A PET detector module using FPGA-only MVT digitizers

A PET detector module using FPGA-only MVT digitizers Daoming Xi, Student Member, IEEE, Chen Zeng, Wei Liu, Student Member, IEEE, Xiang Liu, Lu Wan, Student Member, IEEE, Heejong Kim, Member, IEEE, Luyao Wang, Chien-Min Kao, Senior Member, IEEE, and Qingguo Xie, Member, IEEE Abstract Multi-voltage threshold (MVT) is an amplitudebased sampling method. It takes timing samples when the event pulse crosses the user-defined thresholds. Only a few comparators and TDCs are required when implementing such digitizer. Previously, we have demonstrated an FPGA-only MVT digitizer based on this method. The FPGA-only MVT digitizer employs the differential I/Os in an FPGA as the required comparators and FPGA based TDCs. The implementation of this digitizer is entirely based on the FPGA. We have demonstrated that it is possible to implement a significant number of MVT digitizers by using a single FPGA. It is also flexible, as it allows us to readily modify, or add functions to, the implementation without requiring costly hardware changes. Currently, we are developing a PET detector module using the FPGA-only MVT digitizer. In this paper we describe the design and implementation of the detector module and report its performance properties. The detector module has a total detection sensitive area of 50mm 50mm, an overall energy resolution of 15.1% FWHM at 511keV, and a module-level coincidence timing resolution of 684ps FWHM. In addition, our preliminary imaging with such detector module successfully resolves 1.6mm-diameter rods separated by 3.2mm. Index Terms FPGA, LVDS, MVT, Digitizer, all digital data acquisition (DAQ) system, block detector, PET. I. INTRODUCTION RECENTLY, there is a strong interest in sampling the PET event waveform at the earliest possible stage of the signal detection chain and several methods to achieve this have emerged [1] [4]. By sampling signals early, information loss due to, for example, filtering/shapping and transmission This work was supported in part by the Natural Science Foundation of China (NSFC) Grant #U1201256, #61027006, #61210003, in part by the National Key Technology R&D Program of China Grant #2012BAI13B06, in part by the Key Grant Project of Chinese Ministry of Education #313023, in part by the Ministry of Science and Technology of China (MOST) grant #2012DFG31970, the Research and Development Programme of Hubei Province grant #2011BFA005, the Wuhan Programs for Science and Technology Development grant #201231234461. It was also supported in part by USA Clinical and Translational Science Awards (CTSA) grant CTSA UL1 TR000430 and National Institutes of Health (NIH) grant R01 EB016104-01A1. Qingguo Xie is with the Wuhan National Laboratory for Optoelectronics, Wuhan, Hubei, China and also with the Department of Biomedical Engineering, Huazhong University of Science and Technology, Wuhan, Hubei, China (Tel: +86-27-87793080, email: qgxie@mail.hust.edu.cn). Daoming Xi and Luyao Wang are with the Department of Biomedical Engineering, Huazhong University of Science and Technology, Wuhan, Hubei, China and also with the Raycan Technology Co., Ltd (Suzhou), Suzhou, Jiang Su, China. Wei Liu, Chen Zeng, Xiang Liu and Lu Wan are with the Department of Biomedical Engineering, Huazhong University of Science and Technology, Wuhan, Hubei, China. Chien-Min Kao and Heejong Kim are with the Department of Radiology, the University of Chicago, Chicago, Illinois, USA. can be minimized. By converting the event signals into digital numbers, it also becomes possible to tap the power of modern digital electronics to apply nontrivial signal processing algorithms to achieve many benefits, including enhanced detection performance and easier to perform upgrade and adjust for varying detector properties due to, for example, aging and radiation exposure. Also, the same data acquisition (DAQ) electronics may be used for a variety of detector designs by changing the signal processing algorithms. By using such kind of universal DAQ, one can shorten the PET system development cycle and reduce the development cost. We have previously proposed one such sampling method, which we refer to as the multi-voltage threshold (MVT) method. This method provides amplitude-based sampling in the sense that the signal waveform is sampled with respect to a user-defined set of reference amplitudes [5] [7]. In contrast, the conventional sampling method is time-based in the sense that the signal waveform is sampled at a (approximately) regular interval in time. Modern PET detectors use fast scintillators; therefore, a critical challenge in using the time-based sampling method is the need to have