Nuclear Instruments and Methods in Physics Research A 471 (2001) 80–84
A cylindrical SPECT camera with de-centralized readout scheme F. Habtea,*, P. Stenstro. ma, A. Rillberta, A. Bousselhama, C. Bohma, S.A. Larssonb b
a Department of Physics, Stockholm University, Box 6730, 113 85 Stockholm, Sweden Department of Hospital Physics, Karolinska Institute/Hospital, 171 76 Stockholm, Sweden
Abstract An optimized brain single photon emission computed tomograph (SPECT) camera is being designed at Stockholm University and Karolinska Hospital. The design goal is to achieve high sensitivity, high-count rate and high spatial resolution. The sensitivity is achieved by using a cylindrical crystal, which gives a closed geometry with large solid angles. A de-centralized readout scheme where only a local environment around the light excitation is readout supports high-count rates. The high resolution is achieved by using an optimized crystal configuration. A 12 mm crystal plus 12 mm light guide combination gave an intrinsic spatial resolution better than 3.5 mm (140 keV) in a prototype system. Simulations show that a modified configuration can improve this value. A cylindrical configuration with a rotating collimator significantly simplifies the mechanical design of the gantry. The data acquisition and control system uses early digitization and subsequent digital signal processing to extract timing and amplitude information, and monitors the position of the collimator. The readout system consists of 12 or more modules each based on programmable logic and a digital signal processor. The modules send data to a PC file server-reconstruction engine via a Firewire (IEEE1394) network. r 2001 Elsevier Science B.V. All rights reserved. Keywords: SPECT; Cylindrical detectors; De-centralized readout; Early digitization; Optimal pulse processing; Fire wire
1. Introduction Tomographic tracer imaging systems with good imaging ability are superior tools for accurate diagnostics in nuclear medicine. Gamma and positron radiating tracers are detected by single photon emission computed tomographs (SPECT) and positron emission tomographs (PET) respectively. It is also possible to build combined PET/SPECT systems. During the last few years *Corresponding author. Tel: +46-8-164661; fax: +468164580. E-mail address:
[email protected] (F. Habte).
several important technological innovations [1–3] have made it possible to improve the performance of such systems. The important instrument parameters are spatial and energy resolution, sensitivity, cost, dynamic capabilities and count rate limitations. While optimizing these parameters, one often encounters trade off situations where some parameters are improved at the expense of others. Thus, the optimum choice of instrument parameters depends on the most common applications of the instrument under development. The SPECT camera being developed at Stockholm University and Karolinska Hospital [4] is mainly intended for high-resolution brain imaging.
0168-9002/01/$ - see front matter r 2001 Elsevier Science B.V. All rights reserved. PII: S 0 1 6 8 - 9 0 0 2 ( 0 1 ) 0 0 9 1 9 - 6
F. Habte et al. / Nuclear Instruments and Methods in Physics Research A 471 (2001) 80–84
However, it is also designed to support future PET mode operations as well.
2. Design strategy Our design goals have been to achieve high sensitivity, high-count rate and high spatial and energy resolution while minimizing the cost, and to allow future PET mode with reasonable efficiency. It is obvious that minimizing the cost conflicts with optimizing most of the other parameters. There is also a conflict between spatial resolution and sensitivity. Another conflict is between intrinsic SPECT resolution and PET efficiency. The ability to resolve the position of an absorbed gamma photon is the intrinsic spatial resolution. The task of the collimator is to use the intrinsic resolution in order to obtain spatial resolution and sensitivity. The Stockholm camera is therefore designed for high intrinsic spatial and energy resolution with a crystal thickness slightly larger than the optimum for SPECT resolution to improve the PET performance. The optimization has been achieved by considering a combination of detector and scintillator geometry, PMT size and configuration, readout and signal processing architecture, and positioning methods. The optimization deliberations are discussed below. 2.1. Detector geometry A cylindrical crystal was chosen to optimize the sensitivity. This configuration maximizes the acceptance solid angle, improving the gamma– photon absorption efficiency [5,6]. It also allows the detector to fit closely around the patient, keeping the distance from the photon source short and constant. The area coverage of the detector was also chosen to fit the object of interest, in our case the human brain. A factor of 5–10 improvement in detection sensitivity is expected over conventional single head planar gamma cameras with parallel-hole collimators. The only moving part in the cylindrical configuration is a rotating collimator. This leads to significant simplifications in the mechanical design of the gantry, considerably reducing the cost.
