IEEE TRANSACTIONS ON NUCLEAR SCIENCE, VOL. 52, NO. 3, JUNE 2005 691 Architecture of a Dual-Modality, High-Resolution, Fully Digital Positron Emission Tomography/Computed Tomography (PET/CT) Scanner for Small Animal Imaging Réjean Fontaine, François Bélanger, Jules Cadorette, Jean-Daniel Leroux, Jean-Pierre Martin, Jean-Baptiste Michaud, Jean-François Pratte, Stéfan Robert, and Roger Lecomte Abstract—Contemporary positron emission tomography (PET) scanners are commonly implemented with very large scale integration analog front-end electronics to reduce power consumption, space, noise, and cost. Analog processing yields excellent results in dedicated applications, but offers little flexibility for sophisticated signal processing or for more accurate measurements with newer, fast scintillation crystals. Design goals of the new Sherbrooke PET/computed tomography (CT) scanner are: 1) to achieve 1 mm resolution in both emission (PET) and transmission (CT) imaging using the same detector channels; 2) to be able to count and discriminate individual X-ray photons in CT mode. These requirements can be better met by sampling the analog signal from each individual detector channel as early as possible, using off-the-shelf, 8-b, 100-MHz, high-speed analog-to-digital converters (ADC) and digital processing in field programmable gate arrays (FPGAs). The core of the processing units consists of Xilinx SpartanIIe that can hold up to 16 individual channels. The initial architecture is designed for 1024 channels, but modularity allows extending the system up to 10 K channels or more. This parallel architecture supports count rates in excess of a million hits/s/scintillator in CT mode and up to 100 K events/s/scintillator in PET mode, with a coincidence time window of less than 10 ns full-width at half-maximum. Index Terms—Computed tomography, digital signal processing, field programmable gate array (FPGA), positron emission tomography. I. INTRODUCTION O NE goal of contemporary research in biological imaging, centered on the genetically-modified mouse in this postgenomic era [1], is to achieve better resolution images with Manuscript received June 3, 2003; revised May 12, 2004. The work was supported by the Natural Science and Engineering Research Council of Canada and Le Fonds Québécois de la Recherche sur la Nature et les Technologies. R. Fontaine, F. Bélanger, J.-D. Leroux, J.-B. Michaud, and S. Robertis are with the Université de Sherbrooke, Sherbrooke, QC J1K 2R1, Canada (e-mail: Rejean.Fontaine@USherbrooke.ca). J. Cadorette is with the Centre Hospitalier Universitaire de Sherbrooke (CHUS), Sherbrooke, QC J1H 5N4, Canada. J.-P. Martin is with the Department of Physics, Université de Montréal, Montréal, QC H3C 3J7, Canada. J.-F. Pratte is with the Instrumentation Division, Brookhaven National Laboratory, Upton, NY 11973-5000, USA. R. Lecomte is with the Animal PET Imaging Group, Department of Nuclear Medicine and Radiobiology, Université de Sherbrooke, Sherbrooke, QC J1H 5N4, Canada. Digital Object Identifier 10.1109/TNS.2005.850484 less exposure and shorter acquisition time. Current positron emission tomography (PET) scanners designed for small animal imaging achieve a spatial resolution that makes it possible to detect lesions just over 2 mm in size. For applications in the genetically modified mouse, a 1-mm accuracy would be required all across the field of view of the camera. Furthermore, future PET scanners should be designed to be able to accommodate new incoming detector and signal processing technologies that will help enhance its accuracy. Currently, one major problem in PET scanner designs resides in their low upgrading capacity resulting mainly from their analog electronics architecture. Up until very recently, digital signal processing technologies could not be considered due to size and power limitations, and effort in PET design was focused on the necessity to reduce data processing and power using analog electronics with high-level channel encoding. Unfortunately, this approach makes it difficult to integrate new technologies and new scientific discoveries. In another respect, with the use of new, highly specific PET radiopharmaceuticals that are being developed to very selectively target desired tissues, cells, or cell subsystems, it is becoming increasingly