IEEE Transactions on Nuclear Science, Vol. NS-34, No. 1, February 1987

SCINTILLATION PHOTON DETECTION AND EVENT SELECTION IN HIGH RESOLUTION

POSITRON EMISSION TOMOGRAPHY

B.T. Turko*, G. Zizka*, C.C. Lo* and B. Leskovar*J.L. Cahoon**, R.H. Huesman**, S.E. Derenzo**

A.B. Geyer** and T.F. Budinger**

*Engineering DivisionLawrence Berkeley Laboratory

University of CaliforniaBerkeley, CA 94720 U.S.A.

(415) 486-6101

**Lawrence Berkeley and Donner LaboratoryUniversity of California

Berkeley, CA 94720 U.S.A.

ABSTRACT

The scintillation photon detection and eventselection subsystem for the high spatial resolutionDonner 600-channel positron emission tomograph isdescribed. The tomograph spatial resolution ofbetter than 3.0 mm, FWHM, is obtained by using 600closely-packed bismuth germanate 3 mm thick scintil-lation detectors. Each detector has its own 13.5 mmdiameter photomultiplier and event selection chan-nel, consisting of two fast pulse amplifiers, acharge integrator, a differential pulse height dis-criminator and precision time delay for event stro-bing. The event selection channel is capable ofstarting a timing cycle from the first photoelectronemitted from the photocathode. It is also capable ofdiscriminating between true events and noise and ofeliminating events occurring simultaneously in twoadjacent channels. Adjustments of the event strobingtime and two discriminator levels are performed by acomputer.

INTRODUCTION

In many recently developed positron emissiontomographic systems small scintillators and smalldiameter photomultipliers are used as detectors toobtain a high spatial resolution and sampling den-sityl. Minimizing the coincidence resolving timeto reduce the number of random coincidences is animportant factor in the design of such systems.Generally, such a system requires a large number offast scintillators and high time resolution photo-multipl iers2.

In this paper a new scintillation detection andevent selection subsystem for the high spatial reso-lution Donner 600-channel positron emission tomographis described. The high spatial resolution is achievedby using 600 closely-packed bismuth germanate scin-tillation detectors. Each detector has its own photo-multiplier and event selection channel2. Detectors,including the photomultipliers, are arranged in aclose packed single layer ring having a diameter of60 cm. The dimensions of an individual detector are3 mm x 10 mm x 25 mm. The inner diameter of the ringis 30 cm.

SCINTILLATION PHOTON DETECTION

The major time resolution limitations of a tomo-graphic system are due to variations of the locationof the scintillation events in the scintillators,fluctuations in the light collection process, and thetime resolution capabilities of the photon detector.

Several photon detectors, such as silicon PIN andavalanche photodiodes as well as high gain photomul-tipliers, were initially considered for the scintil-lation detection subsystem. In the first part of1982 when a decision was due on the selection of thesystem's 600 photon detectors, the silicon PIN andavalanche photodiodes did not have a capability forthe first photoelectron detection. (The first photo-electron detection gives the highest time resolutionneeded for coincidence detection in positron emissiontomography4). These devices, when operated at roomtemperature, did not have gain bandwidth product, lownoise level, gain-quantum efficiency uniformity, andlong term stability necessary for the detectionsybsystem. Although the energy resolution of thesolid state photon detector can be significantlyincreased by their operation at cryogenic tempera-tures5 it was unlikely that device adequate timeperformance could ever be achieved for 511 keV anni-hilation photons. Consequently, the solid statephoton detectors were eliminated from further consid-eration for possible application in the 600-channelpositron emission tomograph. Efforts were directedtoward selection of the most suitable high-gainphotomultiplier having a small diameter photocathodeand to determine its optimum operating condi-tions6,7. From several candidates, of which somewere in the experimental stage of development, Hama-matsu R647-01 photomultiplier was finally selectedfor scintillation detection subsystem on basis of itsperformance, cost and availability.

