Nuclear Instruments and Methods in Physics Research A 804 (2015) 127–131
Contents lists available at ScienceDirect
Nuclear Instruments and Methods in Physics Research A journal homepage: www.elsevier.com/locate/nima
Cherenkov TOF PET with silicon photomultipliers R. Dolenec a,b,c,n, S. Korpar a,b, P. Križan c,b, R. Pestotnik b a b c
Faculty of Chemistry and Chemical Engineering, University of Maribor, Maribor, Slovenia Jožef Stefan Institute, Jamova cesta 39, SI-1000 Ljubljana, Slovenia Faculty of Mathematics and Physics, University of Ljubljana, Ljubljana, Slovenia
art ic l e i nf o
a b s t r a c t
Article history: Received 12 March 2015 Received in revised form 17 September 2015 Accepted 18 September 2015 Available online 30 September 2015
As previously demonstrated, an excellent timing resolution below 100 ps FWHM is possible in time-offlight positron emission tomography (TOF PET) if the detection method is based on the principle of detecting photons of Cherenkov light, produced in a suitable material and detected by microchannel plate photomultipliers (MCP PMTs). In this work, the silicon photomultipliers (SiPMs) were tested for the first time as the photodetectors in Cherenkov TOF PET. The high photon detection efficiency (PDE) of SiPMs led to a large improvement in detection efficiency. On the other hand, the time response of currently available SiPMs is not as good as that of MCP PMTs. The SiPM dark counts introduce a new source of random coincidences in Cherenkov method, which would be overwhelming with present SiPM technology at room temperature. When the apparatus was cooled, its performance significantly improved. & 2015 Elsevier B.V. All rights reserved.
Keywords: PET Time-of-flight Cherenkov radiation Silicon photomultipliers Lead fluoride
1. Introduction The time resolution in time-of-flight positron emission tomography (TOF PET) measurements can be improved by basing the detection method on the use of Cherenkov light, produced promptly in a suitable Cherenkov radiator material. In this way, the contribution from scintillation light production mechanisms can be avoided and the time resolution becomes limited predominantly by the photodetector response and the optical photon travel time spread in the radiator. This method and an experiment, demonstrating a coincidence resolving time of 87 ps FWHM, were presented in our previous work [1–4]. Such a fast detection was achieved using lead fluoride (PbF2) crystals with a thickness of 15 mm. At such thickness, the stopping power of PbF2 is comparable to 20 mm of LSO scintillator. In addition, the PbF2 has a higher photofraction due to its high Zeff. The crystals were coupled to microchannel plate photomultiplier (MCP PMT) photodetectors, which were selected due to their very fast time response. The prototype Hamamatsu MCP PMTs used had a single photon time response of 50 ps FWHM [2], but had a relatively low quantum efficiency (QE) with peak value of 20%. Furthermore, due to the microchannel plate collection efficiency of approximately 60%, the photon detection efficiency (PDE) was only about 12%. With superbialkali photocathode with peak QE of 35% and MCP collection n Corresponding author at: Jožef Stefan Institute, Jamova cesta 39, SI-1000 Ljubljana, Slovenia. Tel.: þ 386 1 477 3157; fax: þ386 1 477 3166. E-mail address:
[email protected] (R. Dolenec).
http://dx.doi.org/10.1016/j.nima.2015.09.059 0168-9002/& 2015 Elsevier B.V. All rights reserved.