a high sampling rate (above 1 Gsps). Several research groups have demonstrated the Domino Ring Sampler (DRS) can be used to provide such a high sampling rate but the count-rate performance is a concern that remains to be addressed [1], [8]. On the other hand, we have shown that it is sufficient to use only a small number of reference amplitudes with the MVT method [5]. We have also recently demonstrated that it is possible to implement a significant number of MVT digitizers by using a single FPGA [9]. We believe that this sampling technology can provide low-cost and versatile DAQ for developing PET detectors. In this work, we developed a PET detector module using the FPGA-only MVT digitizers and LYSO/SiPM detector block. In Sec. 2, we show the design of the detector module. In Sec. 3 we measured the performance of the detector module and also obtain a imaging by using of a stable dual-head detection system which is consisted of such 2 detector modules. Conclusions and discussion is followed in Sec. 4. II. DETECTOR DESIGN Previously, we have implemented a FPGA-only MVT digitizer by using of the differential I/Os in a FPGA which is configured to work in the low-voltage-differential-signaling (LVDS) receiver mode (refer as LVDS comparators in the

Fig. 1. A 4-level MVT digitizing channel in the FPGA-only implementation. Fig. 2. The digital PET detector module. following paragraph). Such LVDS receiver in the FPGA has a positive input and a negative input and its output is determined by the relative voltages appearing on these inputs. It sends a logic 1 when the voltage at its positive input is higher than that at the negative input; otherwise, it sends out a logic 0.. The schematics of the digitizer is shown in Fig. 1. It has four LVDS comparators for providing four reference voltages. As shown, the PET event pulse is split into four signals and each of which is fed to the positive node of an LVDS receiver. The negative node of the LVDS receiver is connected to a DAC to provide a programmable voltage threshold. The logic output of the LVDS comparator is then connected to two TDCs for determining the digital times of its positive and negative transitions. Thus, the four-level channel generates eight samples for a pulse that has a sufficient amplitude. The TDCs are also implemented inside the FPGA by using the Song s method [10]. In [9], we have reported a single channel implementation on the Altera Cyclone II EP2C70F896C7 FPGA and described methods for calibrating the implementation. Using this FPGA-only 4-channel MVT digitizer for an 2.0 2.0 10mm 3 LSO coupled to a SensL SiPM, we obtained an energy resolution of 16%@511keV and a coincidence timing resolution of 500 ps. By collaborating with the Raycan Technology Co., Ltd (Suzhou), we have recently developed a FPGA-only MVT digitizer board that has 48 4-level MVT digitizer channels. In this work, we setup a detector by connecting 3 such digitizer boards, which provide a total of 144 digitizing channels, to a 2x2 array of LYSO/SiPM detector blocks, each of which has 36 output signals (hence a total of 144 output signals). As shown in Fig.2, the detector module consists of two physical units, including the detector unit (DU) and the digital acquiring unit (DAU). The scintillator pulses generated from the DU are digitized by the DAU. The resulting samples are packaged into event frames and fed to a computer via an Ethernet interface. The event information is then extracted by digitally analyzing the samples on the computer. The energy calibration, crystal identification and coincidence detection are also performed on the computer. An important feature of our detector design is that the DAU can be readily used for other DUs that may use different scintillation materials, crystal sizes, or photodetectors, through Fig. 3. The LYSO/SiPM detector block. Left: A 6 6 SiPM array. Middle: A 6 6 LYSO matrix. Right: A LYSO/SiPM detector block the use of adequate adaptors. This is possible because, as discussed above, different processing algorithms can be readily developed and implemented to handle the different characteristics of the signal pulse once it is converted into digital form by the DAU. A. The Detector Unit The DU consists of a 2 2 array of LYSO/SiPM detector blocks. As shown in Fig. 3, each LYSO/SiPM detector consists of a 6 6 SiPM array and a 6 6 LYSO matrix using one-to-one coupling. The SiPM is SensL FM30035. In the array, each SiPM pixel has an 3.0 3.0 mm 2 active area with a 4.2 4.2 mm 2 pitch. The LYSO matrix is consisted of 3.95 3.95 20 mm 3 crystals with a 0.3 mm gap. Currently, the LYSO matrix is directly coupled to the SiPM array using epoxy optical adhesive (using GHJ-01(Z) from Chenguang research institute of chemical industry). Since the crystal size is larger than the SiPM s active area, we are considering using taped crystal bars to provide a better optical coupling between the LYSO and SiPM in future. B. The digital acquiring unit The DAU is consisted of 3 digitizer boards. We show one of the digitizer boards in Fig. 4. It contains 4 Altera Cyclone II EP2C50U484C6 FPGAs and each of them provides 12 4-level MVT digitizing channels described above, yielding a total of 48 digitizing channels. The waveform samples generated by the EP2C50U484C6 FPGAs are readout by another Altera Cyclone II EP2C50F484C7 FPGA and output via a Gigabit

