head-on. The correlated delay in constant fraction time pickoff with charge collection is well modeled by a straight line. The effect of interaction depth on detector response is evident in the response to the monoenergetic photo-peak photons. 511 ke V photons interacting further from the PMT result in attenuated and delayed signals as compared with those generated by photons interacting close to the PMT. The effect is manifested in the monotonic trend of time pickoff with energy for photo-peak events (figure 7). For scattered photons, depositing an unknown amount of energy, time pickoff shows no correlation with collected charge. The correlation of time pickoff collected charge can be exploited to improve pixel timing resolution. A pulse-by-pulse subtraction of the the linear trend from the time pickoff results in improved coincidence timing resolution of 192ps (table III). Charge-based correction to time pickoff results in partial recovery of the timing infonnation lost due to DOl dispersion. Figure 6: Constant fraction time pickoffvs. interaction distance from PMT. The coincidence timing resolution of events interacting at a fixed distance from the PMT ranged from 140ps to 168ps, averaging 166ps over all depths. Coincidence timing resolution degraded by less than 30ps for events interacting furthest from the PMT, despite the 30% decrease in light collection measured at that depth. The fixed depth timing measurement allows us to distinguish the effect of DOl dispersion from that of light transport on the timing perfonnance of long pixels. The altered light transport in a long pixel accounts for the degradation in coincidence timing resolution of the fixed depth 4x4x30mm 3 pixel measurement (166ps) from that measured with the 4x4x5mm 3 crystal (89ps). When irradiated head-on, the depth-dependent dispersion in time pickoff (figure 6) results in pixel timing 4x4x5mm' 4x4x30mm' 4x4x30mm3 4x4x30mm3 Fixed DOl Head-On Head-On Energy Corr. Coinc. Tres. vs. 164ps I 89ps 211ps 20lps reference detector Exp. Tres. vs. 89ps 166ps 212ps 192ps identical detector Table III: Comcldence tlmmg resolution for 4x4x30mm 3 LaBr 3 [30%Ce] crystal. The top row shows the timing resolution measured against the BrilLanCe 380 crystal coupled to a Photonis XP20DO PMT. The bottom row shows the expected timing resolution for two identical test detectors. The improvement in timing resolution with charge correction, from 212ps to 192ps, demonstrates the ability to reduce the deleterious impact of pixel length on timing resolution. VI. IMPROVED DOl METRICS The high aspect ratio of scintillator pixels results in multiple interactions with crystal sides for most scintillation photons. 1994

The number and nature of these interactions affect light collection as well as the temporal profile of the light signal. Scintillation photons generated from energy depositions further from the PMT will show greater sensitivity to surface effects, due to the increased probability of reflection they experience before detection. We tested the effect of two surface treatments on depth dependence of signal response. A coarser treatment was achieved using a 1500 grit micro-mesh polishing kit, and a finer surface treatment using a 4000 grit micro-mesh polishing kit. The surface treatments were applied sequentially to the same 4x4x30mm 3 LaBr3[30%Ce] crystal. The finer surface treatment resulted in decreased sensitivity of the collected charge and constant fraction time pickoff to interaction depth. For the coarser surface treatment, collected charge decreased by 30% for events interacting furthest from the PMT in comparison to a decrease of 25% for the finer surface treatment. The delay in time pickoff with increasing interaction distance from PMT decreased from 390ps for the coarser surface to 300ps for the finer surface. Both measurements are markedly higher than 180ps, the shift expected from direct photon propagation path-length alone. Sensitivity of the measured signal risetime to DOl is limited by the risetime of the PMT (700ps for the R4998 PMT). The combined effect of PMT risetime and signal noise result in reduced precision of the measured risetime. Reduction in noise susceptibility can be achieved by