efficiency of 60% a single side gamma detection efficiency of about 10% would be possible. In efforts to improve the efficiency, the silicon photomultiplier (SiPM) was considered as a photodetector for the Cherenkov TOF PET method. Compared to MCP PMTs, the SiPMs have significantly higher peak photon detection efficiency that can typically reach 40%. However, with SiPMs this peak is shifted to higher wavelengths and the efficiency drops more abruptly at lower wavelengths, where more Cherenkov photons per unit wavelength are produced (Fig. 1). According to simulation prediction, the net effect is still an approximately four fold increase in coincidence detection efficiency; in addition, the improvement in UV sensitivity has already been demonstrated [6] so a further improvement in gamma detection efficiency can be expected. The SiPMs also have other benefits: they are insensitive to strong magnetic fields (suitable for PET/MR scanners) and might soon become more cost effective than other photodetectors currently used in PET. However, compared to MCP PMTs the SiPMs have a worse time response – the single photon time resolution of larger area SiPMs is about 200 ps FWHM [7] – and have high dark count rates on the order of 100 kHz/mm2 at room temperature. The dark count rate represents a challenge for the use of SiPMs in Cherenkov TOF PET: on average, only a few Cherenkov photons reach the photodetector after 511 keV gamma interactions with the radiator and the method has to rely on detection of single photons, which produce pulses indistinguishable from the pulses due to dark counts. To
128
R. Dolenec et al. / Nuclear Instruments and Methods in Physics Research A 804 (2015) 127–131
reduce the dark noise background, the presently available SiPMs have to be cooled. In this paper, the first Cherenkov TOF PET measurements using SiPMs are presented. Hamamatsu S10931-050P 3 3 mm2 SiPMs were used in combination with PbF2 Cherenkov radiators in a back-to-back set-up presented in the next section. The measurements were performed at temperatures between 25 °C and þ25 °C. The effects of the temperature and overvoltage on the investigated detector performance are shown and discussed in Section 3, followed by the conclusion in Section 4.
2. Experiment and methods Two gamma detectors, consisting of a 3 3 mm2 active area SiPM optically coupled to a 5 5 15 mm3 PbF2 crystals, were positioned in a back-to-back configuration (Fig. 2). This crystals were already used in our previous measurements with MCP PMTs as photodetectors. Hamamatsu S10931-050P SiPM samples with 50 μm pixels were used due to their low dark count rates. According to producer specifications, at room temperature they have a dark count rate of 0.77 MHz per device. Their specified single photon time resolution is 100
0.9
PDE,Transmission [%]
0.8
80 70
Cherenkov photons produced
0.7
60
PbF2 Transmission PDE (SiPM) PDE (MCP PMT)
0.6
50
0.5
40
0.4
30
0.3
20
0.2
10
0.1
0 200
300
400
500
600
700
Ch. photons produced [a.u.]
1
90
0 800
wavelength [nm] Fig. 1. The PDE of MCP PMT photodetector used in previous experiments [1–4], compared to the PDE of Hamamatsu S10931-050P SiPM. The former was calculated as quantum efficiency collection efficiency, while the latter is the standard SiPM PDE as reported by the producer, excluding the effects of crosstalk and afterpulses. Also shown are the optical transmission of 25 mm thick PbF2 crystal [5] and the 1=λ2 distribution, indicating the wavelength dependence of produced Cherenkov photons.
