Index terms: PET, Silicon Photo-multiplier, Small Animal PET insert for MRI scanner. Size: 1.2x1.2 mm, Pitch: mm Thickness: 4 mm

Transcription

1 Measurement of Energy and Timing Resolution of Very Highly Pixellated LYSO Crystal Blocks with Multiplexed SiPM Readout for Use in a Small Animal PET/MR Insert Christopher J. Thompson, Senior Member IEEE, Andrew L Goertzen, Member IEEE, Poitr Kozlowski, Fabrice Retière, Greg Stortz, Vesna Sossi, and Xuezhu Zhang 1 Abstract - Arrays of silicon photo-multipliers (SiPMs) are good candidates for the readout of detectors in PET MR inserts due to their high packing density, efficiency, low bias voltage and insensitivity to magnetic fields. In this study we report the readout performance of SensL SiPM arrays in terms of their ability to resolve all elements of pixellated lutetium oxyorthosilicate (LYSO) crystals, and their energy and timing resolution. A SensL SB CER SiPM array consisting of sixteen 3 x 3 mm elements were used as light sensor. An LYSO crystal block consisting of 10x mm x 1.2 mm x 6.0 mm crystals on the bottom layer and 9x9 1.2 mm x 1.2 mm x 4.0 mm crystals on the top layer (which is offset by Ω the crystal width) was mounted on the SensL array, and covers over 95% of its area. A DPC encoding multiplexor with HDMI cable readout made by the Triumf Instrumentation group with a footprint suitable for allowing 16 modules to form a ring inside a Brooker 7T MR rodent imager was used as a readout. The HDMI cable supplies power, and provides readout of four channels and the device temperature. The average energy resolution for all 100 crystals in the lower layer was 11.3±1.8% and 10.8±1.2% for the 81 crystals in the upper layer. The average timing resolution for all 100 crystals in the lower layer was 2.52±0.23 nsec. and 2.55±0.23 nsec. for the 81 crystals in the upper layer. Index terms: PET, Silicon Photo-multiplier, Small Animal PET insert for MRI scanner Manuscript date November 19, This work was 1 supported by a grant from the Natural Science and Engineering Council of Canada (OGP ) to Dr. Thompson and a Manitoba Health Research Council operating grant to Dr Goertzen. C J Thompson is an emeritus professor of Montreal Neurological Institute, McGill University. Present address: rue Lavigne, Montreal, QC, Canada H4J 1X8. Phone +1 (514) Christopher.Thompson@McGill.Ca A L Goertzen :Department of Radiology, University of Manitoba, Winnipeg, MB, Canada F Retière: Detector Development Group, TRIUMF, Vancouver, BC, Canada P Kozlowski: Department of Radiology, University of British Columbia, Vancouver, BC, Canada G Stortz and V Sossi: Department of Physics and Astronomy, University of British Columbia, Vancouver, BC, Canada X Zhang: National Research Council of Canada, Winnipeg, MB, Canada I. INTRODUCTION Our group is presently building a small animal PET scanner which can be installed within a 7 Tesla MRI scanner within the 106 mm bore gradient coil using dense arrays of lutetium-yttrium oxy-orthosilicate (LYSO) crystals coupled to silicon photo-multipliers[1] as photo-sensors which are insensitive to magnetic fields. This particular study addresses two critical issues related to the overall project: 1) The ability to resolve a large number of small crystals mounted on SensL[2] B-series silicon photo-multipliers (SiPMs); 2) the ability of these detectors to provide good energy and timing resolution. Space for the PET detectors is very limited, and this severely limits the choice of detectors and supporting equipment. The configuration presently envisaged consists of rings of 16 detectors with a face-to-face distance of 64 mm. If each detector is in coincidence with seven others the fully sampled field of view diameter is 32 mm. The prototype scanner will have two rings. Other aspects of the project are described more fully in the following presentations made at the 2013 IEEE MIC: [3,4] II. MATERIALS AND METHODS A. Description of Crystal Arrays The crystal arrays purchased from Proteus Inc. according to the following specifications: LOWER LAYER: Material: LYSO: Ce Matrix: 10x10 elements 2 Size: 1.2x1.2 mm, Pitch: mm Thickness: 6 mm Overall footprint: 13.1 x 13.1 mm 2 UPPER LAYER (offset by Ω crystal pitch in both directions): Material: LYSO: Ce Matrix: 9x9 elements 2 Size: 1.2x1.2 mm, Pitch: mm Thickness: 4 mm Surfaces: all polished Gap/Reflector: mm/bonded ESR Two layers of ESR reflector are used for the boundary between the outer rows and columns in both layers Crystal density: 99.8 crystals/cm 2 The SensL ArraySB CER SiPM array used in this study has the following specifications: Number of pixels: 16 arranged in a 4 x 4 array Active area: mm 2