81
2.2. Scintillator type and configuration The spatial resolution is sensitive to the crystal and light guide configuration. A continuous cylindrical 12-mm NaI (Tl) crystal optically coupled with a 12-mm light guide gives an intrinsic spatial resolution better than 3.5 mm [10]. For SPECT application, simulation results [7] indicate that a crystal thickness of 6 mm is slightly more optimal when considering the intrinsic resolution. However, for PET mode an increased thickness has definite advantages. NaI (Tl) was used in the prototype camera for its good light output, but other scintillating crystals (e.g. YAP : Ce [8]), with other preferred properties, will be considered in a design upgrade. A different approach is to use an assembly of small discrete crystals coupled to the light guide. This will give a high intrinsic spatial resolution provided that the crystals are small enough. 2.3. PMT size and configuration Optimization of the PMT size and configuration can also significantly improve the detector efficiency [7,9]. Using hexagonal 60 mm PMTs with curved front ends provide better area coverage and an improved light collection efficiency than using round tube configurations. In our case, the energy resolution improved by about 15% using 60 mm hex-shaped tubes [10]. Smaller PMTs would of course ideally also improve the intrinsic spatial resolution, this must be balanced against reduced area coverage due to edge effects, dark currents and steeply increasing cost. 2.4. Collimator design Several studies [11] have shown that an optimized collimator design can improve both sensitivity and spatial resolution to a certain point. The rotating collimator is very flexible. It allows combinations of different types of collimator segments: parallel, diverging, pinhole and slit collimators. These degrees of freedom can be used to emphasize the information from the central regions, in order to compensate the information
82
F. Habte et al. / Nuclear Instruments and Methods in Physics Research A 471 (2001) 80–84
loss due to photon absorption and distancedependent spatial resolution. 2.5. Readout electronics and signal processing An efficient readout system improves the overall system performance. A fast data collection system will increase the count rate and give a good time resolution, facilitating PET mode imaging. Early digitization of the analog signals with the ability to perform digital processing with optimal filters [12,13] improves the precision of amplitude and timing estimates. A considerable effort has been made to build an efficient readout system based on programmable logic and digital signal processors. The system is used to extract optimal amplitude and timing information of the detected signal. The details are described in Ref. [14] and in Section 4. 2.6. Positioning algorithm The ability to precisely determine the location of scintillation events is one of the key issues in both SPECT and PET. Since only PMTs in the immediate neighborhood of the event will capture the scintillating light, positioning is done using the individual PMT responses in a region around the event. The method of maximum likelihood (ML) is used to position the event. Previous work [15,16] has shown that the ML method maps the physical event space much better than the conventional Anger method. The ML estimator also models better the underlying statistical process. It is also able to take crystal edge effects into account extending the axial field of view.
face of the PMTs is polished to the same cylindrical diameter as the light guide. A 1-mm thick aluminum cover on the inner side of the scintillator protects it from humidity and shields it from light. It also has a reflection coating on the side facing the PMTs to reflect light back towards the PMTs. The collimator, driven by a DC motor by means of a geared belt, rotates inside the cylindrical aluminum cover. The collimator position is monitored via a dedicated control module. Fig. 2 shows the physical view of the camera.
4. The readout system The readout system (Fig. 3) consists of 12 data acquisition (DAQ) modules. Each DAQ module supports six PMT channels. One or more additional modules can also be included for control, positioning and image processing applications. A
Fig. 1. SU-SK camera configuration.
3. System description The mechanical design including the detector unit (crystal and PMTs) has been adopted without modification from a previous prototype unit described in Ref. [4]. It consists of a 180 mm long cylindrical NaI (Tl) scintillator with a 305 mm inner diameter and 12.5 mm thickness (Fig. 1). The scintillator is optically coupled via a 12.5-mm light guide to 72 hexagonal PMTs organized in four rows of 18 densely packed 60 mm PMTs. The front
Fig. 2. SU-KS SPECT camera picture.
F. Habte et al. / Nuclear Instruments and Methods in Physics Research A 471 (2001) 80–84
83
Fig. 4. Localized readout pattern.
Fig. 3. Readout system.
