New-generation small animal positron emission tomography system for molecular imaging

Abstract.

The next generation of discoveries in molecular imaging requires positron emission tomography (PET) systems with high spatial resolution and high sensitivity to visualize and quantify low concentrations of molecular probes. The goal of this work is to assemble and explore such a system. We use cadmium zinc telluride (CZT) to achieve high spatial resolution, three-dimensional interaction positioning, and excellent energy resolution. The CZT crystals are arranged in an edge-on configuration with a minimum gap of $50\mu \mathrm{m}$ in a four-sided panel geometry to achieve superior photon sensitivity. The developed CZT detectors and readout electronics were scaled up to complete significant portions of the final PET system. The steering electrode bias and the amplitude of the analog signals for time measurement were optimized to improve performance. The energy resolution (at 511 keV) over 468 channels is $7.43\pm 1.02\%$ full-width-at-half-maximum (FWHM). The spatial resolution is $0.76\pm 0.1\mathrm{mm}$ FWHM. The time resolution of six CZT crystals in coincidence with six other CZT crystals is 37 ns. With high energy and spatial resolution and the relatively low random rate for small animal imaging, this system shows promise to be very useful for molecular imaging studies.

1. Introduction

Imaging low concentrations and small structures is a driving force in both academia and industry to develop preclinical positron emission tomography (PET) scanners with ultrahigh spatial resolution ($\le 1{\mathrm{mm}}^3$). Including different factors contributing to blurring, a spatial resolution of 0.4 mm is achievable for small animal PET.¹ To reach such a resolution, different scintillator segmented detectors with crystals having dimensions as small as 0.8 and 0.5 mm have been investigated.^2,3

In addition, high-density semiconductor materials, such as cadmium zinc telluride (CZT), have gained particular interest for gamma-ray imaging due to their excellent energy resolution, room temperature operation, and intrinsic spatial resolution defined by an electrode pattern fabrication. The latter eliminates the need for saw cutting of scintillator crystals into tiny elements, which would then require assembly into arrays and a labor-intensive process. The reliable manufacturing of CZT has progressed over the past 30 years. Different manufacturers, such as Redlen Technology (Victoria, Canada) and Imarad Imaging System Ltd. (Rehovot, Israel), have been manufacturing CZT since the 1990s. Several laboratories have investigated the spatial, energy, and timing resolution of individual CZT crystals (with different dimensions and electrode design) as a detector technology for PET scanners.^4–7 Energy resolution better than 2.0% full-width-at-half-maximum (FWHM) at 511 keV and time resolution of about 10 ns FWHM using a 10-mm-thick CZT detector have been shown.^4,5 The results indicate that CZT is a promising detector technology for a high spatial resolution small animal PET system. However, a complete small animal PET based on CZT has not yet been demonstrated.

Such a system is currently under development at the Molecular Imaging Instrumentation Laboratory (MIIL) at Stanford University. Three-dimensional (3-D) positioning CZT detectors comprising $4\mathrm{cm}\times 4\mathrm{cm}\times 0.5\mathrm{cm}$ crystals metalized with cross-strip electrode pattern were developed in our lab.^8–10 The cross-strip configuration reduces the number of electronic readout channels required for the same detector area compared to a fully pixelated anode while providing excellent spatial resolution ($\sim 1\mathrm{mm}$ in plane and $<1\mathrm{mm}$ axial). The ratio of cathode to anode signal amplitude is used to determine the third coordinate, allowing precise 3-D positioning of photon interactions. The electrode pattern design was fully optimized for intrinsic sensitivity, spatial, and energy resolution.¹⁰ In this work, 12 of such CZT detectors and the readout electronics were assembled into a structure to complete a significant portion of the final PET system and to demonstrate successful operation of the system, and algorithms for system calibration and optimized performance.