The R647-01 is a 13.5 ± 0.5 mm-diameter,10-stage, head on, flat-face plate type photomul-tiplier with a bialkali photocathode having a S-11response. Because of its small size, the photo-multiplier is particularly suitable for high spatialresolution tomography systems. The photocathode hasa peak response of 420 nm and a useful photocathodediameter of 9 mm. The multiplier utilizes a box typein-line dynode structure.

Our measurements have shown that the photomulti-plier has a gain of 1 to 1.2 x 106 at 1000 V andaverage dark current at this voltage of 350 pA. Theaverage quantum efficiency was found to be 28% at 410nm for five photomultipliers. Furthermore, thephotomultiplier operating at 1000 V, which cor-

responds to a gain of 10b, has measured the darkpulse count:

16 photoelectrons

= 54 counts per second

1/8 photoelectron

0018-9499/87/0200-0326$01.00 1987 IEEE

LBL-21042

326

327

Time resolution capabilities of the photomul-tiplier are essentially determined by the randomdeviations in the transit time of electrons travelingfrom photocathode to collector.

The electron transit time spread is mainly causedby fluctuations of individual times of flight ofphotoelectrons and secondary electrons due to theirdifferent trajectories and their initial velocitydifferences. The factors that contribute to thetransit time spread are differences in trajectorylengths and in electrical field strengths in dif-ferent portions of the photocathode-first-dynoderegion, and between various dynode sections. Inaddition, the transit time spread depends upon thenumber of photoelectrons released from the photo-cathode since the time spread varies approximatelyinversely as the square root of the number of photo-electrons. The time behavior information of singlephotoelectrons can be used for predicting the transittime spread for a arbitrary number of photoelectrons.Consequently, the best characterization of the photo-multiplier electron transit time spread is given bythe single photoelectron time performance3.

The R647-01 single photoelectron pulse response,using a modified manufacturer voltage divider toachieve a better match to the 50 Ohm external load,and the single photoelectron time spread with fullphotocathode illumination, are shown in Fig. 1 and 2,respectively. The 10-90% rise time of a singlephotoelectron pulse of the photomultiplier was foundto be 2 ns ±0.3 ns at 1000 V. The singlephotoelectron time spread was 1.22 ns, FWHM6.

ANALOG PROCESSING OF EVENTS

The detector photomultiplier output contains manyunwanted signals. Each detector is served by anindividual analog processing channel, and thiscircuit should pass only events meeting requiredpreset criteria and eliminate all others as early aspossible. The criteria for elimination of the"nonevents" are preset and each channel is individ-ually matched to its assigned detector for bestperformance. In order to compensate for the detectoraging, the critical parameters are individuallycontrolled by computer, allowing remote alignment andtest of the whole system.

Each of the system's 600 detectors requires anindividual analog processing channel. The detectorsignal is brought in directly from the photomultiplierby a 50-ohm cable. Since the signal level receivedacross the matching termination resistor is in themillivolt range, great care was taken to minimizenoise pick-up and crosstalk between the channels.

The analog processing channels are organized ingroups of eight, each group mounted on a single 19" x10" printed circuit board. There are 75 such boardsin the system. Each board also contains digitalcircuits for the control of individual channels bythe computer. Stable on-board power regulators weredesigned to maintain a constant supply voltage suchthat the analog circuits are independent of changesin the bus voltage and variations in contact resis-tance. These regulators also reduce noise on thepower lines, which in turn results in a stable andjitter-free performance of the analog circuits.Special care has been taken to keep power dissipationat a minimum.

