Count Rate Performance and Deadtime Analysis of The New 3D PETRRA PET Camera

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 11

Count rate performance and deadtime analysis

of the new 3D PETRRA PET Camera

K.Wells *a, C. Kakanab, R.J. Ott **c, M. A. Flowerc, A Divolic, S Meriauxc, J.E.Batemand,
R Stephensond, D Duxburyd , E Spilld
a
School of Electronics, Computing & Mathematics, University of Surrey
b
School of Physical Sciences, University of Surrey
c
Joint Dept of Physics, Royal Marsden NHS Trust Hospital/Institute of Cancer Research
d
Rutherford Appleton Laboratories

ABSTRACT

We report on the count-rate performance of the unique PETRRA positron camera at activities up to 60MBq. The
camera consists of two large area detectors, each comprising a tiled array of 10mm thick BaF2 scintillation crystals
interfaced to a multi-step avalanche chamber filled with 4.2mbar of pure TMAE vapour. Preliminary results
demonstrate coincident count rates of over 80kcps for a cylindrical (20x20cm3) phantom with 50MBq of F-18 in the
field-of-view using a 20ns coincidence time window. Each component of the readout cycle has been characterised in
terms of dead-time loss. The camera’s dead-time related count loss is well-described by a paralysable model with a
dead-time of ~500ns. Other sources of count rate loss are also discussed.

Keywords: PET, PETRRA, count rate, dead-time, BaF2, TMAE

1. INTRODUCTION

Over the last two decades there have been a number of attempts to harness multi-wire proportional chamber (MWPC)
technology for use medical imaging, in particular, Positron Emission Tomography (PET). Among these, the HIDAC
system[1] and the MUP-PET system[2] have been the most well-known. Within the Institute of Cancer Research/ Royal
Marsden NHS Trust Hospital, the MUP-PET system has shown that clinically useful PET imaging can be achieved
using this technology[3]. This system has also shown the benefits of using true 3D PET reconstruction and imaging.
However, the very modest sensitivity of the system has proved to have severe consequences for imaging. We have
therefore been developing a new MWPC PET system, based on BaF2-TMAE (Tetrakis (dimethyl-amino)ethylene)
technology originally developed for use in high energy physics calorimetry[4] .

Detection of annihilation radiation in BaF2-TMAE detectors is a two stage process. Energy deposition by ionising
radiation, generates UV scintillation light in the BaF2, which when emitted, is then absorbed in the sensitive TMAE
vapour. Successful detection of events relies on the overlap of the BaF2 ultraviolet scintillation spectra with the TMAE
absorption spectra[4]. TMAE also has a low ionisation threshold (7.5 eV) which means that the absorbed energy of the
scintillation light can be efficiently converted into energetic free photoelectrons. These primary conversion electrons
are then gas multiplied in a multi-step gas avalanche chamber (MSA) and position determined using conventional
MWPC readout techniques.

The PETRRA BaF2-TMAE system has been in development for a number of years, progressing from bench top hybrid
through to full-size dual head camera system[5,6]. The final system consists of two gantry mounted, large area
(60x40cm2) detectors that rotate about 1800 . Clinically, the main advantage of this technology is the 40cm field-of -

* k.wells@ecm.surrey.ac.uk; phone 44-1483 686036; fax 44 1483 686031; http://www.ee.surrey.ac.uk/CVSSP; School of Electronics, Computing
& Math, University of Surrey, Guildford Surrey, GU2 7XH, UK
** bob@icr.ac.uk; phone 44-208-642 6011; fax 44 208 643 3812; http://www.icr.ac.uk; Royal Marsden Hospital /InstituteofCancerResearch, Sutton,
Surrey SM2 5PT UK

Penetrating Radiation Systems and Applications III, H. Bradford Barber, Hans Roehrig,
F. Patrick Doty, Richard C. Schirato, Edward J. Morton, Editors, Proceedings of SPIE 123
Vol. 4508 (2001) © 2001 SPIE · 0277-786X/01/$15.00
view, allowing large organs such as lung or liver to be imaged without patient movement. We now report on further
detailed tests of count-rate performance and the various losses in the electronic read-out system.