2. Methods

The final system’s field of view (FOV) is formed from four detector panels in a box-shaped configuration with maximum dimensions of $10\times 10\times 8{\mathrm{cm}}^3$ as shown in Fig. 1. A subsystem was configured as two opposing detection panels containing a total of six CZT detectors in each panel arranged edge-on with respect to incoming photons. CZT crystals with a cross-strip electrode pattern were assembled with flexible circuits according to a selected design¹⁰ to form dual-CZT detector modules (Fig. 2) where each module consists of two CZT crystals. Each $4\times 4\times 0.5{\mathrm{cm}}^3$ CZT crystal in a dual-CZT module has 39 anode strips ($100\text{-}\mu \mathrm{m}$ width and 1-mm pitch), 38 steering electrode strips ($400\text{-}\mu \mathrm{m}$ width), and 8 cathode strips ($4900\text{-}\mu \mathrm{m}$ width and 5-mm pitch). The cathodes and anodes electrodes’ bias voltages are $-500$ and 0 V, respectively. The steering electrode bias was optimized for the highest system energy resolution.

Fig. 1.

Schematic of box-shaped CZT-based small animal PET system.

Fig. 2.

(a) Two CZT crystals are assembled to flexible circuit and stacked based on anode–cathode–cathode–anode configuration to form a CZT module ($4\mathrm{cm}\times 4\mathrm{cm}\times 1\mathrm{cm}$). (b) Schematic of a CZT crystal with cross-strip pattern showing anodes, cathodes, and steering electrodes. (c) Drawing of cross-section of the CZT crystal.

Each CZT crystal is connected to a front-end data acquisition board via an intermediate board, which provides the detector high voltage bias and signal ac coupling. The connections to the detectors are achieved using Kapton flexible circuits, which allow for tight packing of the detector modules and reduce the effect of dead space to increase photon sensitivity. Each front-end data acquisition board uses two commercial RENA-3 application-specific integrated circuits (ASIC) by NOVA R&D Inc. (Riverside) for energy and time measurement of a CZT crystal. The front-end electronics of CZT crystals in each panel are connected to a common backplane data bus in a simple tree structure via a fan-in board. The backplane bus is implemented in hardware using the Xilinx Virtex-5 field programmable gate array. Figure 3 shows a picture of the subsystem.

Fig. 3.

Small animal PET system subassembly comprising two opposite-facing panels of CZT detectors and data acquisition electronics. Inset shows the inside of the detector and board housing.

2.1. CZT Detector Characterization

In this work, CZT crystals were used without blocking layers. A 150-nm-thick gold (Au) layer was deposited on the polished surface of CZT crystal for both anode and cathode electrodes. Conductive silver epoxy was used as the bonding material to assemble the flexible circuit to Au electrodes. The precise alignment of flex bonding pads to the electrode was performed at VJ Technologies. The long-term stability of the current for 12 CZT crystals was monitored over 1 year. Each month the leakage current of the CZT crystals was measured on different days and compared over a 12-month period.

The digital-to-analog converter-based trigger thresholds of each channel’s comparators in the RENA ASIC were optimized with respect to the noise level in the system. ${}^{68}\mathrm{Ge}$ and ${}^{137}\mathrm{Cs}$ point sources were used for calibrating the digitized signal amplitude in analog-to-digital converter (ADC) units and converting to corresponding energy (keV) values. A linear ADC unit-to-keV map was obtained by identifying the spectral features of both sources.

2.2. System Energy Resolution

The subsystem was set to acquire data in singles mode. The energy spectrum from all anodes using a ${}^{68}\mathrm{Ge}$ point source (placed at the center of the FOV) was measured for the 12 CZT crystals covering 468 anode channels and 48 cathode channels. The energy resolution of each channel defined as the FWHM of the photopeak divided by the energy at the center of photopeak was calculated. The energy resolution of the system was defined as the average energy resolution of the 468 anode channels in the system. In order to calibrate for the well-known depth-dependent anode signal due to hole trapping, a scatter plot of anode signal ($A$) versus cathode-to-anode signal ratio ($C/A$) was calculated for each channel and the photopeak cluster was fit using a quadratic function, $f(C/A)=a_0{(C/A)}^2+a_1(C/A)+a_2$. Then each anode signal amplitude is multiplied by $511\mathrm{keV}/f(C/A)$. The energy resolution of the system was optimized for steering electrode bias.