about 500 ps FWHM [8], while recent reports indicate a better resolution of about 200 ps FWHM with the same SiPM type [7]. The crystals were polished and either bare or black painted, while their exit surface was centered with SiPM active surface using mechanical supports, machined out of 10 mm thick Teflon. In case of bare PbF2, the reflectivity of the Teflon supports also contributed to an increased number of detected Cherenkov photons. The two detectors were positioned very close to a 1.8 MBq 22Na point source, so that the distance between individual crystal entry surface and the source was approximately 5 mm. Such small distance was used so that even at room temperature the rate of true coincidences was not significantly smaller than the rate of random coincidences due to SiPM dark counts. The source and both detectors were enclosed in lead radiation shielding, which in turn was stationed in a light tight, temperature controlled freezer. The SiPMs were soldered to custom electronic boards with a NEC uPC2710TB preamplifier. Preamplified signals were lead out of the freezer box and into a leading edge discriminator (Phillips Scientific Mod.708), after additional amplification (Ortec FTA820) and a passive signal splitter. Discriminator threshold was set to 0.5 single photoelectron signal height. Logic signals from the discriminator were used for time information (Kaizu works KC3781A TDC) and coincidence logic (Phillips Scientific Mod.752). The other outputs of the signal splitter were used for the charge information (CAEN Mod.V965 QDC), which was needed for the correction of time-walk due to the use of a leading edge discriminator. The measurement was triggered by an AND output of the coincidence logic, with a coincidence resolving time set to 10 ns. To accommodate the time needed to form the coincidence trigger, the signals led to TDC and QDC were delayed by approximately 70 ns. The coincidence time was calculated as the difference between the time-walk corrected TDC measurements from the two detectors. The obtained distributions were fitted with a sum of a constant and two Gaussian functions and the time resolution was expressed as full width of the peak at one half between the constant noise floor and the peak maximum (e.g. FWHM of the double Gaussian component of the fit function). This value will be reported simply as FWHM throughout this paper. To quantify the effects of SiPM dark counts in Cherenkov TOF PET, a signal-to-noise (S/N) value was defined as the ratio between the number of detected Cherenkov coincidences and the number of random coincidences due to dark counts. These were estimated from the integral of the double Gaussian and the constant components of the fit function, respectively. The ratio was calculated for the coincidence time windows (integration intervals) of710 ns, 74 ns and 72 ns. The shortest 4 ns ( ¼ 7 2 ns) wide time window corresponds to a field-of-view of 60 cm. This is more or less the smallest practical time window for full-body PET, especially considering the axially tilted lines of response.
3. Results and discussion 3.1. TOF PET measurements Fig. 3 shows the coincidence time distributions in cases with and without the 22Na source present between the two detectors. Here, bare PbF2 crystals were used, while the SiPMs were biased at a producer recommended overvoltage of V ov ¼ 1.5 V and the temperature was set to þ25 °C. As can be seen, the random coincidences due to SiPM dark counts result in a constant background.1 The measurement Fig. 2. The experimental set-up: two PbF2 Cherenkov radiators coupled to SiPM photodetectors, positioned very close to a 22Na point source. Optical crosstalk between the two detectors was suppressed using a thin, black plastic foil, which is not shown in this figure.
1 The count ripple in both cases, with and without the source, was caused by crosstalk between the channels of the discriminator. This crosstalk has no significant effect on the measured efficiency and timing resolution.
R. Dolenec et al. / Nuclear Instruments and Methods in Physics Research A 804 (2015) 127–131
Counts
4000 3000
Coincidence Timing
Entries 1109742 Const 1108 ± 0.0 G1_Const 3398 ± 0.3 G1_Mean -0.1337 ± 0.0000 G1_Sigma 0.3241 ± 0.0000 G2_Const 439.1 ± 0.3 G2_Mean -0.001323 ± 0.000448 G2_Sigma 1 ± 0.0
With source W/o source FWHM = 824 ps S/N (± 10ns) = 0.2 S/N (± 4ns) = 0.4 S/N (± 2ns) = 0.9
900 Coincidence Timing FWHM [ps]
5000
2000 1000 0 -10
-5
0
5
129
800 700 600 500 400 300 200 100 0
10
-20
-10
0
Coincidence Time [ns]
Fig. 3. The coincidence time distribution with bare PbF2 crystals, V ov ¼ 1:5 V and T ¼ þ25 °C, for cases with and without the 22Na source present. In both cases the events were collected for 30 min.
3.2. Gamma detection efficiency The 511 keV gamma detection efficiency of the Cherenkov detector using a SiPM was estimated in a similar way as for our previous 2 The noise was removed from coincidence window by introducing a longer power supply line for one of the two channels.