Fig. 4. The 48 channel 4-level MVT digitizer board. Fig. 6. Count distribution obtained by placing a point-like source in front of the detector module. Fig. 5. The experiment setup for measuring the performance of the detector modules and obtaining the initial imaging result. Each module has a detection sensitive area of 50mm 50mm Ethernet communicator. This FPGA also controls the DACs (using DAC7678 from TI) to provide programmable reference voltages. C. The samples analyzing The event samples acquired by the detector module are first stored on a computer. The resulted samples are fitted, as described in [9], by a bi-exponential function for estimating the event energy and time. The bi-exponential function is given by ( y (t) =a exp t t ) [ ( 0 1 exp t t )] 0, (1) b d where a is determined by the pulse amplitude, b and d by the rise and decay time of the pulse, and t 0 by the occurrence time of the pulse. The event energy and timing is obtained with the re-generated pulse from the fitted function. III. PERFORMANCE MEASUREMENTS AND INITIAL IMAGING As shown in Fig. 5, we set up a dual head detection system by using a pair of detector modules. The distance between the two heads is 26mm. The resulted samples from each digitizer board is collected and analyzed. A clock distributor is used to synchronize all the six digitizer boards. The results reported below are all measured by using this detection system. A. Crystal Identification Since the LYSO crystals are coupled to the SiPMs one by one and the SiPMs output are individually sampled, we can directly get the events position by identifying the digitizer channel number. Fig. 6 shows the counts distribution in each crystal. As a point-like source is placed at the front of the detector, the count in the center crystal is higher than those in the border crystals. This result indicates that all the channels in the detector modules work properly. B. Energy Resolution The top graph in Fig.7 shows the pulse height spectra for individual crystals in a detector module. A Gaussian function fitting is applied to each spectra to obtain the photo-peak position and energy resolution for individual crystals. Then, based on the obtained photo-peak position the pulse height measurement is calibrated to event energy measurement. The energy resolutions of the crystals vary from 11.2% to 24.6% FWHM at 511keV for a detector module. The bottom plot of Fig. 7 shows the summed energy spectra of one detector module, after applying energy calibration. The overall energy resolution is 15.1% FWHM at 511keV. C. Timing Resolution Figure 8 shows the measured coincidence timing histogram (using the events between 450keV and 650keV ) of a pair of detector modules using the dual-head detection system shown in Fig.5. This result is obtained by filling a single 1.6 mmdiameter channel of the phantom shown in Fig. 9 with FDG and placing the phantom at the center of the setup, by using an energy window of 450keV to 650keV. By fitting a Gaussian function to this histogram, a module-level coincidence timing resolution of 684ps FWHM is obtained.

Fig. 9. A homemade micro-derenzo phantom consisting of six groups of hollow channels having diameters of 2.4mm, 2.0mm, 1.6mm, 1.2mm, 1.0mm, 0.7mm. Also shown is the transverse slices of the reconstructed image of the phantom generated by an maximum likelihood expectation maximization algorithm. The 1.6mm-diameter rods can be resolved. Fig. 7. Top: The pulse height spectra obtained for the 12 12 crystals in a detector module. The energy resolutions of individual crystals range from 11.2% to 24.6%. Bottom: The average energy spectra which indicates an over-all energy resolution of 15.1% FWHM at 511keV. D. Phantom Imaging Figure 9 shows the the reconstructed images, and a photo, of the homemade micro-derenzo phantom that contains six groups of hollow channels having diameters ranging from 0.7mm to 2.4mm. The center-to-center spacing between adjacent channels in the same group is twice the channel