2.3. System Coincidence Time Resolution

To measure the photon interaction time, the RENA ASIC provides a coarse time stamp (CTS) and a fine time stamp for every event. The CTS is recorded by a digital up counter that is incremented on every clock edge using a 50-MHz clock (corresponds to about 20 ns). Whenever a photon interacts and a shaped signal passes the leading edge discriminator, the value of the counter is recorded as the CTS number. The fine time stamp circuit relies on the sampling of two analog signals ($U$ and $V$), which are 90 deg out of phase sine waves. The amplitude ($A_{UV}$) of the sine waves and their frequency ($T_{UV}$) is adjustable with a maximum of 3 V and 1 MHz, respectively. As the two sinusoids are 90 deg out of phase, the loci of points traversed by the point ($U$, $V$) form a circle with radius $A_{UV}$. This is shown in Fig. 4. If the point ($A_{UV}$, 0) on the $U-V$ plane circle is selected to be the $t=0$ reference point, then the time of photon interaction of an event with time measurement ($U_i$, $V_i$) is determined using Eq. (1), where $i=\sqrt{-1}$

$$t=\mathrm{angle}\mathrm{of}\left(\frac{U_i+i{V}_i}{A_{UV}}\right)\frac{T_{UV}}{2\pi }.$$ (1)

The time difference ($\mathrm{\Delta }t$) between two photon interactions can be determined using

$$\mathrm{\Delta }t=\mathrm{angle}\mathrm{of}\left(\frac{U_i+i{V}_i}{U_j+i{V}_j}\right)\frac{T_{UV}}{2\pi }.$$ (2)

A large $A_{UV}$ is desirable for detecting a small angle, and consequently, a small time difference. However, the internal crosstalk between channels in the RENA ASIC causes the trigger threshold to slightly fluctuate with the $U$ and $V$ signals. As the $A_{UV}$ increases, the number of noise triggers increases. Therefore, the $A_{UV}$ was optimized to minimize the number of noise triggers in the system.

Fig. 4.

Continuous time measurement scheme of RENA ASIC-based on $U$ and $V$ sine waves.

For measuring the time resolution of the system, coincidence data were acquired for six CZT crystals in one panel in coincidence with six CZT crystals in the opposite panel using a ${}^{68}\mathrm{Ge}$ point source. Cathode signals with wide electrode strips (5 mm) tend to be sensitive to charge carriers moving deep within the detector and, therefore, trigger more instantaneously after photon interaction.⁸ As a consequence, for implementation of time coincidence event identification algorithm, the CTS number and digitized $U$ and $V$ values of cathodes were used. A time window of $1\mu \mathrm{s}$ based on the CTS number was chosen to select the coincidence events. It is critically important to correct for the center and skew of $U-V$ planes of all the channels. Due to channel-to-channel and board-to-board variation, this correction must be done to calculate the coincidence time information based on Eq. (2). After calculating the coincidence time, the system time resolution was calibrated to correct for systematic errors due to photon interaction location (depth dependence), constant line delays, and crosstalk.¹¹ The time resolution is quantified by the FWHM of the timing histogram peak.

2.4. System Point Spread Function

In order to calculate the system point spread function (PSF), data were acquired from a point ${}^{22}\mathrm{Na}$ source, which was attached to a translation stage and stepped at 0.1-mm increments along the system axis. The PSF was calculated by electronic collimation of the number of detected coincidences over a fixed time window at each point source position. To account for charge sharing, neighboring anodes whose summed energy fell within the photopeak were considered. In the case of charge sharing, the event is counted toward the anode with the highest energy contribution.

3. Results and Discussion

3.1. CZT Detector Characterization

Figure 5 shows the $I–V$ characteristics of CZT detectors used in the system. AC coupling is necessary because the leakage current of our CZT detectors is higher than what can be tolerated by the RENA ASIC. Due to the choice of coupling capacitor (470 pF capacitors rated for 630 V), we have not exceeded a biasing voltage of 500 V. Figure 6 shows the average leakage current of CZT crystals biased at 500 V over 1 year. The error bars represent the standard deviation of the measurement at six different days in each month. The current of the CZT crystals are very stable with a mean value of $\sim 270\pm 20\mathrm{nA}$ per CZT crystal. An automated trigger threshold finder tool was developed to adjust the RENA’s internal 8-bit digital-to-analog converters of each channel above the noise floor. Due to internal variation of 24 RENA ASICs (likely manufacturing tolerance) used in the subsystem, a 2.52% variation in standard deviation, as a percentage of the mean, was observed in mapping the ADC to keV value of 468 channels. After the ADC to keV conversion, the trigger levels in the scaled-up system were found to be in the range of 100 keV. Signals below this level would not be registered.