2. DETECTOR STRUCTURE

Fig. 1 shows a cross sectional schematic diagram of the main elements of the two large area detectors. Visvikis[7]
discusses the construction of the first large area detector, whilst Duxbury [8] discusses some of the further engineering
features of the technology.

Below the tiled array of BaF2 scintillation crystals is a MSA structure divided into 3 main regions: the primary detection
region, the gated transfer region and the coincident readout section. The primary detection region is divided into six
parallel sectors, each with its own preamplifier-discriminator unit. The output from each of these units is used to control
the opening of a corresponding section of the gate structure within the transfer region. This minimises deadtime losses
at the front of the detector, and also minimises corruption in the readout cycle, which is discussed in more detail in
section 3.

Within the absorption region approximately 1-2 photoelectrons are detected per gamma photon interaction. Although
there are intrinsically around 5-6 photoelectrons produced per 511keV deposited, the structure of the chamber is such
that only the first one or two photoelectrons created in the absorption region are likely to undergo sufficient gas
multiplication to ensure successful completion of the readout cycle. These initial photoelectrons are drifted across a
high electric field initiating a small amount of gas multiplication. Further, more substantial, gas multiplication occurs in
the preamplification region. As the resulting electron shower passes out of the primary conversion region, their passage
is sensed by one of the six preamplifier/discriminator modules coupled to the final wire plane in the primary detection
region. This is used to arm one side of a coincidence unit. If a corresponding event is found in temporal coincidence in
the second detector, then an enable signal is produced to initiate the readout cycle.

A screened gate structure is positioned in the middle of the transfer region that efficiently blocks unwanted singles
events from passing into the readout cycle. The gate is also segmented into six parallel sections corresponding to those

124 Proc. SPIE Vol. 4508


in the primary detection region.

Assuming a coincidence event has been detected in both detectors, then the electric fields within one of the
corresponding gate regions is pulsed for 100ns to allow free passage of the electron cloud through to the MWPC
readout section. Position readout is accomplished using orthogonal delay lines coupled to the x and y MWPC cathodes.
Constant fraction discriminators (CFDs) connected to Time to Analogue Converters (TACS) determine the difference in
arrival times of the pulses at the end of each delay line. The resulting data is digitised and used to compute the x,y
position of the coincidence event in each detector. Before passing the coordinate data to the computer system several
checks are made to ensure that the data has not been corrupted. Assuming the data is valid, then a flag is set initiating
data transfer to the computer system.

3. COINCIDENCE COUNTING

Counting events in time coincidence has a number of implications for the way in which pulse information is handled.
We now describe some of the main issues for coincidence counting. For further details, refer to Bateman [8] and
Visvikis[9] .

The finite time, τ, taken to process an analogue pulse in any practical counting system means that there are inevitable
losses of the true count rate, R, due to the insensitivity of counting electronics whilst this processing takes place. This
fractional loss of events, F, becomes particularly serious at high data rates and can usually be characterised by one of
two models[9]: either the paralysable (sometimes referred to as extendable) model, described by [1] below, or the non-
paralysable (or non-extendable) model described by [2],

τ
Fp = e - R [1]

Fn = 1 / (1 + Rτ) [2]

Although exhibiting similar behaviour at low rates, where counting losses are ≤10% level, at higher rates the
paralysable model will describe an ever decreasing event rate tending to zero, following a maximum at 1/τ, whilst the
non-paralysable model will tend to approach a value of Rτ at high count rates. Further details may be found in [9] .