2 Pixel size: mm 2 Cells per pixel: 4774 Breakdown Voltage: 24.5 V Gain: Fill Factor of crystal block on SiPM 95.6% Fig 1. Top view of crystal block. Fig 2: Bottom view of block. Two views of the crystal block are shown in Figs. 1 and 2. The SensL array is actually covered with a 0.5 mm protective layer which provides some sharing of the light. Crystals in an outer row or column of the 10x10 crystal array shine on only two of the 16 SiPMs, enough for crystal identification on one axis only. Several different strategies were tried to optimize the light spread from all the small crystals in order to 1) uniquely identify all crystals, and 2) make the flood images as uniform as possible in order to facilitate automated segmentation for the crystal identification matrix (CIM). In a previous paper [5] we concentrated on being able to resolve a centred two layer array of 8x8 and 7x7 1.2 mm crystals. In order to extend the identification to the edge and corner crystals of the 10x10 array we first changed the light barrier between the outer and penultimate rows and columns adding an extra layer of ESR. This has the effect of steering more light from these crystals towards the outer SiPM elements. We then tested various sizes of light guides to optimize the uniformity of the flood images. B. Signal Processing and Data Acquisition The signals from the 16 elements of each array are multiplexed down to four channels using a discretized positioning circuit (DPC)[6] encoding multiplexor with HDMI cable readout made by the Triumf Instrumentation group [7]. It has a footprint suitable for allowing 16 modules to form a ring inside a Brooker 7T MR rodent imager. The HDMI cable terminates on a controller board [3] equipped with a Raspberry Pi processor which monitors the SiPM temperature and can control the SiPM bias. This board has a set of five SMA connectors for each detector, four for the DPC position encoding signals and one for the sum of all which is sent to a Canberra Model 1428 constant fraction discriminator (CFD) to produce a stop signal to a Canberra 2145 time to amplitude converter (TAC) for each event. The crystal block mounted on the SiPM array attached to the MPX device is mounted in a light-tight diecast box with the HDMI cable extending from one end. In order to facilitate the testing of different crystal block and light guide combinations, a two part optical cement without the curing part added allowed the parts to be changed easily. The crystal block was held down on the SiPM face with a piece of 1 mm plywood which is tensioned by tightening two screws to lock everything in place for each test. The box was placed 5 cm from a Scanwell Systems PET timing alignment probe [8] which contained a Na 12.4 kbq source at the time of these experiments. This served as a source 511 kev photons. When the Na source emits a positron a signal is sent to a CFD whose output provides a start signal for input to the TAC. In order to acquire the flood CIM images and identify the region associated with each crystal in a detector, data are acquired with the four DPC signals (which have been filtered with a spectroscopy amplifier with a 200 nsec time constant) input to a six channel ADC. Another ADC channel acquires the TAC signal. The ADC is triggered twice for each event, first at the peak of the DPC signals, and 1 microsecond later to encode the TAC signal. Data were collected in list-mode by a program which provides three outputs, 1) a list of all the ADC readings for each event, 2) a series of flood images, binned according to the time after the positron was detected, and 3) a global spectrum derived from the sum of the filtered DPC signals. Energy discriminators can be set up so as to accept only events which are within the global 511 kev photo-peak, and a timing window can be set up such that only those events which occur at the selected time after the positron was detected by the timing probe are assigned to an image. The list-mode file is then processed by several other programs which enable the production of energy and timing spectra for all the crystal elements in the block. The file containing the flood images has 64 individual images, the first of which is the sum of all data collected. This image is used to segment the data into square regions corresponding to each crystal s response. The segmentation is done manually, first by pointing to the region, then to the crystal to which it corresponds. Since there are two layers, the protocol requires the lower layer crystals be assigned to odd: odd locations in the CIM, and those in the upper layer to even:even locations. Next the list-mode file is played back into a program which uses the data from each of the regions to create a spectrum for each crystal. Next the photo-peaks are identified and fitted to a triple Gaussian from which the photo-peak location and the FWHM of the energy spectrum are calculated. Then the listmode file is played back into another program which uses the data from each of the regions and the relative photo-peak position to provide an energy-windowed timing spectrum in which the events a re-binned according to their perceived detection delay since the positron emitted by the Na nucleus was detected by the timing probe. Finally the timing spectra from each crystal are fitted with a triple Gaussian and the FWHM of the timing spectrum and the delay after the positron detection are calculated. C. Experimental Studies Data were acquired with the crystal block directly coupled to the SiPM array and with an intervening layer of 1.0 mm thick glass which was cut from a microscope slide. The glass pieces were made of squares 13.1, 13.7, 14.4 and 24.4 mm on the side. In each case, the edges were roughenedwith #200 grade sanding cloth.