difficult to accurately localize the uptake structure highlighted by the PET tracers within the body [2], [3]. One must rely on anatomical details obtained with another imaging modality, either on a different scanner or with dual-modality systems incorporating a PET camera and a computed tomography (CT) scanner [4]–[6]. This is currently achieved by using two sets of sensors placed side by side: one for PET imaging and the other one for CT imaging. These devices are cumbersome and expensive. Moreover, they suffer limitations due to sequential imaging and possible subject motion between successive scans. To overcome these problems, the concept of multimodality imaging on the basis of a common detection system based on high-performance avalanche photodiode (APD) detectors was first proposed in 1998 [7]. However, due to the complexity of the signal processing from the common PET/CT detector front-end, such an approach would be very difficult to implement in a fully analog scanner design. We propose to realize this original scanner concept using a novel electronic architecture relying heavily on digital signal processing. This approach offers much more flexibility and 0018-9499/$20.00 © 2005 IEEE 692 IEEE TRANSACTIONS ON NUCLEAR SCIENCE, VOL. 52, NO. 3, JUNE 2005 Fig. 1. PET/CT scanner basic operation principle. opens new possibilities for more sophisticated signal processing. The counterpart is higher power consumption and a huge amount of data to process in real time. II. PET AND CT IMAGING PRINCIPLES PET radiopharmaceuticals enable the investigation of biochemical processes at the cellular and molecular level in vivo. -fluorodeoxyglucose, an analog of glucose, For example, will accumulate faster in cancer cells that are more metabolically active than normal cells, making it possible to detect tumors and determine their grade. The disintegration of the PET radioisotope within tissues produces a positron that will soon annihilate with an electron, emitting two collinear 511-keV photons 180 apart. The goal of the PET scanner is to detect and count a large number of these annihilation photon pairs in coincidence to eventually reconstruct the image of the radioactivity distribution using tomographic techniques (Fig. 1). Typically, to a few mCi (10–100 MBq) of radioacseveral hundreds tivity is injected to the subject under study, yielding individual detector singles count rate in the range of 10-100 kcps. In CT imaging, the transmission of X-rays from an external source through the subject is used to obtain an image of the tissue density. Tomographic data are acquired by rotating the X-ray source around the subject while recording the X-ray flux transmitted through the tissues in opposite detectors, as illustrated in Fig. 1. As the flux per unit area can be fairly large photons [8]), CT detec(typically tors conventionally operate in current mode. III. PET/CT SCANNER REQUIREMENTS A prerequisite for using the same detection system for PET and CT imaging is to be able to detect and count individual highenergy annihilation photons (511 keV) and low-energy X-rays (30–60 keV) with the same detector and front-end electronics. This has been shown to be achievable with APD-based multilayered crystal detectors [7] and fast, low-noise, high dynamic range front-end electronics [9]. One further requirement is that the signal from every individual detector must be processed independently to cope with the high-event rate in CT mode and the crystal identification in PET mode. The accuracy and contrast of PET images are dependent on a number of physical and electronic parameters, including the number and size of sensors, the detector speed and front-end electronic noise, and the signal processing sophistication and computational power which affect the ability to perform correction on acquired data. The size of individual detectors essentially determines the intrinsic spatial resolution in PET, which is given to a first approximation by half the detector dimension [10], [11]. In small diameter scanners used for animal PET, the parallax error resulting from the depth of interaction (DOI) within detectors has a nonnegligible effect on resolution. Parallax is the uncertainty on the position of the virtual line traced between thin and tall coincident detectors when viewed at an angle. Image