The observed singles count rate of each sector, No, in the primary detection region can be related to the actual singles
rate, Ns, as

No = Ns F1 [3]

where F1 is the count rate dependent deadtime factor for the CFDs, described with reference to one of the two models in
[1] or [2] above. Assuming that each sector sees approximately the same solid angle flux of photons, then the overall
effective deadtime , τ, is effectively reduced by ~factor 6 due to the segmentation of the primary detection region.

The total observed coincidence counting rate registered in the primary detection region is given by the sum of the fast
(trues + scatter) coincidences Nf, and the random or accidental coincidence rates Nr:

Ntot = Nf + Nr . [4]

Once the electron shower has been recognised as coincidence event, and so passes from the primary detection region to
the MWPC for readout, there is a chance for some of these electrons to be lost due to, for example, slight mistiming
with the opening of the gate field. This may produce a loss of events across the transfer region as the attenuated shower
may fail to trigger the anode discriminator. For a given detector, this non-rate dependent loss, referred to as L1, can be

Proc. SPIE Vol. 4508 125


determined from :

Na = L1(Ntot + Fg) , [5]

where Na is the counting rate recorded at the detector’s anode plane. The extra term Fg appears because whilst the gate
is open, there is also a chance for unrelated avalanches which are temporally coincident with the gate opening time to
also pass through into the MWPC. In practice it can shown that for gate opening times of 100ns, and given that each
front sector sees approximately 1/6th of the singles photon flux, then Fg (≈ NtotNs x 100ns / 6) [6] is expected to be of the
order of a few per cent of Ntot.

Successful readout of x,y positional information is subject to two further sources of loss. The first of these, L2 , refers to
non-rate dependent losses from incomplete coordinate sets. This may be due to pulse attenuation in the delay lines
causing a failure to trigger one of the CFDs. As with L1 , L2 is also a non-rate dependent loss. The final source of loss,
F2 describes rate-dependent event corruption due to extra pulses in the delay lines (such as arising from the term Fg in
[5] above). So assuming Na is similar in both detectors, the final coincidence count rate passed to the data acquisition
computer, RDAQ is given by

RDAQ = [L2 ] 2F2 Na [6]

The squaring arises because RDAQ is the result of using two independent detectors. Note that F2 is not squared because
the pile-ups caused by deadtime are synchronised in both detectors as a result of counting coincidence events.

4. METHODS & MATERIALS

4.1 DECAYING SOURCE METHOD

In order to determine the dead time and other losses associated with the PETRRA detector and counting electronics, we
have used the classic decaying source method. A full description of this method is contained in Knoll[9] , but we
summarise the main points below.

Assuming that the background counting rate of a counting system is negligible, we can measure the temporal decay of a
short-lived radioactive source by recording an initial count rate, Ni, and then systematically measuring the drop in count
rate, N(t) at some time t later (where λ is the decay constant associated with the particular source):

λt
N(t) = Ni e - [7]

By combining this with the non-paralysable model, we obtain

λτ
N(t) e = -NiN(t) τ + No [8]

By similar reasoning we can obtain an expression for the paralysable case:

−λt
λt + ln N(t) = -Ni τ e + lnNi [9]

The dead time, τ, of the component under investigation is determined using a straight-line plot from [8] or [9], where
the gradient will be equal to -Niτ.

4.2 EXPERIMENTAL TECHNIQUE

In order to study the count rate performance and deadtime characteristics of the PETRRA camera system, we uniformly

126 Proc. SPIE Vol. 4508


filled a 20cm x 10cm cylindrical phantom with Fluorine-18. Before dilution, the aliquot containing the activity was
placed in a calibrated well counter. The absolute activities at various points during the experiment were then determined
by extrapolation from this initial reading.

The phantom was placed in the centre of the field of view of the two large area detectors separated by 95cm between
crystals. Over a period of hours the singles and coincidence count rates were recorded from different points in the
readout system. The random coincidence component was calculated using [5], with a coincidence window of 20ns.