A general block diagram of an 8-channel analogprocessing subsystem is shown in Fig. 3. Eachchannel (No. 0 to No. 7) can be remotely controlledby three 8-bit digital-to-analog converters (D/A 1 toD/A 3) with data latches. D/A 1 controls the strob-ing time of the event being processed, and D/A 2 andD/A 3 control the lower and upper discriminatorlevels of the single channel pulse height analysermonitoring the event's charge. All 24 D/A converterson the board have a common Data bus (DO to D7). Eachconverter is strobed by one of the correspondingdecoder outputs (D1 to D3). The three decoders arein turn controlled by three Crystal Number bus lines(AO to A2) and strobed one at a time by a FunctionDecoder (D). The fourth decoder function controls anoctal Veto 3 latch (L). When a latch is set, thecorresponding analog channel is out of operation.The decoder (D) is addressed by two Function lines(FO and Fl). A 13-bit word is thus needed to definethe command controlling any of the 24 D/A convertersand 8 veto latches. The command is executed upon thereceipt of a Strobe pulse. All vetoes can be removedsimultaneously by resetting the veto latches with aGlobal Clear pulse.

Each analog channel also responds to twoadditional veto inputs. The function of these vetoesis to block certain analog functions in the proces-sing of events which will be discussed later. Theeight pairs of corresponding veto inputs are bussedand control all eight analog channels simultaneously.

The photomultipliers, signals from which arebrought to the analog channel inputs, are labelledPMT No. 0 to PMT No. 7. An event that meets all thepreset conditions will produce a fast, preciselytimed ECL pulse at the channel output (Event Out).The eight event outputs are fed by 1OO-ohm microstriptransmission lines to the pins of an edge connector.From there the signals are transferred to the centralfast logic processor.

A group of fast logic gates (Gi to G3), providedthe Veto 1 signal is not applied, is designed toinhibit the event output of any channel if either itsleft or right neighbor channel is also processing anevent at the same time. This means that probably thesame gamma ray has hit two adjacent detectors and theevent should be aborted. The end channel on eitherside of each board has a provision for controlling ofthe end channel of the adjacent card. Of course itsend channels can be controlled by the adjacentboards. All 600 analog channels thus are daisychained, making a circle where each one controls itsleft and right hand neighbor and is being in turncontrolled by them.

Analog Processor Description

The photomultipliers generate random signals ofvarious magnitude and duration. The purpose ofanalog processing is to examine all signals andimmediately eliminate those not meeting the requiredamplitude level and time duration. The selectiontakes a very short time, allowing the processor tosort out all incoming signal pulses in search for apotential event. Analog processing channel blockdiagram is shown in Fig. 4. Detailed schematicdiagram is given in Fig. 5.

Referring to Fig. 4, an input signal is immedi-ately and evenly split upon its arrival between twowide-band amplifiers, separating two parallel lanesof analog processing. The gain of timing amplifier(TA) is high and a few millivolts of signal amplitude

328

is sufficient to drive the output to saturation. Theleading edge of the output pulse is thus made verysharp and it serves as a time reference marking thestart of an event. The second half of the signal isamplified by a fast linear amplifier (CA) and furtherintegrated in a fast gated charge integrator (CI).The integrator is enabled when a clocked event latch(EL) is set by the leading edge of the timing portionof the signal passing TA.

Selection of signals of various amplitude andduration is illustrated by the timing diagram in Fig.6. The upper sequence shows two short random "noise"signals, each having sufficient level to saturate thetiming amplifier (line A). Event latch is set immedi-ately, which in turn starts integration of the chargedefined by the magnitude and waveform of the signal(line Ap). A delay timer is also started by EL,preset to produce a pulse fed back to its own clockinput 50 ns later (lines C and D). The trailing edgeof the clock checks the status of the latch D input,tied also to the signal A. Since both "noise"signals have already decayed below the triggeringthreshold at this time, the latch is reset by theclock and the integrator quickly discharged. Only afew nanoseconds are required for the recovery beforea new signal can be tested.

If the signal duration exceeds 50 ns, the latchremains set and the charge integration continues(Event Cycle portion of Fig. 6). Bold lines A and Apshow a signal lasting several hundred nanoseconds,and the integrator output increases accordingly(line L). A precision 500 ns timer was started whenthe latch EL was set. The purpose of this timer isto produce a clock pulse exactly 500 ns after thestart of the event. Each of the 600 processingchannels must have this identical delay and be verystable. Small adjustments in delay of up to + 30 nsdue to the system time propagation differences can bemade by the computer. An 8-bit D/A convertercontrols the delay in 0.25 ns steps (Fig. 4).