3 B. Crystal identification images Each of the crystals in both layers must be identified in order to measure its energy resolution for 511 kev annihilation photons and form its timing spectrum. This is done manually, first identifying the region associated with each crystal, then assigning this region to a specific crystal as illustrated in fig. 5. Fig. 3 Profiles through two rows of crystals illustrating the effects of different coupling methods. Data from all combinations of glass coupling thickness were displayed and analysed according to the method described above. Only the set which gave the best crystal identification method (the 13.1 glass square) is discussed further and processed to provide the energy and timing spectral analysis. III. RESULTS A. Profiles through crystal identification matrix images Dot-join profiles through the block with two layers of 1.2 mm crystals are shown in Fig. 3. The first image is for direct coupling with no light guide. In this case, the crystal response dots are the smallest and the crystal responses are spread over the biggest area. However, the outer crystals are not resolved. W hen the 1 mm glass light guides are inserted between the crystal block and the sensor array, the distribution of crystal response dots gets more uniform, but the dots become bigger, and the region of the identification matrix they occupy becomes smaller. The resolvability index (ratio of average crystal response FW HM to the crystal response s average separation in the flood image) for the best and worst are 0.51 and The average peak to valley ratios are 11.2 and 7.2 for the same rows. Fig. 5 Each crystal s response in the CIM (left side) is assigned to an individual crystal in the matrix (right side) Fig. 6 Two dimensional histograms of the photo-peak location and energy resolution for each crystal in each layer. C. Energy spectra for individual crystals The photo-peak position in its 256 channel spectrum and the measured energy resolution at 511 kev are shown for each crystal in both layers in Fig. 6. The average energy resolution is 11.3±1.8% for the 100 crystals in the lower layer, and 10.8 ±1.2% for the 81 crystals in the upper layer of the block. Fig. 4 Profiles through crystal responses in the bottom layer in green and top layer in blue for the case with a 13.1 mm square light guide. The best result, shown in Fig. 4 is with a 13.1 mm light guide with its surface roughened, and surrounded with black tape. This combination was then used for further study of the energy and timing resolution. Fig. 7 Two dimensional histograms of the timing resolution and location of the peak in the timing spectrum for each crystal in both layers.

4 Fig. 8 Arrival time image. Each image in the matrix represents a time bin of 210 psec. The image in the top left corner is the sum of all events, and the next and last images are the events which fall outside the time window. D. Timing spectra for individual crystals The timing peak position in its 256 channel spectrum and the measured timing resolution at 511 kev are shown for each crystal in both layers in Fig. 7. The average timing resolution was 2.52±0.23 nsec for the 100 crystals in the lower layer, and 2.55±0.23 nsec for the 81 crystals in the upper layer. There is some variation in the perceived arrival time due to the resistive nature of the multiplexer and the intrinsic capacitance of the individual SiPM elements and associated wiring. As seen in Fig. 8, the crystals in the corners show the annihilation photons arriving first, then those in the top and bottom crystals, of while those in the centre are later. The spread of arrival times is characterized by a standard deviation of 1.12 nsec for crystals in the lower layer and 0.95 nsec for crystals in the upper layer. This dispersion of the arrival times has the effect of blurring the global timing spectrum of the block. VI. DISCUSSION We have successfully mounted a 181 crystal dual layer block on a SensL 4 x 4 B-series array and were able to identify all crystals with the 16 sensor outputs multiplexed to only four signals for encoding. Profiles through the crystal response dots went down to almost background between almost all the crystals suggesting that it may be possible to use even smaller crystals on these sensors, since the crystal size is the dominant factor contributing to the resolution blurring. However, the corner crystals are not fully resolved. Our method of using a second layer of crystals which are offset by Ω of the crystal spacing and simple geometric decoding based on the crystal s response blob requires that the response falls into a pattern which can be interpreted easily to assign the responses to their appropriate crystal. For this reason it is important that they first be resolved, and