diameter. All the hollow channels in the phantom is filled with FDG, and placed at the center of the dual-head detection system with its axis vertical to the detector module. When acquiring the data, both the phantom and the detection system are kept stationary. The coincidence event are then obtained by using a 10ns coincidence timing window and a 450keV -650keV energy window. A model-based image reconstruction approach is used to obtain the tomographic image of the phantom [11]. As shown in Fig.9, the 1.6mm-diameter rods can be resolved in the reconstructed image. In this initial study, no normalization, attenuation correction or random correction is employed. Fig. 8. The coincidence timing histogram obtained for a pair of detector module. By fitting this histogram with a Gaussian function (red curve), we obtain a module-level coincidence timing resolution of 684ps FWHM. IV. CONCLUSIONS AND DISCUSSION In this work, we develop a digital PET detector module. It employs FPGA-only MVT digitizers to generate samples from which event energy and timing are estimated by digital signal processing. The digital signal processing algorithms are implemented in the computer. The experimental results show that our detector module provides an energy resolution in the range of 11.2% to 24.6% FWHM at 511keV (with an overall energy resolution of 15.1%) and a module-level coincidence timing resolution of 684ps FWHM. These performance properties are adequate for human and small-animal imaging. Our

initial imaging result also shows that a PET scanner based on these modules has the capability to resolve 1.6mm structures. V. ACKNOWLEDGE The authors thank Dr. Peng Xiao and Mr. Xiaoqing Cao for discussion. We also thank the PET center of the Union Hospital, Tongji Medical College of Huazhong University of Science and Technology for their generous support with radioisotopes and other resources. REFERENCES [1] H. Kim, C.-M. Kao, S. Kim, and C.-T. Chen, A development of waveform sampling readout board for PET using DRS4, in Nuclear Science Symposium and Medical Imaging Conference (NSS/MIC), 2011 IEEE, 2011, pp. 2393 2396. [2] W. J. Ashmanskas, B. C. LeGeyt, F. M. Newcomer, R. Van Berg, R. Wiener, and J. Karp, Waveform-sampling electronics for timeof-flight PET scanner, in Nuclear Science Symposium and Medical Imaging Conference (NSS/MIC), 2011 IEEE, 2011, pp. 3347 3350. [3] R. Fontaine, F. Belanger, J. Cadorette, J. D. Leroux, J. P. Martin, J. B. Michaud, J.-F. Pratte, S. Robert, and R. Lecomte, Architecture of a dual-modality, high-resolution, fully digital positron emission tomography/computed tomography (PET/CT) scanner for small animal imaging, Nuclear Science, IEEE Transactions on, vol. 52, no. 3, pp. 691 696, 2005. [4] Q. Xie, C.-M. Kao, Z. Hsiau, and C.-T. Chen, A new approach for pulse processing in positron emission tomography, Nuclear Science, IEEE Transactions on, vol. 52, no. 4, pp. 988 995, 2005. [5] Q. Xie, Y. Chen, J. Zhu, J. Liu, X. Wang, W. Liu, X. Chen, M. Niu, Z. Wu, D. Xi, L. Wang, P. Xiao, C.-T. Chen, and C.-M. Kao, Implementation of LYSO/PSPMT block detector with all digital DAQ system, Nuclear Science, IEEE Transactions on, vol. 60, no. 3, pp. 1487 1494, 2013. [6] Q. Xie, C.-M. Kao, X. Wang, N. Guo, C. Zhu, F. H., M. W. W., and C.- T. Chen, Potentials of digitally sampling scintillation pulses in timing determination in PET, Nuclear Science, IEEE Transactions on, vol. 56, no. 5, pp. 2607 2613, 2009. [7] H. Kim, C.-M. Kao, Q. Xie, C.-T. Chen, Z. L., F. Tang, H. Frisch, M. W. W., and W.-S. Choong, A multi-threshold sampling method for TOF-PET signal processing, Nuclear Instruments and Methods in Physics Research Section A: Accelerators, Spectrometers, Detectors and Associated Equipment, vol. 602, no. 2, pp. 618 621, 2009. [8] H. Kim, C.-T. Chen, H. Frisch, F. Tang, and C.-M. Kao, A prototype tof pet detector module using a micro-channel plate photomultiplier tube with waveform sampling, Nuclear Instruments and Methods in Physics Research Section A: Accelerators, Spectrometers, Detectors and Associated Equipment, vol. 662, no. 1, pp. 26 32, 2012. [9] D. Xi, C.-M. Kao, W. Liu, C. Zeng, X. Liu, and Q. Xie, FPGA-only MVT digitizer for TOF PET, Nuclear Science, IEEE Transactions on, vol. 60, no. 5, pp. 3253 3261, 2013. [10] J. Song, Q. An, and S. Liu, A high-resolution time-to-digital converter implemented in field-programmable-gate-arrays, IEEE Trans. Nucl. Sci., vol. 53, no. 1, pp. 236 241, Feb. 2006. [11] C.-M. Kao, Q. Xie, Y. Dong, L. Wan, and C.-T. Chen, A high-sensitivity small-animal pet scanner: Development and initial performance measurements, Nuclear Science, IEEE Transactions on, vol. 56, no. 5, pp. 2678 2688, 2009.