Fig. 5.

$I-V$ measurement of the CZT crystals at different biasing voltages (controlled by the cathode bias). Current density is calculated as the total current divided by the effective cross-sectional area of the CZT crystal.

Fig. 6.

The average current density of CZT crystals over 1 year (September 2014 to August 2015).

3.2. System Energy Resolution

Figure 7 shows the global energy spectrum of all 468 anodes before and after correction for interaction depth-dependent (i.e., $z$) photopeak tailing observed in CZT crystals. Figure 8 shows the histogram of anode energy resolution for all channels before and after correction. The mean FWHM energy resolution across all 468 anode channels before and after correction is $11\pm 2.04\%$ and $7.43\pm 1.02\%$, respectively. The anodes at the edge of the crystals have one neighboring steering electrode as oppose to two (one on each side) and as a consequence have worse energy resolution.

Fig. 7.

Global energy spectrum of 468 channels before and after interaction depth correction.

Fig. 8.

Distribution of FWHM anode energy resolution for 468 channels from ${}^{68}\mathrm{Ge}$ source (a) before and (b) after anode-to-cathode interaction depth-dependence correction.

Table 1 summarizes the detector performance characterization of the prototype system compared with just two CZT crystals previously reported in Ref. 8. The energy resolution results show that in the scaled-up system, the global energy resolution performance is worse than the results from a single CZT crystal.^8,9 This can be attributed to the overall increased level of electronic noise in the system due to longer trace lengths and cross-coupling between traces on the flexible circuits in the detector modules. In addition, the previously reported energy resolution was from two of our CZT crystals with the highest performance.^8,9 The preinspection of the CZT crystals provided by the vendor classified the performance of CZT crystals into 12 grades based on the number of pixels with worse than average energy resolution. The lower grade number has the best energy resolution. The scaled-up system contains mixtures of 12 crystals with three different grades so that the best crystals are close to the center of FOV and higher-grade crystals are at the edge of FOV. The crystals with similar grades were chosen for each panel.

Table 1.

The energy resolution at 511 keV, PSF, and time resolution for the present system and the previously reported results from two CZT crystals in Ref. 8.

	$E_{\text{res}}$ at 511 keV (%)	PSF (mm)	$t_{\text{res}}$ (ns)
Two CZT crystals tested in Ref. 8	$3.06\pm 0.39$	$0.76\pm 0.1$	Not determined
Twelve CZT crystals	$7.43\pm 1.02$	$0.78\pm 0.1$	$37\pm 5$

The energy resolution of the system was optimized for the steering electrode’s bias. The lower steering electrode reduces the dark current but it drops the photon sensitivity. Table 2 shows the global energy resolution as a function of steering electrode bias before and after correction. The energy resolution of the scaled-up system is very sensitive to the steering electrode bias and improves significantly by increasing steering electrode bias from 60 to 80 V. Beyond this point, the improvement becomes negligible.

Table 2.

Global energy resolution of the system at different steering electrode bias.

	Before calibration	After calibration
Steering electrode bias (V)	FWHM energy resolution (%)
60	$12.74\pm 2.43$	$8.67\pm 1.04$
70	$11.78\pm 2.11$	$7.68\pm 1.06$
80	$11.0\pm 2.04$	$7.43\pm 1.02$
90	$10.94\pm 2.00$	$7.35\pm 1.08$

3.3. System Time Resolution

Figure 9 shows the FWHM energy resolution of the system using a test pulse as a function of $U–V$ signal’s amplitude. The test pulse was chosen to eliminate the effect of the crystals’ quality in the energy resolution. The internal crosstalk between RENA ASIC channels at larger $A_{UV}$ increases electronic noise of the system and worsens the energy resolution. In addition, the coupling of the high frequency digital signal on the $U–V$ signal adds noise to the measured $U$ and $V$, which leads to error in timing information from channel to channel. The $A_{UV}$ was selected to be 1.6 V since at 2 V the system is triggering five times more due to noise. Lower $A_{UV}$ is not desirable since based on Eq. (1), a larger $A_{UV}$ provides a finer time resolution between two points in the $UV$ circle.