90000
Detector 1
80000
[counts per second]

70000 mean

60000
Detector 2

50000

40000
Count rate / 10

30000

20000

10000

0
0 10 20 30 40 50 60

Activity (MBq)

Fig. 1 Plot of single count rates using a 10cm x 20cm head phantom

5. RESULTS & DISCUSSION

5.1 SINGLES COUNT RATES

Fig. 1 shows the mean uncorrected singles count rates for the 10cm x 20cm cylindrical phantom along with the separate
count rates for the two detectors. Clearly there is some different in performance, with detector 1 decreasing in count
rate slightly at activities above 35MBq. By contrast detector 2 shows a more slowly varying loss of count rate. The
decaying source method was employed to investigate the dead time behaviour of both detectors, and analysed by fitting
both models to the singles data. Figure 2 shows the effect of applying the non-paralysable model to the data for detector
1 using a least squares fit. This clearly shows a poor fit, with similar behaviour exhibited by detector 2. By contrast, Fig.
3 shows the result when a paralysable model is applied to the data for detector 1. This visually shows a very good fit to
the data, with a corresponding r2 value of .999. This leads to an experimentally determined value for the deadtime of
500ns. When considered that this represents approximately one sixth of the total count rate, this leads to an estimate for
the singles deadtime of 3 µs.

This is a rather surprising result, as the pre-amplifier discriminator combinations are expected to behave an non-
paralysable sources. It has been suggested that this apparent paralysable behaviour may be due to a loss of detector gain
at high counting rates due to slow ion clearance at the front of the detector[10] and possibly in part due to saturation
induced deadtime in the front preamplifiers[7].

Proc. SPIE Vol. 4508 127


3000000

2500000
Singles D1*exp(lamda*t)

2000000

1500000

1000000

500000

0
0 100000 200000 300000 400000 500000 600000 700000 800000

Singles Rate from Detector 1 / counts per second

Fig. 2 Non-paralyzable model fitted to singles data from detector 1

15

14.8
y = -1.5732x + 14.916
R2 = 0.9987
14.6
λ*t+ln(singles D1)

14.4

14.2

14

13.8

13.6

13.4
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Exp(−λ*t)

Fig. 3 Paralysable model fitted to data for singles rates from Detector 1

128 Proc. SPIE Vol. 4508


It is impossible to analytically solve [1] to correct for the singles dead time loss. Instead, we resort to using a Taylor
expansion as an approximation [10] , so that from [7], ignoring higher terms, we may write:

τ
R /Ro =e - R ≈ 1 + (-Rτ) + ...... [10]

Fig. 4 shows the effect of correcting the raw data from detector 1 using the above approximation. The predicted singles
were estimated by extrapolation from data occurring at low count rates, where dead time was thought to be
insignificant. This shows that use of the Taylor expansion provides a good correction up to approximately 28MBq in the
field of view, equivalent to singles count rates up to approximately 1.2 Mcps. At higher count rates and activities, the
approximation underestimates the actual singles rates as higher terms of the expansion become significant. Thus, the
Taylor approximation is only suitable for correcting paralysable behaviour when Rτ<<1. A similar analysis was
undertaken for detector 2. This resulted in similar behaviour with a value for the paralysable deadtime of approximately
3 µs.

2500000

predicted
actual
Singles rates (Counts per second)

2000000 singles

1500000 corrected
singles

1000000
raw
singles

500000

0
0 10 20 30 40 50 60

Activity (MBq)

Fig. 4 Effect of singles correction by paralysable model using Taylor expansion


Data relates to detector 2

5.2 COINCIDENCE COUNT RATES

Fig. 5 shows the coincidence rates associated with using the head phantom. The random rate was calculated using a
coincidence time window of 2T = 20ns, and the fast coincidence rate was found by subtracting the randoms from the
total coincidence rate. The plot also shows the recorded data acquisition rate. The difference between this curve and the
total coincidence rate arises because of deadtime and other losses in the detector and counting electronics. These losses
are analysed in the following sections.