fuzziness increases trigonometrically with parallax as the coincidence occurs further away from the center of cylindrical scanners, unless DOI is measured to reduce location uncertainty roughly to the same value in all directions. One way to measure DOI is to use a phoswich assembly of scintillation crystals having different decay times, which can be identified by analyzing their signal pulse shape [12]–[17]. The major challenge with this approach resides in the capacity to perform accurate DOI identification with minimal signal processing and CPU power. The detector coincidence timing resolution and time skew between detectors mostly determine the time window width that can be used for coincidence detection of annihilation photon pairs in PET. A narrow time window is important to limit the rate of random events, which can seriously degrade the image signal to noise ratio at high-count rates. With APD coupled to bismuth germanium oxide (BGO) detectors and conventional analog timing techniques, a time window larger than 20 ns must be used [18]. A narrower window of less than 10 ns can be considered with faster, high-luminosity scintillators, such as APD coupled to lutetium oxyorthosilicate (LSO) [19], [20]. Using digital techniques, a time window of the order of 10 ns is targeted with APD-BGO detectors [21]. In single photon counting CT mode, only the number of lowenergy X-ray events per pixel needs to be recorded. Besides spatial resolution, which is determined by the size of individual detector pixels and the geometry of the X-ray source, the most critical parameter of a CT scanner based on PET detectors will thus be its count rate capability. To deal with realistic X-ray fluxes from a low-power X-ray tube, the single photon count rate . of individual detectors should reach at least 2 As the X-ray energy is one order of magnitude lower than the PET 511 keV radiation, the front-end analog electronics must be designed with a wide dynamic range and a short pulse decay time to accommodate the large amplitude range and high rate of detector signals. IV. EXISTING PET ARCHITECTURES Most existing PET scanners are built around analog subsystems in order to extract only fairly basic information from the detected events, like detector address, timestamp, and energy FONTAINE et al.: ARCHITECTURE OF A DUAL-MODALITY, HIGH-RESOLUTION, FULLY DIGITAL PET/CT SCANNER 693 Fig. 2. A typical analog PET scanner front-end architecture. (Fig. 2). After an annihilation photon hits a crystal detector (like BGO or LSO), the scintillation light re-emitted by the crystal is collected by a photomultiplier tube or an APD that transforms and amplifies this faint light signal into a measurable charge pulse. This charge pulse is further amplified and converted into a voltage pulse by means of a charge sensitive preamplifier (CSP), and then fed to the front-end analog processing electronics. A timestamp is generated by a leading edge discriminator (LED) or a constant fraction discriminator (CFD) [13]. The LED uses a fixed threshold to trigger the timing pulse, which results in a simple design but the timestamp is amplitude dependent. An amplitude independent timestamp can be obtained with the CFD by setting the triggering threshold at a fixed fraction of the maximum signal amplitude, but at the cost of a more complex design. Moreover, the CFD internal delay must be tailored to the scintillator decay time used. The performance of these two time discrimination techniques degrades in presence of noise [13], [19], [20]. The fast timing pulse generated by the discrimination circuit is fed to analog or digital units for coincidence verification. Most PET scanners also extract the energy of the signal to select true photoelectric events and reject events scattered in the object or detection system. V. NOVEL PET/CT ARCHITECTURE The most promising approach to overcome several of the problems enumerated above and to achieve sufficient system flexibility to be able to perform CT scans and PET scans using the same detectors is to use digital signal processing. The proposed PET/CT architecture, depicted in Fig. 3, consists of three main subsystems. The data acquisition subsystem (DASS) includes the analog front-end, the analog-to-digital converters (ADC), the digital signal processing units, a 500-V APD