5 second the location of their response can allow the response to be assigned to the correct crystal. This was done manually in the present study, as shown in Fig. 5, but an automated procedure will be required in the final scanner. We have been very encouraged by the ability of the SensL detectors to resolve the 181 crystals in our arrays. In our previous work [9], we successfully identified only 113 crystals in a dual layer block. The individual crystals were the same size (1.2 mm) as in this study, but only occupied 57% of the 4x4 SiPM array. In this study we have added two extra rows and columns which has allowed us to increase the fill factor in the planned scanner from 57% to 95%. In addition the energy resolution has been improved from 16.5% previously obtained to an average of 11.1% with many more crystals. We believe this improvement to be due to the use of the recently introduced B-series SiPM arrays which have significantly lower noise than the M-series used in the previous study. Prior to choosing this block design, several others were evaluated. Other designs made the identification of the outer rows and columns of crystals difficult, and many crystals were unresolved, while others caused significant reductions in the light received from the outer rows and columns of crystals. We feel it is important to keep the apparent arrival time and the photo-peak location (light collection efficiency) as closely matched as possible so that when these detectors are used in a real scanner the global timing and energy windows can be kept narrow. This will minimize the number of events which might be valid, but end up being rejected once the crystal in each detector has been identified and appropriate corrections for apparent arrival time and photo-peak position are applied. No attempt has been made in this work to use only non-magnetic materials in order to test the system in a high field magnet. Instead we concentrated on demonstrating the detector s utility for high resolution PET imaging. V. CONCLUSION W e have shown that the current version of the SensL SPM4 array is capable of resolving 181 crystals in a dual layer configuration. The energy and timing resolution of individual crystals of this detector are very suitable for use in a PET scanner for imaging the brains of rats or mice. With the use of entirely non-magnetic materials in the final detector, we believe this would provide quality PET images simultaneously with any MRI pulse sequence, and thus be suitable for use in the PET insert for the Brooker 7T MRI which our group is presently developing. Our previous studies [9], with a smaller number of crystals in each block, and two detectors in coincidence suggest that a small animal PET scanner with these detectors should achieve about 1.0 mm FWHM near the center for the field of view. Thus it would appear that this type of detector would perform well in a PET ring small enough to fit in the bore of a MRI. VI. ACKNOWLEDGEMENTS The glass light guides were made by Pierre Verrette. This work was supported from a personal grant from the Canadian Natural Science and Engineering Research Council of Canada (NSERC) # to CJT. and from a Manitoba Health Research Council Operating Grant to ALG. VI. REFERENCES [1] B. Dolgoshein,,V. Balagura, P. Buzhan, et al. (Calice SiPM collaboration) Status Report on Silicon Photomultiplier Development and Applications Nucl. Instr. & Meth-A 563: (2006) [2] Sensl Inc. Cork, Ireland: See device specifications at: [3] E. Shams, J. D. Thiessen, D. Bishop, P. Kozlowski, F. Retiere, V. Sossi, G. Stortz, C. J. Thompson, A. L. Goertzen A PET Detector Interface Board and Slow Control System Based on the Raspberry Pi IEEE MIC conference record 2013: M [4] J. D. Thiessen, C. Jackson, K. O'Neill, D. Bishop, P. Kozlowski, F. Retire, V. Sossi, C. J. Thompson, A. L. Goertzen Performance Evaluation of SensL SiPM Arrays for High-Resolution PET IEEE MIC conference record 2013: M [5] C. J. Thompson, A. L. Goertzen, F. Retière, P. Kozlowski, L. Ryner, G. Stortz: Evaluation of Very Highly Pixilated Crystal Blocks with SiPM Readout as Candidates for PET/MR Detectors in a Small Animal PET Insert IEEE MIC Conference Record 2012: M [6] P D Olcott, J A Talcott, C S Levin, F Habte and A M K Foudray Compact Readout Electronics for Position Sensitive Photomultiplier Tubes IEEE Trans Nucl. Sci. 52: (2005) [7] A. L. Goertzen, J. D. Thiessen, X. Zhang, C-Y Liu, E Berg, D Bishop, P. Kozlowski, F. Retière, V. Sossi G. Stortz, G. Stortz, C. J. Thompson: Application of HDMI Cables as an MRI Compatible Single Cable Solution for Readout and Power Supply of SiPM Based PET Detectors IEEE MIC Conference Record, 2012: M [8] Thompson C J, Camborde M-L, Casey M E: "A Central Positron Source to Perform the Timing Alignment of Detectors in a PET Scanner" IEEE Trans Nucl Sci. 52: (2005) [9] C J Thompson, E J Berg, A L Goertzen, P Kozlowski, F Retière, L Ryner, G Stortz and V Sossi: Evaluation of High Density Pixellated Crystal Blocks with SiPM Readout as Candidates for PET/MR Detectors in a Small Animal PET IEEE Trans Nucl. Sci. 59: (2012)