Proc. SPIE Vol. 4508 129


100000
Total
90000

Coincidence rates (counts per second)


Fast
80000

70000 Recorded
Coincidences
60000

50000

40000

30000

20000

10000
Randoms

0
0 10 20 30 40 50 60

Activity (MBq)

Fig 5 Plot of coincidence count rates as a function of activity

5.3 COUNT RATES IN THE READ OUT SYSTEM

Transfer Efficiency

In order to examine the efficiency with which electron showers are transported across the transfer region, [5] can be re-
written as

Na/Ntot = L1(1 + qNs) [11]

where q is equal to the gate opening time, Tg divided by 6. Thus plotting [11] in Fig. 6 demonstrates the virtually
horizontal line that is expected provided corruption by singles breakthrough at the gate is insignificant. The intercept
value for L1 =0.97 suggests that only about 3% of the events are lost in a detector, between primary detection and initial
readout at the multiwire section of the chamber.

Delay line Losses

It is worth noting that the gate could be set up in two modes: first, the gate responds to every coincidence event detected
at the front of the detector regardless of the state of the readout system; secondly, the gate only allows coincidences
through to the MWPC when the readout system is free, and not busy processing a prior event.

In the first case, this will result in non-paralysable behaviour because every coincidence event will always be spaced at
least τ2 apart. This of course ignores any other source of corrupting event that might cause extendable deadtime period.
However, Fig. 6 suggests that such corrupting effects can largely be ignored, as the gate works extremely efficiently at
rejecting uncorrelated singles events.

In the latter case, the readout system can be expected to behave as a paralysable source of deadtime, as any new
coincidence will feed straight into the delay lines and cause extension of any existing readout cycle. In fact, in this
situation, the effective deadtime is >τ2 in each detector as the readout cycle will be longer whilst the system waits for

130 Proc. SPIE Vol. 4508


1.2
Anode rate / raw coincidence rate

0.8

0.6

0.4

0.2

0
0 200000 400000 600000 800000 1000000 1200000 1400000 1600000
Corrected singles D2 (cps)
Fig. 6 Plot showing the efficiency of coincidence event collection at the
MWPC after the electron avalanches from coincidence
events have crossed the transfer region.

1.6

1.4

1.2

1
Na / NDAQ

0.8

0.6

0.4

0.2

0
0 20000 40000 60000 80000 100000
Ntot, / counts per second

Fig. 7 Plot of the ratio of anode rate to data acquisition rate as a function of
total coincidence count rate.

Proc. SPIE Vol. 4508 131


the corrupting pulses to propagate through the system. The system is therefore set up to operate as a non-paralysable
source of deadtime. Ignoring any other source of corruption (such as Fg), and assuming that both anodes count at similar
rates, then using [2], [6] is expressed as

Na /RDAQ = (1+ Ntotτ2)/ L22 [12]

A value for L22 is estimated by considering the reciprocal of the intercept from Fig. 7, so that L2 = 0.93. This suggests
that about seven per cent of the valid events are lost due to corruption from noise and incomplete coordinate data in
each detector’s delay lines. The gradient, τ2/L22 = 2.0 x 10-6 indicates thatτ2 is approximately 1.7µs. At the highest
levels of activity (~60MBq) this resulted in around 25% total loss of detected coincidence events of which around half
were due to rate dependent pileups in the in the delay line readout system.

6. CONCLUSION

The above results have shown that the PETRRA system is capable of acquiring coincident count rates >90kcps for a 20
x 10cm uniformly filled head phantom. However, losses in the counting system reduce this to approximately 65kcps.
This work has highlighted the efficient functioning of the MSA part of the PETRRA detectors. However, the dominant
sources of event loss is associated with the primary detection region, where >50% loss in counting rate is experienced at
activities >>30-35MBq. Future modifications to the design of the detectors and front-end electronics should help to
reduce these losses. If such improvements can be made then we confidently expect to be able to realise recorded
coincident event rates >100kcps.