bias supervisor unit, a clock controller, and a power unit. The coincidence detector subsystem (CDSS) performs the secondary sorting of data for coincidence identification. Finally, the image processing software performs the data normalization, corrections, and rebinning for tomographic reconstruction of the image. In the DASS, each sensor has its own data processing channel composed of the scintillation crystal and APD, the CSP followed by an anti-aliasing filter, the ADC, and a field programmable gate array (FPGA)-based processor unit (Fig. 3). The MAX1198 100-MHz ADC from Maxim [22] was selected for its low power consumption and high sampling rate, which is necessary for accurate timing with high-speed crystals like ) or LuAP ( ) [12], [13]. LSO ( To perform all needed computations for PET and CT data processing, the FPGA architecture is also critical. The digital signal processing unit is based on the XC2S300E Spartan-IIE from Xilinx Inc. This FPGA includes 64 kb of distributed RAM, four delay lock loops, and between 93–300 K equivalent gates [23]. This FPGA can handle clock frequencies beyond 200 MHz, enough for the first phase of the prototype system. The FPGA is preferred to a digital signal processor for its capacity to perform simple distributed computational tasks and its low cost per I/O. The ACQuisition FPGA architecture (Fig. 4, exploded from Fig. 3) is composed of 16 free running ADC controllers based on a local 150-B first in, first out (FIFO), some I/O port controllers, and a level detector. When the level detector detects a signal with enough energy, the local FIFO stores up to 150 samples as one acquisition frame. The acquisition length is based on the minimum time needed to sample the signal rising edge of the slowest for BGO – at the output of the CSP. scintillator – At the end of data sampling, a finite state machine will transfer the data from the FIFO to the main RAM where different processes may take place. In the case of multilayer crystals, the first process consists in determining the time stamp generation and energy computation. Subsequently, DOI is performed by identifying the crystal of interaction using signal theory methods [24]. Although, the proposed system design could possibly be applied to a wider range of DOI solutions, only phoswich detectors are being considered for now. For the time stamp generation, digital versions of analog LED and CFD circuits, as well as more complex techniques that are accessible in a digital architecture can be implemented and assessed [21]. It is worth pointing out that the DOI and timing discrimination circuits can be readily modified and reprogrammed in the FPGAs in order to experiment various strategies or optimize data processing for other crystal scintillators. Once the computation is completed, the channel number is attached to the data and sent through a parallel 8-b port to the sorting FPGA. This XC2S300E SpartanIIE FPGA performs a preliminary timestamp sorting of the data and prepares data words for transmission through a high-speed serial port to the CDSS. In CT mode, data are simply histogrammed in counters activated by the level detector. Counters can be read out on request from the main computer. 694 IEEE TRANSACTIONS ON NUCLEAR SCIENCE, VOL. 52, NO. 3, JUNE 2005 Fig. 3. High-level architecture of the dual-modality PET/CT scanner and exploded view of one DASS board architecture. VI. SELF-ADJUSTING APD POLARIZATION APDs need to be properly reverse biased (between about 350 and 500 V) for optimum timing performance with slow scintillators like BGO, and for maximum signal-to-noise ratio in measuring low-energy X-rays. The DASS includes a 500-V regulator that can adjust the bias voltage for APDs. The system is supervised by a microcontroller implemented with a voltage protection filter against sudden drop or surge of the bias voltage. The subsystem design is depicted in Fig. 5. The microcontroller controls the polarization voltage with an 8-channel, low-power, 10-b TLV5608 digital-to-analog converter (DAC) serially linked with a SPI protocol. DAC voltage is amplified by a factor of 200, and regulated by a linear regulator built around an operational amplifier and a low-pass filter. The low-pass filter acts as an additional security in the system to ensure slow APD biasing at a slew rate