20
10
20
Singles Rate
Rate [kHz]
800 700
Source + dark counts
600
Dark counts
500 400 300 200 100 0 -20
-10
0 Temperature [ ° C]
Signal-to-noise Ratio ± 2ns window
50
± 4ns window
40 S/N
±10ns window
30 20 10 0 -20
-10
0
10
20
Temperature [ °C]
Fig. 4. The measured temperature dependence of coincidence timing (top), single side event rate (middle) and signal-to-noise ratios at different time windows (bottom) with bare crystals and V ov ¼ 1.5 V.
25000
935845 105 ± 0.0 5321 ± 1.6
Entries Const G1_Const
FWHM = 717 ps S/N (± 10ns) = 10.1
20000
S/N (± 4ns) = 25.2 S/N (± 2ns) = 50.4
Counts
without the source has a small central peak from time correlated noise via the shared preamplifier power supply line, which was removed for later measurements.2 When the source of annihilation gammas is introduced, a significant central peak due to Cherenkov photons generated in PbF2 crystals with the timing resolution of 824 ps FWHM can be observed. A separate measurement was performed in order to evaluate the contribution of events, where the annihilation gamma is detected in the SiPM itself (the devices used have a 0.3 mm thick epoxy window). Black plastic foil was inserted between the Cherenkov radiator and the SiPM in one detector and the event rate was measured. According to this measurement, only about 2% of single side events are not produced due to a detection of Cherenkov photons originating from a 15 mm thick, bare PbF2 crystal. A set of measurements was performed while lowering the temperature from þ25 °C to 25 °C in 10 °C steps. Due to the large mass of the lead shielding, the system was left overnight to stabilize for each temperature step before measuring. The SiPM breakdown voltage at different temperatures was determined from the data provided by the producer, the value at room temperature (¼70.33 V) and the temperature coefficient (¼ 56 mV/°C) [8]. The sensor was operated at a producer recommended overvoltage of 1.5 V; the resulting signal gain was measured to be the same at all temperatures. The temperature dependences of the coincidence timing, single side event rate and S/N ratios are shown in Fig. 4. Despite the fact that SiPM overvoltage was kept constant, the coincidence timing resolution improved from 824 ps FWHM at þ25 °C to 717 ps FWHM at 25 °C (Fig. 5). It follows from the measured single side event rates that the SiPM dark count rate decreases by a factor of approximately 2.4 for every 10 °C of cooling. As can be seen from the S/N ratios, the contribution from SiPM dark counts decreases significantly with temperature – the S/N ratios increase almost by a factor of 60 after cooling from þ 25 °C to 25 °C. At T ¼ 25 °C measurements were repeated with black painted crystals and with different SiPM overvoltages. It can be seen in Fig. 6 that the black paint improves the time resolution to 581 ps FWHM at 1.5 V overvoltage (Fig. 7, top). The timing improves further with increasing overvoltage, to 422 ps FWHM at V ov ¼ 2:5 V (Fig. 7, bottom). However, dark count rates increase with overvoltage and consequently the S/N ratio becomes less favorable.
10
Temperature [ ° C]
15000
G1_Mean G1_Sigma
-0.07354 ± 0.00003 0.6187 ± 0.0001
G2_Const G2_Mean G2_Sigma
1.936e+04 ± 1.531e+00 -0.07721 ± 0.00001 0.2668 ± 0.0000
10000 5000 0 -10
-5
0
5
10
Coincidence Time [ns]
Fig. 5. The coincidence time distribution with bare crystals, V ov ¼ 1:5 V and T ¼ 25 °C.
detector [3]. The Cherenkov detector on one side of the back-to-back setup was replaced with a reference scintillation detector, consisting of a different sample of Hamamatsu S10931-050P SiPM and a 3 3 30 mm3 Teflon wrapped LYSO crystal. The reference detector was positioned further away from the source, ensuring a tight collimation of coincident gammas incident on the Cherenkov detector.