Fig. 9.

Best-case FWHM energy resolution using a test pulse (equivalent to a signal generated by 511 keV photon) as a function of $U-V$ signal amplitude. As the $A_{UV}$ increases, the crosstalk between RENA ASIC channels broadens the spectral peaks and increases the lower bound of the energy resolution of the system.

Figure 10 shows cathode–cathode time difference spectrum for the six CZT crystals in one panel in coincidence with six other CZT crystal in opposite panel before correction and after correction based on Ref. 11. Figure 11 shows six CZT crystals’ time resolution versus their energy resolution where a lower number (G1) indicated a better crystal quality reported by the vendor. The crystals with poorer energy resolution have worse time resolution. The measured global time resolution of the system is $37\pm 5\mathrm{ns}$ FWHM, a parameter that we are continuing to optimize. The time resolution of the system compared to a scintillation-based detector is relatively poor, fortunately the randoms fraction for small animal imaging is relatively low. In order to improve the time resolution, we will revise the RENA boards to reduce the noise to be able to increase the amplitude of the $U–V$ signals and re-evaluate the time resolution. In addition, utilizing the high energy and spatial resolution in the CZT detector, we are developing algorithms to use the kinematics of Compton scatters in the CZT crystals to make an estimate of the incoming photon incidence angle as a way to geometrically reject randoms in order to compensate for the lower coincidence time resolution of CZT.

Fig. 10.

Cathode–cathode time difference histogram before and after correction for systematic errors. FWHM coincident time resolution of $37\pm 5\mathrm{ns}$ was calculated by fitting a Gaussian curve to the coincident peak (circle points). The offset in the mean value of the histogram before calibration is due to the line delays intrinsic in data acquisition hardware.

Fig. 11.

Cathode–cathode time resolution of six CZT crystals in panel 1 as a function of theirs energy resolution. The grading of the crystals is labeled as G1, G4, and G10, where a lower number indicates better quality reported by the vendor.

3.4. System Point Spread Function

The coincident events obtained from pairs of anodes facing each other across the FOV while moving the source a total of 4 mm along the $x$-axis in $100\mu \mathrm{m}$ steps are shown in Fig. 12. The PSF was $0.76\pm 0.1\mathrm{mm}$ FWHM, including the $250\mu \mathrm{m}$ diameter of the point source. The figure shows a clear crossing point below the half-maximum levels. The FWHM value of the Gaussian fit including the effect of charge sharing was $0.78\pm 0.1\mathrm{mm}$ (dashed lines in Fig. 12). As shown in Table 1, the resulting PSF is comparable to previously reported results. As the system was scaled up, assuming negligible misalignment, we do not expect the PSF to be affected.

Fig. 12.

Coincidence PSF across three anode strips without (solid lines) and with (dashed lines) considering the effect of charge sharing.

4. Conclusion

In this study, we developed a prototype subsystem consisting of 12 CZT crystals as part of a CZT-based small animal imaging PET system. The successful operation of full data acquisition chain and data processing algorithms was demonstrated. The values of steering electrode bias (80 V) and sine wave amplitude (1.6 V) for time measurement were optimized for energy resolution. The initial data from the prototype system illustrate good energy resolution ($7.43\pm 1.02\%$) and high spatial resolution ($0.76\pm 0.1\mathrm{mm}$). The global time resolution of the system based on six CZT crystals in coincidence with six other CZT crystals is $37\pm 5\mathrm{ns}$ FWHM. We are continuing to optimize the time resolution. With high energy and spatial resolution and the relatively low random rate for small animal imaging, this system shows promise to be very useful for molecular imaging studies.

Biographies

Shiva Abbaszadeh is a postdoctoral fellow in the Department of Radiology at Stanford University. Her research interests include radiation detection and instrumentation for molecular imaging, computational problem solving, and quantitative characterization of biological processes. She is the author of 20 peer-reviewed journal articles and has a patent on an x-ray imaging device structure.

Craig S. Levin is a professor of radiology and, by courtesy, of physics, electrical engineering, and of bioengineering at Stanford University. His research interests include the development of instrumentation and software algorithms for in vivo imaging of molecular signatures of disease in humans and small laboratory animals.

Disclosures

No conflicts of interest, financial or otherwise, are declared by the authors.