not exceeding 5 V/s. A analog multiplexer is used to sense the regulator output low voltage and a 12-b ADC embedded in the microcontroller samples the individual APD bias through a voltage divider to ensure that the biasing circuit is working correctly. The microcontroller can also send data to the scanner main computer for operator monitoring. VII. ACCURACY OF THE TIMESTAMP To achieve coincidence timing resolution better than 10 ns, the accuracy of the timestamp is very important. In analog PET systems, the detector ring is divided into sectors within which no coincidence is allowed. All events triggered within one sector are ORed and forwarded to coincidence units that compare events from opposite sectors through AND gates [18]. This approach requires a huge number of connectors and properly shielded cables to ensure noiseless transmission paths between the detector front-end and the coincidence units. In the architecture proposed, high-frequency, low-cost Ethernet cables replace analog cables. The timestamp synchronization is FONTAINE et al.: ARCHITECTURE OF A DUAL-MODALITY, HIGH-RESOLUTION, FULLY DIGITAL PET/CT SCANNER Fig. 4. Exploded view of the acquisition FPGA (DASS subsystem), showing the various digital processing blocks. Fig. 5. The APD bias regulator and voltage supervisor. the critical part for the accuracy of the coincidence system. For a 1024-sensor PET scanner, 16 DASS boards will be equally distributed around the ring path and each board will be precisely synchronized to provide accurate timestamps. Three factors can influence the accuracy: the delay induced by components and cables; the rise time and fall time of the clock signal; and noise. The architecture proposed to overcome this problem is to distribute a single low-frequency, low-jitter, high-precision clock source (16 MHz) to the entire system through an off-the-shelf, low-voltage positive emitter-coupled logic (LVPECL) buffer located on the CDSS (Fig. 6). Noise immunity is ensured by the use of differential clock distribution within the system. On the DASS board, a differential LVPECL frequency synthesizer is multiplying the 16 MHz to the required 100 MHz. Then low voltage zero delay buffers provide the 100 MHz clock to all FPGAs and ADCs. With such a topology, the worst theoretical clock skew between components on different DASS boards is 850 ps, while the typical value should not exceed 200 ps. Since 695 events are saved with timestamp and detector addresses, clock skew between different detector channels can be estimated by analyzing the coincidence time distribution for each detector pair. A software calibration will be performed in order to correct residual time skew originating from cable length and component delay. VIII. CONCLUSION A dual-mode PET/CT concept using the same detectors has been proposed. An architecture offering the opportunity to integrate new digital signal processing principles in the medical imaging field has been designed. The new proposed PET/CT scanner design, which is characterized by excellent accuracy, modularity, flexibility, and self-adjusting systems, will provide unprecedented performance and reliability for molecular imaging research applications. 696 IEEE TRANSACTIONS ON NUCLEAR SCIENCE, VOL. 52, NO. 3, JUNE 2005 Fig. 6. Distribution of the clock for a scanner with 1024 channels. ACKNOWLEDGMENT The authors would like to thank the Université de Sherbrooke for the technical support in this project. Finally, the authors want to acknowledge the scientific contribution of P. O’Connor, Brookhaven National Laboratory, in the first phase of this project. REFERENCES [1] T. Doetschman, “Interpretation of phenotype in genetically engineered mice,” Laboratory Animal Sci., vol. 49, no. 2, pp. 137–143, 1999. [2] B. Solomon, G. A. McArthur, C. Cullinane, J. R. Zalcberg, and R. J. Hicks, “Applications of positron emission tomography in the development of molecular targeted cancer therapeutics,” Biodrugs, vol. 17, no. 5, pp. 339–354, 2003. [3] T. L. Collier et al., “Assessment of cancer associated biomarkers by positron emission tomography: Advances and challenges,” Disease Markers, vol. 18, no. 5–6, pp. 211–247, 2002. [4] J. Siedel, J. J. Vaquero, J. Pascau, M. Desco, C. A. Johnson, and M. Green, “Features of the NIH ATLAS small animal PET scanner and its use