130
R. Dolenec et al. / Nuclear Instruments and Methods in Physics Research A 804 (2015) 127–131
S/N (± 10ns) = 6.4
800
S/N (± 4ns) = 16.1
4000
700
S/N (± 2ns) = 32.2
600 500 400
3000 2000 1000
200 100 1
1.2
1.4
1.6
1.8
2
2.2
0 -10
2.4
-5
0
Singles Rate 60 Dark counts
50
4500
FWHM = 422 ps
4000
S/N (± 10ns) = 1.3
3500
S/N (± 4ns) = 3.3
Counts
30
Entries Const G1_Const G1_Mean G1_Sigma G2_Const G2_Mean G2_Sigma
S/N (± 2ns) = 6.6
3000
40
5
10
Coincidence Time [ns]
Overvoltage [V]
Rate [kHz]
154163 25.63 ± 0.04 386.6 ± 0.9 0.02792 ± 0.00041 0.6776 ± 0.0009 4506 ± 0.9 0.04456 ± 0.00002 0.2346 ± 0.0000
300
0
2500
155410 80.53 ± 0.04 3857 ± 0.7 0.04715 ± 0.00002 0.1696 ± 0.0000 354.3 ± 0.7 0.06173 ± 0.00038 0.5635 ± 0.0008
2000 1500
20
1000 500
10
0 -10
0 1
1.2
1.4
1.6
1.8
2
2.2
-5
2.4
Overvoltage [V]
0
5
10
Coincidence Time [ns]
Fig. 7. The coincidence time distribution with black painted crystals and T ¼ 25 °C, for cases with SiPM overvoltage of V ov ¼ 1:5 V (top) and V ov ¼ 2:5 V (bottom).
Signal-to-noise Ratio ± 2ns window
50
Entries
2200
± 4ns window
40
Counts
30 20 10
1191255
Measurement
2000
± 10ns window
S/N
Entries Const G1_Const G1_Mean G1_Sigma G2_Const G2_Mean G2_Sigma
FWHM = 581 ps
5000
Counts
Coincidence Timing FWHM [ps]
Coincidence Timing 900
1800
± 1 σ Photopeak Cut
1600
1275 keV Compton Estimate
1400 1200 1000 800 600
0 1
1.2
1.4
1.6
1.8
2
2.2
2.4
Overvoltage [V]
Fig. 6. The dependence of the coincidence timing (top), single side event rate (middle) and the signal-to-noise ratios at different time windows (bottom) on overvoltage. Here, black painted crystals were used and the temperature was set to 25 °C.
Due to much larger signals resulting from scintillation pulses, the reference SiPM was biased at a very low overvoltage and the amplification chain for the reference detector was modified. The measurement was triggered by the reference detector. Only one change was made to the Cherenkov detector – instead of machined Teflon, 3Dprinted black plastic was used for mechanical support. This enabled a more precise alignment of the two detectors and the source, but resulted in a slight decrease in the number of detected Cherenkov photons in the case of bare crystals. Fig. 8 shows the energy spectrum obtained with the reference detector. Clearly visible are the photopeaks and Compton distributions due to 511 keV and 1275 keV gammas, originating from the 22Na source. Also of note are events due to LYSO self-activity and the nonlinearity, caused by the SiPM saturation. The region of the spectrum from 511 keV photopeak to approximately the upper edge of 1275 keV Compton distribution was fitted with a sum of Gaussian, linear and exponential functions. The Gaussian function described the photopeak, while the sum of a linear and exponential functions closely followed the 1275 keV Compton distribution distorted by a finite energy resolution and SiPM saturation.
400 200 0 0
500
1000
1500
2000
2500
Charge [200 fC/bin]
Fig. 8. The energy spectrum measured with a reference SiPM-LYSO detector.