However, these results do show the benefits of using a segmented gated detector. The division of the primary detection
region into effectively six independent counting units minimises electronic deadtime losses at the front of each large
area detector. Dividing the gate into 6 sectors reduces to insignificant levels the fraction of unwanted singles showers
transmitted through to the MWPC. Thus, whilst the front of the detector is counting at the full singles rate, the gate
situated midway through the transfer region can block unwanted uncorrelated electron showers and only allow through
the desired coincidence events, sparing the MWPC from singles saturation. This means the back of the detector, which
has an intrinsically lower count rate capability due to the readout system, counts at approximately 10-1 slower compared
to the primary detection region at the front. These results have then demonstrated that a large area detector can be read
out efficiently, without significant corruption of singles events, which have been seen to be a significant handicap in the
previous large area MUP-PET system [2,8].

It is worth noting that had the PETRRA system been developed without the segmented gate structure, then despite much
improved detection efficiency (~30% compared to 7% in MUP-PET), the singles could be expected to saturate the
readout electronics at coincidence rates of around 15-20kcps. This is in stark contrast to the >80 kcps coincidence rates
which have been observed in the current PETTRA system, demonstrating the enhanced count rate capability of
segmented gated readout at higher levels of activity. It should also be noted that the deadtime of the TAC system could
be reduced substantially (to <1 µsec) using modern time digitisers. Further work will concentrate on developing a
clinical deadtime correction for quantitative imaging studies, and investigation of the discrepancies between the
performance in the two detectors.

REFERENCES

[1] Townsend, D.W. (1988) Clinical PET with the HIDAC camera?; Nucl Inst Meth A269 pp 433 - 450
[2] Marsden et al (1989) The performance of a multiword proportional chamber for clinical use; Phys Med Biol, 34,
No. 8; pp 1043 - 1062

132 Proc. SPIE Vol. 4508


[3] Ott, R.J.et al (1988) Clinical PET with a large area multiwire proportional chamber PET camera; Nucl Instr Meth
A269; pp 436 -442
[4] Anderson, D.F. et al (1983) Coupling of a BaF2 scintillator to a TMAE photocathode and a low pressure wire
chamber; Nucl Instr Meth; 271; pp 217 - 223
[5] Wells K. et al (1994) Performance of a BaF2 - TMAE prototype detector for use in PET; IEEE Trans Nucl Sci; 41
pp 2737 - 2742
[6] Visvikis, D et al (1995) PETRRA: Preliminary experimental results from the first full size detector and dead time
simulation of the count rate performance of a unique whole body PET camera; IEEE Trans Nucl Sci 42; pp 1031 -
1037
[7] Duxbury, D.M. et al (1999) Preliminary results from the new large area PETRRA positron camera; IEEE Trans
Nucl Sci 46; pp 1050 - 1054
[8] Bateman, J.E. et al (1989) Development studies for a high rate positron camera based on a BaF2/TMAE system,
Nucl Inst Meth, A283 ; pp 463 - 444
[9] Knoll, G.K. Radiation Detection and Measurement, 2nd edition Wiley & Sons, New York
[10] Cherry, S.R. et al (1989) Image quantification with a large area multiwire proportional chamber positron camera
(MUP-PET); Euro. J. Nucl . Med. 15; pp 694 – 700
[11] Bateman, J.E. (2000) DRAL Internal report

ACKNOWLEDGEMENTS

This work is funded by the Engineering & Physical Sciences Research Council, the Cancer Research Campaign and the
Institute of Cancer Research. The authors would like to acknowledge workshop assistance from Mr Paul Collins and his
staff at the ICR Mechanical workshop and Mr Richard Symonds-Tayler also at the Institute of Cancer Research.

Proc. SPIE Vol. 4508 133

You might also like