with a coaxial small animal volume CT scanner,” in Proc. IEEE Int. Symp. Biomedical Imaging, Bethesda, MD, 2002, pp. 545–548. [5] A. L. Goertzen, A. K. Meadors, R. W. Silverman, and S. R. Cherry, “Simultaneous molecular and anatomical imaging of the mouse in vivo,” Phys. Med. Biol., vol. 47, pp. 4315–4328, 2002. [6] M. Khodaverdi, F. Pauly, S. Weber, G. Schroder, K. Ziemons, R. Sievering, and H. Halling, “Preliminary studies of a micro-CT for a combined small animal PET/CT scanner,” in Proc. 2001 IEEE Nuclear Science Symp. Conf. Rec., vol. 3, San Diego, CA, pp. 1605–1606. [7] A. Saoudi, D. Rouleau, and R. Lecomte, “A novel APD-based detector module for multi-modality PET/SPECT/CT scanners,” in Proc. 1998 IEEE Nuclear Science Symp. and Medical Imaging Conf. Rec., Toronto, pp. 1089–1094. [8] R. Birch, M. Marshall, and G. M. Ardran, “Catalogue of Spectral Data for Diagnostic X-Rays,” Hospital Physicists’ Association, London, Scientific Report Series-30, 1979. [9] J.-F. Pratte, J. Mouine, C. M. Pepin, D. Rouleau, and R. Lecomte, “Design of a fast shaping amplifier for PET/CT APD detectors with depth-of-interaction,” IEEE Trans. Nucl. Sci., vol. 49, no. 5, pp. 2448–2454, Oct. 2002. [10] S. E. Derenzo, W. W. Moses, R. H. Huesman, and T. F. Budinger et al., “Critical instrumentation issues for resolution smaller than 2 mm, high sensitivity brain PET,” in Quantification of Brain Function, Tracer Kinetics and Image Analysis in Brain PET, K. Uemura et al., Eds. New York: Elsevier, 1993, pp. 25–37. [11] R. Lecomte, “Technology challenges in small animal PET imaging,” Nucl. Instrum. Meth. Phys. Res. A, to be published. [12] M. Streun et al., “Pulse shape discrimination of LSO and LuYAP scintillators for depth of interaction detection in PET,” IEEE Trans. Nucl. Sci., vol. 50, no. 3, pp. 344–347, Jun. 2003. [13] D. B. Crosetto, “A modular VME or IBM PC based data acquisition system for multi-modality PET/CT scanners of different sizes and detector types,” in Proc. 2000 IEEE Nuclear Science Symp. Conf. Rec., vol. 2, Lyon, France, pp. 12/78–12/97. [14] M. Schmand, L. Eriksson, M. E. Casey, K. Wienhard, G. Flugge, and R. Nutt, “Advantages using pulse shape discrimination to assign the depth of interaction information (DOI) from a multilayer phoswich detector,” IEEE Trans. Nucl. Sci., vol. 46, no. 4, pp. 985–990, Aug. 1999. [15] J. Seidel, J. J. Vaquero, S. Siegel, W. R. Gandler, and M. V. Green, “Depth identification accuracy of a three layer phoswich PET detector module,” IEEE Trans. Nucl. Sci., vol. 46, no. 3, pp. 485–490, Jun. 1999. [16] A. Saoudi et al., “Investigation of depth-of-interaction by pulse shape discrimination in multicrystal detectors read out by avalanche photodiodes,” IEEE Trans. Nucl. Sci., vol. 46, no. 3, pp. 462–467, Jun. 1999. [17] R. Lecomte, “A new dual crystal depth sensitive detector for high resolution PET cameras,” J. Nucl. Med., vol. 27, p. 974, 1986. [18] R. Lecomte, J. Cadorette, P. Richard, S. Rodrigue, and D. Rouleau, “Design and engineering aspects of a high resolution positron tomograph for small animal imaging,” IEEE Trans. Nucl. Sci., vol. 41, no. 4, pp. 1446–1452, Aug. 1994. [19] R. Lecomte, C. M. Pepin, M. D. Lepage, J.-F. Pratte, and D. M. Binkley, “Performance analysis of phoswich/APD detectors and low noise CMOS preamplifiers for high resolution PET systems,” IEEE Trans. Nucl. Sci., vol. 48, no. 3, pp. 650–655, Jun. 2001. [20] M. E. Casey, C. Reynolds, D. M. Binkley, and J. M. Rochelle, “Analysis of timing performance for an APD-LSO scintillation detector,” Nucl. Instrum. Methods Phys. Res. A, vol. 504, no. 1–3, pp. 143–148, May 2003. [21] J.-D. Leroux et al., “Time determination by digital signal processing with BGO-APD detectors in positron emission tomography,” in Proc. IEEE/NPSS 13th Real Time Conf., Montreal, QC, Canada, 2003, pp. 1723–1727. [22] MAX1198, Dual 8 bits, 100 Msps, 3.3 V, Low Power ADC with Internal Reference and Parallel Output. Maxim. [Online]. Available: www.maxim-ic.com [23] “Spartan-IIE 1.8 V FPGA Family,” Xilinx Inc., DS077-1, 2.0 ed., 2002. [24] J.-B. Michaud, R. Fontaine, and R. Lecomte, “Real-time pole-andzero-space analysis for discrimination of multilayer detectors used for depth-of-interaction measurement in positron emission tomography,” in Proc. IEEE/NPSS 13th Real Time Conf., Montreal, QC, Canada, 2003.