The Gaussian component of the fit was used to select events inside a 71σ wide part of the 511 keV photopeak. The fraction of such events that had a hit also on the Cherenkov detector was the basis for the estimation of detection efficiency. The final values shown in Fig. 9 were also corrected for the overestimation due to SiPM dark counts and underestimation due to the 1275 keV Compton events. A single side detection efficiency of 14% was measured with a bare crystal and 2.5 V overvoltage. Note, however, that these measurements were performed with 5 5 mm2 cross-section Cherenkov radiators coupled to 3 3 mm2 active area SiPMs and as a result, a large fraction of Cherenkov photons available for detection was lost. Therefore an improvement in detection efficiency is expected with a 1:1 coupling between the radiator and the SiPM. From a simulation of the device we estimate an improvement by about a factor of 2.1 for black painted crystals, while for bare crystals this factor would be 1.8 (Fig. 9). 3.3. Discussion According to results for efficiency and timing presented in this work, a Cherenkov based TOF PET scanner using SiPMs as
R. Dolenec et al. / Nuclear Instruments and Methods in Physics Research A 804 (2015) 127–131
Bare (1:1 coupling) Bare Black painted (1:1 coupling) Black painted
30
Efficiency [%]
25 20 15 10 5 0 1
1.2
1.4
1.6
1.8
2
2.2
2.4
Overvoltage [V]
Fig. 9. The measured single side efficiency with bare and black painted crystals (solid lines). Also shown are the estimated efficiencies for the case of 1:1 coupling between the Cherenkov radiator and SiPM (dashed lines).
photodetectors could achieve performance competitive to a scintillation based scanner. For the same scanner length, the Cherenkov approach could also be less expensive mainly due to the lower cost of PbF2 radiator compared to typically used scintillators. This statements, however, do not take into account the issue of SiPM dark counts. We estimate that in order for the random rate due to SiPM dark counts in a full body scanner to reach the same order of magnitude as random rates in traditional scanners (100 kcps), the dark count rate should be reduced to about 1 kHz per channel. Since this is about 2 orders of magnitude below the noise level of presently available SiPMs, a certain amount of cooling or improvement in SiPM technology would be needed.
4. Conclusion TOF PET measurements were performed for the first time using exclusively the Cherenkov photons detected with SiPMs. A higher
131
peak PDE of SiPMs compared to previously used MCP PMTs resulted in a higher single side detection efficiency of up to 14%, despite non-optimal coupling between the Cherenkov radiator and SiPM. With 1:1 coupling, single side efficiency of over 25% seems possible. The SiPMs used for measurements presented here were not the best in terms of timing and limited the coincidence time resolution to 422 ps FWHM in the best case presented. SiPMs with better timing properties are available and could improve the results. Measurements with different SiPM samples and with optimized coupling with PbF2 crystals are already under way and will be presented in our future publication. The SiPM dark counts introduce a constant background in the coincidence data. With SiPMs used for the presented measurements, this background is overwhelming at room temperature, but can be decreased with cooling.
References [1] R. Dolenec, et al., in: 2010 IEEE Nuclear Science Symposium Conference Record (NSS/MIC), 2010, p. 280. [2] S. Korpar, et al., Nuclear Instruments and Methods in Physics Research Section A: Accelerators, Spectrometers, Detectors and Associated Equipment 654 (2011) 532. [3] S. Korpar, et al., Physics Procedia 37 (2012) 1531. [4] S. Korpar, et al., Nuclear Instruments and Methods in Physics Research Section A: Accelerators, Spectrometers, Detectors and Associated Equipment 732 (2013) 595. [5] P. Achenbach, IEEE Transactions on Nuclear Science NS-48 (2001) 144. [6] K. Sato, et al., Nuclear Instruments and Methods in Physics Research Section A: Accelerators, Spectrometers, Detectors and Associated Equipment 732 (2013) 427. [7] S. Gundacker, et al., Nuclear Instruments and Methods in Physics Research Section A: Accelerators, Spectrometers, Detectors and Associated Equipment 718 (2013) 569. [8] Hamamatsu Photonics Catalogue No. KAPD1023E05 Nov. 2009 DN.