During the 500 ns integration time the integratoroutput is monitored by two discriminators (LD andUD). Each discriminator is controlled by an 8-bitD/A converter, which allow each to be set by thecomputer to any level within the range of the inte-grator. When the charge exceeds the lower discrim-inator level (Fig. 6, line M), the first criterionfor recording an event is satisfied. The D input ofthe event latch timer (ET, Fig. 4) is armed through agate G3 and set by the delay clock exactly 500 nsafter the event started. A short time delay circuitresets the latch 30 ns later. The latch thus gene-rates a 30 ns wide event pulse (Fig. 6, line N) whichis sent to the tomograph coincidence circuits insearch of a matching pulse that may result in con-firming the occurrence of a real event.

If the charge integrated during the 500 ns periodexceeds the expected level, the upper discriminatorfires (dash-dot lines L and S), closing the anticoin-cidence gate G3 which prevents the firing of theevent latch. No event pulse is sent out in this case.

The remote control of discriminator levels canturn each analog processor into a multichannel pulseheight analyzer. When using an appropriate radio-active calibrating source, the computer scans thediscriminators automatically through the integratorrange. The gain of each charge amplifier is thenadjusted and discriminator levels are set to center

the energy band of interest into the desired ampli-tude range. Valid events are those falling withinthe range.

Although the processing is completed after theevent pulse is sent out and event latch timer reset,the processor must remain paralyzed to prevent retrig-gering by photons in the tail of the scintillatorlight pulse. The end of the 500 ns delay starts asecond, 1.5 us deadtime timer. A pulse is generatedat the end of 1.5 P's which clears the latches anddischarges the integrator. The processor is thenimmediately ready for a new event.

There is a large category of signals long enoughto pass the 50 ns duration test. Their amplitudehowever may be too small to produce charge suffi-ciently large to exceed the lower discriminatorthreshold. Since this condition can be detectedearly, the processing cycle deadtime can be madeshorter than 1.5 ps (dotted lines in Fig. 6). Thesignal lasts long enough to pass the 50 ns durationtest (lines A and D). At this moment an amplitudediscriminator LL is tested which clocks the chargeamplifier signal (Fig. 4). If the signal level atthe 50 ns point is below the preset threshold, thediscriminator does not fire (line E in Fig. 6) andthus does not enable the deadtime delay timer. Theconsequence is that the deadtime is reduced to 0.1 Psinstead of the normal duration of 1.5 ps. The pro-cessing rate of signals is therefore substantiallyincreased.

When processing a regular event the signalamplitude is sufficiently large to set the LL latch(line E) and enable the deadtimer to full 1.5 psdelay. Also, a 40 ns wide pulse is generated at thismoment by a left/right veto timer (lines F and G).This pulse enables gates Gl and G2 for a short time(Fig. 3). The gate Gl is controlled by the vetotimer of the processing channel on the left and thegate G2 by the processing channel on the right.Should the neighbors produce an overlapping pulse,the gate G3 passes this coincidence back to theanalog processor setting a left/right veto latch(lines Q and R). This latch in turn keeps reset theevent timer. The signal processing continues, butthe event will be aborted regardless. How effectivethis method is can be tested by applying an externalVeto 1. The veto timer will be disabled and allevents can be processed regardless of possible over-lap between adjacent channels. When applied, Veto 1affects all 600 analog channels.

Veto 2 controls also all analog processors of thetomograph. When applied, the noise eliminationfeature is disabled, allowing the processing of allsignals as though all were potential events. Thesevetos are useful in testing the effectiveness of theevent filtering procedure.

Veto 3 prevents the operation of the whole proces-sing channel. Since each channel has an individualVeto 3 latch, addressable by computer, any combi-nation of the tomograph detectors can be temporarilyshut down when running test routines.

A number of test points throughout the analogprocessing channel makes alignment and monitoringeasy. The timing and levels of all parametersdiscussed above can be individually preset so thatall channels can be made identical. The adjustmentrange is sufficiently wide to permit realigning anyparameter, if required, without the need to changecircuit components.