DAQ module performs early digitizing and subsequent digital processing using a data driven distributed data acquisition technique [14]. The modular approach apart from cost reduction also allows the acquisition of events detected simultaneously at different locations in the scintillator, which significantly lowers the system dead time. The DAQ modules are interconnected via a dedicated trigger network to form flexible localized readout patterns (FLIRP). When a PMT channel triggered by a sufficiently large pulse distributes trigger pulses to its immediate neighbors to force the read out of an environment of 7, 13, or 19 PMT channels as shown in Fig. 4. Once a FLIRP is formed, samples of all relevant waveforms are stored and analyzed. All readout modules are also connected to each other, to the control module and to the PC-file server engine via a high-speed serial bus IEEE-1394 (Firewire) as illustrated in Fig. 3. This bus is used for transmitting the results to the PC. The de-centralized readout system can be extended to include modules to handle computation-intensive signal and image processing algorithms. 4.1. Hardware implementation Three boards have been designed to implement a DAQ module (Fig. 3). The main board (10 20 cm) consists of a DSP from Texas Instruments (TMSC6202) and a XILINX FPGA (Virtex XCV100) to handle detection, acquisition and processing of pulses. Using the generalpurpose blocks of memory and DLLs, an internal
FIFO is implemented within the FPGA. An LVDS backplane network is used to implement the trigger network, to connect the DAQ modules and to distribute clock and synchronization signals. The ADC board consists of two 3-channel, 8-bit video ADCs (TDA8752AH/8 100 MHz) to sample the pulses. Each channel has programmable gain to compensate for individual PMT variations. The Firewire board is designed to interface the main board to the PC via a Firewire cable. It supports a full 400 Mbytes/s of firewire bandwidth. The board uses the link layer and physical layer controllers TSB21LV01B and TSB41AB3, respectively. 4.2. Software implementation A software framework based on the Firewire platform is described in Ref. [17]. It consists of a Firewire stack implemented in a DSP with the API that custom software can use to access the Firewire bus. Control and configuration of a module is also handled by the platform.
5. Preliminary performance test To evaluate the performance of the camera, incremental performance tests and simulation have been carried out [7,9,10,12]. Motivated by the result obtained, a new final version of readout electronics has been designed and tested. The complete implementation is in progress. Using the prototype of the DAQ that supports three PMTs a time resolution of 5.2 ns and 12.6% energy resolution was obtained for 99mTc [12]. The energy resolution will improve when more PMT channels are used. An intrinsic spatial resolution of around 3.5 mm (140 keV) was measured with data
84
F. Habte et al. / Nuclear Instruments and Methods in Physics Research A 471 (2001) 80–84
obtained from the old data acquisition system. The result is expected to improve using the new version of the readout system. A data rate of 12.5 Mbytes/ s or better can be achieved per module using the new version of the readout system which is expected to improve considerably the count rate of the system.
6. Summary and discussion A cylindrical SPECT camera optimized for brain studies is being designed. The design strategy is to optimize essential design parameters to achieve high sensitivity, high-count rate, high resolution at lower cost, and to include a future PET mode with reasonable efficiency. An optimized detector geometry and configuration improve both sensitivity and intrinsic resolution of the camera system. An efficient digital front-end readout system will improve the position resolution and the count rate. Optimal pulse processing methods have been used to obtain estimate of amplitudes and time of arrival. The measured time resolution motivates the extension of the design to a hybrid SPECT–PET imaging system.
Acknowledgements This work has been supported by a grant from the Knut and Alice Wallenberg Foundation, Stockholm, Sweden.
References [1] S.S. Spencer, et al., Mag. Res. Imag. 13 (8) (1995) 1119. [2] R. Pani, et al., Nucl. Instr. and Meth. A 409 (1998) 524. [3] A.D. Guerra, et al., Nucl. Instr. and Meth. A 409 (1998) 537. [4] S.A. Larsson, et al., IEEE Trans. Nucl. Sci. 38 (2) (1991) 654. [5] S. Kimiaei, et al., J. Nucl. Med. 37 (8) (1996) 1417. [6] J. Liu, et al., IEEE Trans. Nucl. Sci. 42 (4) (1995) 1147. [7] M. Bergqvist, A SPECT light transport simulation program, USIP 98–05. [8] C.W.E. van Eijk, Nucl. Inst. and Meth. A 392 (1997) 285. [9] C.Bohm, M. Bergqvist, Sensor configuration for local sampling of light in scintillation detectors, IEEE Trans. Nucl. Sci., USIP 98–06. [10] S.A. Larsson, C. Bohm, et al., The influence of PM-tubes configuration and intrinsic resolution on the imaging properties of a new cylinderical spect camera, USIP Internal Report 98–04. [11] M.C. Wrohel, et al., Nucl. Inst. and Meth. A 376 (1996) 477. [12] P. Stenstrom, et al., Evaluation of a data acquisition system for SPECT(PET), IEEE Trans. Sci., 2000, 1655. [13] G. Bertuccio, et al., Nucl. Inst. and Meth. A 322 (1992) 271. [14] P.Stenstrom, et al., A new scalable modular data acquisition system for SPECT(PET), IEEE Trans. Nucl. Sci. (1998) 1117. [15] M. Bergqvist, et al., Analysis of the intrinsic spatial resolution for the Stockholm University Karolinska Hospital Cylindrical SPECT camera, Conference Record, IEEE Nuclear Science Symposium and Medical Imaging Conference, 1997. [16] C.S. Butler, M.I. Miller, IEEE Trans. Med. Imag. 12 (1) (1993) 84. [17] A. Rillbert, et al., A General Firewire Data acquisition platform applied to SPECT, IEEE NSS Conference Record 1999.