measuring the cumulative light collection. The fraction of total charge collected in the first IOns of the pulse (figure 9) shows improved precision as compared with the direct risetime measurement. VII. CONCLUSIONS Our results show that digital pulse sampling of detector response allows extraction and use of intracrystal position information. The correlated dependence of time pickoff and energy can be exploited to improve coincidence timing resolution from 212ps to 192ps in a 30mm long pixel. The measurable change in risetime with DOl offers a novel method for extracting DOl information in a single-sided readout of a pixelated detector. Preliminary results show that DOl discrimination is possible for events interacting close to the PMT. Surface treatment and DOl metrics may be further optimized to offer improved DOl resolution while maintaining detector energy and timing resolution. The ability to extract DOl information and improve pixel timing resolution hold promise for reduction in the deleterious effect of pixel length on detector performance. These results emphasize the importance of intrinsic scintillator timing properties -light output and decay time -in achieving superior system timing performance. VIII. ACKNOWLEDGMENT Figure 8: 10%-50% risetime vs. DOL The coarser surface treatment results in faster risetime measurement for events interacting close to the PMT. The combined effects of interaction depth and surface finish on light transport result in a depth-dependence of signal risetime. The increased dispersion in detection time of scintillation photons generated further from the PMT is manifested in a slower risetime of the detector response. Measurement of the 10%-50% risetime of the signal for the coarser surface treatment shows faster risetime for events interacting 0-7.5mm from the PMT. The finer surface treatment results in little DOl sensitivity. Figure 9a: (top) Fraction of total charge collected in the first IOns of the signal vs interaction distance from PMT. The coarser surface treatment results in a more pronounced DOl dependence. For the coarser surface, the percentage of total charge collected in the first IOns varied from 40.0±1. 3 close to the PMT to 35. 4±1. 1 30mm from the PMT. Figure 9b: (bottom) DistributiOl of fraction of total charge collected in the first IOns of the signal for events interacting 30mm, 15mm and Omm from PMT for coarse finish crystal. The plot shows depth discrimination for events interacting close to the PMT. We would like to thank the research members at Saint-Gobain and Radiation Monitoring Devices for their continued support. This work was supported by NIH ROI CAI13941. [I] REFERENCES van Eijk, C. W. E., Inorganic scintillators in medical imaging. Physics in Medicine and Biology, 2002. 47: p. R85-RI06. [2] Lecoq, P., et ai.,factors influencing time resolution of scintilla tors and WlryS to improve them, Nuclear Science Symposium Conference Record (NSSIMIC), 2009 IEEE, vol., no., pp. 1880-1885, Oct. 24 2009-Nov. I 2009 [3] Kuhn, A., et ai., Design of a lanthanum bromide detector for time-o.f flight PET. IEEE Trans. Nucl. Sci., 2004. 51(5): p. 2550-2557. [4] Karp, ls., et ai. Characterization oftof PET scanner based on Lanthanum Bromide. in 2005 IEEE Nuclear Science Symposium and Medical Imaging Conference. 2005. San Juan, Puerto Rico: IEEE Nuclear & Plasma Sciences Society. [5] L.G. Hyman, Time resolution of photomultiplier systems, Rev. Sci. Instr., vol. 36, No. 3, pp. 193-196, Feb. 1965. [6] R. F. Post and L. 1. Schiff, Statistical limitations on the resolving time of a scintillation counter, Phys. Rev., vol. 80, p. 1113, Oct. 1950. [7] H. Nyquist, Certain topics in telegraph transmission theory, AlEE Trans., p. 617, Apr. 1928. [8] Valley, George E. Jr. & Wallman, Henry (1948). Vacuum Tube Amplifiers, MIT Radiation Laboratory Series 18. New York: McGraw Hill. [9] Walden, R. H. Analog-to-digital converter survey and analysis. IEEE Journal on Selected Areas in Communic. 17,539-550 (1999). [10] D.A. Gedcke, W. l McDonald, A constant fraction of pulse height trigger for optimum time resolution, Nuclear Instruments and Methods, Volume 55, 1967, Pages 377-380, ISSN 0029-554X, DOl: 10. 1016/0029-554X(67)90145-0. 1995