329

CONCLUSION

Reliability was the principal consideration indesigning the analog processing circuits serving the600 photodetectors of the tomograph. The potentialstatistical failure of a system comprising such alarge number of electronic components was a majorconcern. The building blocks of each of 600 analogprocessing channels are two wide band amplifiers, afast gate charge integrator, two computer controlledlevel discriminators, a precision computer controlledtime delay, a programmable deadtimer, three fastpulse generators, a fast amplitude discriminator, afast time discriminator and logic circuits includingfast coincidence circuits and gated latches.

Analog processors are combined in groups of eightmounted on a single printed circuit card, including atotal of 24 D/A converters and digital computerinterface logic. All analog circuits use only +5Vand -5.2V power lines in order to minimize powerdissipation. Precision regulators maintain thesevoltages stable and noise free, requiring only 6Vsupplies. The system cost was kept low by usinginexpensive standard integrated circuits and discretecomponents.

A test station was designed and an alignmentprocedure devised to ensure that all 600 channels canbe made identical. The procedure, aided by Hewlett-Packard 86 computer, includes the adjustment of 13controls per channel for proper setting of the timersand discriminators and includes the trimming of theD/A converters for optimum tracking.

Final adjustment of gain and threshold isrequired in order to match each channel to itsassigned detector as was described earlier. Oncealigned, the analog processors have exhibitedexcellent stability. No measurable drift of presetparameters was found so far when random channels wererechecked. The photomultiplier drift and aging isthe only potential source of need for periodicadjustment.

Only a few analog channel failures were reportedso far (during the first thousand hours of operation).They were caused by a couple of failed integratedcircuit chips and some capacitors shorting the powerlines. In the system comprising over 7,500 inte-grated circuits, 7,500 discrete semiconductors,63,000 resistors and 26,000 capacitors the numbLer offailures was surprisingly low. The performance ofthe tomograph system is summarised in Ref. 8.

ACKNOWLEDGEMENTS

This work was performed as part of the program ofthe Engineering Division Electronics Research andDevelopment Group of the Lawrence Berkeley Laboratoryand the Biology and Medicine Division ResearchMedicine Group of the Lawrence Berkeley and DonnerLaboratory. The work was supported in part by theU.S. Department of Energy under Contract No.DE-AC03-76SF00098 and in part by the U.S. NationalInstitute of Health, National Heart, Lung and BloodInstitute under grant No. P01 HL25840 and R01 CA38086. The authors would like to express theirappreciation to the Hamamatsu TV Co. for the loan ofthe R 1635 photomultiplier. Reference to a company

or product name does not imply approval orrecommendation of the product by the University ofCalifornia or the U.S. Department of Energy to theexclusion of others that may be suitable.

REFERENCES

1. S.E. Derenzo, T.F. Budinger and T. Vuletich, HighResolution Positron Emission Tomography UsingSmall Bismuth Germanate Crystals and IndividualPhotosensors, IEEE Trans. in Nucl. Sci., Vol.NS-30, No. 1, pp. 665-670, February 1983.

2. J.L. Cahoon, R.H. Huesman, S.E. Derenzo, A.B.Geyer, D.C. Uber, B.T. Turko and T.F. Budinger,The Electronics for the Donner 600-CrystalPositron Tomograph, IEEE Trans. in Nucl. Sci.,Vol. NS-33, No. 1, pp. 570-574, February 1986.

3. B. Leskovar and C.C. Lo, Single PhotoelectronTime Spread Measurement of Fast Photomultipliers,Nucl. Instr. and Methods, Vol. 123, No. 1, pp.145-160, 1975.

4. H. Murayama, A Simple Timing Discriminator for aBGO Scintillation Detector, Nucl. Instr. andMethods, Vol. 177, pp. 145-160, 1980.

5. S.E. Derenzo, Gamma-Ray Spectroscopy Using Small,Cooled Bismuth Germanate Scintillators andSilicon Photodiodes, Nucl. Instr. and Methods,Vol. 219, pp. 117-122, 1984.

6. C.C. Lo and B. Leskovar, Performance Studies ofHamamatsu R647-01 Photomultiplier, IEEE Trans. onNucl. Sci., Vol. NS-31, No. 1, pp. 413-416,February 1984.

7. C.C. Lo and B. Leskovar, Evaluation of HamamatsuR1635 Photomultiplier, Review of ScientificInstruments, Vol. 55, No. 7, pp. 1100-1103, 1984.

8. S.E. Derenzo, R.H. Huesman, J.L. Cahoon, A.Gayer, T. Vuletich, and T.F. Budinger, The Donner600 Crystal Positron Tomograph, LBL-22363, To bepresented at the IEEE 1986 Nuclear ScienceSymposium, October 1986, Washington, D.C.

330

-a4 x 103

z0 I |~~~~~~~

. _ ~~~~~~~~~~~~~~~~~~L

XBB831-798 2

Fig. 1. Single photoelectron pulse response using |modified voltage divider and a preamplifier 0having 30 dB gain.

SINGLE PHOTOELECTRONTIME SPREAD = 1.22ns FWHM

XBB 835-4246

Fig. 2. Single photoelectron time spread of R647-01with full photocathode illumination.

OALEFT VETO IN

*15 IN OUT IN IN OUT INLifR L/RVETO VETO

REGULATOR -S.2V POWER ANALOG ANALOGStV RREOLGULATOR PROCESSING PROCESSINo--S f . >-IS.2 CHANNEL CHANNEL

VO #1

-.VETO 1 IN

LATCHES XBL 863-1192

Fig. 3. 8-channel PET signal processing subsystem.

331

R101 L/R VETO_VETO 3

1

-5.2V OUTVETO 1 1,5K.~~~~~~~~~~~~~~~~~~~~~~~~~~~~-5

(ON: GND, ~~~~~~ ~ ~~~~~~~~MC10231 12

N4CU 1200 1 5 16

QC 60n 3680>856

h~~~~~U Rl2 57l

1~~~~~~~~~~~~~~-R - j L 2 452

R 6 1

"52V

~ ~ ~ ~ ~~3-5. 68 47 ii

QR Ac95 R T5 12lOFl8

3~~~~~~~~~~~~~~6 0 51 U 6 21f 8C016100RR575p22 2 K f R1 2971

SM8 1K8i2.46016 R-K186

-5.2V TPI lOOpF 510n R44 R62 R-~~~~~~~~.29~RI 476906.

1 0R48Q

R46 ~ ~~~~~ ~ ~~~~~~~~~~~I3C17 R3O RT8 77A7 51

R20018 0

U2 D Q R61~~~~~~~~~~~~~~~8

5.CQO R1 5.29V--,

R 1 3.8 0-17 )3I O NT F6

Fig. 5. Analog processing channel schematic diagra

, VETO 3 DATAo DOVEAE

A5MHF'CLOCKED EVERATI-EENTT

3OnD

AEASTIME CLE R

EVN QNDATLORAY

U LEFT/RIGHT_J:BL

DETECTOR VETO 2ATASonTDATAO

VETTOR-N TOICRL

Fig. 4. Analog processing channel block diagram.

EXAMPLE U3 OF CHANNEL 6 WOULD BE U63.

XBL 863-1195

a. NOISE CYCLEA

Ap _-_ ___

B=_

b. EVENT CYCLE.+. . ...i.---...

A- i - __Bi

D -50ns

.......................... .................................................... . . ...........

H 1O500ns s ; ----LI j ~~~~~~~~~~~~~~~140ns. 1.5jusLKL --LL

D ----V L

N MJ.EVENT OUT

L/R VETO IN-'R

SCALE I I l l l li .0 100 200 300 400 500 600 700 2000ns

XBL 863-1194Fig. 6. PET timing diagram.

vtiv .5

L/R VFTn IN

rry.,.

n

T -1--