Validation of the design of a high-sensitivity and high-resolution PET system for a preclinical PET/EPR hybrid scanner

Abstract

We report the development of a high-sensitivity and high-resolution PET subsystem for a next-generation preclinical PET/EPR hybrid scanner for investigating and improving hypoxia imaging with PET. The PET subsystem consists of 14 detector modules (DM) installed within a cylindrical supporting frame whose outer and inner diameters are 115mm and 60mm, respectively. Each DM contains eight detector units (DU) in a row and each DU is made of a 12×12 array of 1×1×10mm³ LYSO crystals (with a 1.05mm pitch) coupled to a 4×4 silicon photomultiplier (SiPM) array that has a 3.2mm pitch (Hamamatsu multi-pixel photon counter (MPPC) array 14161–3050HS-04). The PET subsystem has a 104mm axial field-of-view (AFOV) that is sufficient for full-body mouse imaging, therefore enabling temporal and spatial correlation studies of tumor hypoxia between PET and EPR. It employs 1mm-width crystals to support sub-millimeter image resolution that is desired for mouse imaging. Al-though a DM contains 1,152 LYSO crystals, by use of a newly devised signal readout method only six outputs are produced. Recently a partial prototype of this subsystem consisting of four DMs is built. In this paper, we present performance measurement results obtained for the developed DMs and initial imaging results obtained by the prototype. The developed DMs show uniformly superior performance in identifying the hit crystal and detector unit, in energy resolution, and in coincidence time resolution. The images obtained for a ²²Na point source and a ¹⁸F-filled U-shaped tube source show an image resolution of about 1.1mm and 1.2mm FWHM in the transverse and axial directions respectively, and demonstrate successful imaging over the entire 104mm AFOV of the prototype. This estimated image resolution however includes the contribution by the source size.

1. Introduction

As hypoxic tissues [1, 2] are more resistant to radiation treatment, elevating the radiation dose to the hypoxic region of a tumor may lead to improved cancer treatment [3]. In the clinic, hypoxia imaging can be achieved by PET with ¹⁸F-fluoromisonidazole (FMISO) [4, 5]. However, the effectiveness of dose boosting to hypoxic tumor regions delineated on FMISO-PET images in improving tumor control cannot be established [6]. It is hypothesized that the failure is due to the inaccuracy of FMISO-PET hypoxia imaging due to the multi-factorial dependence of the FMISO uptake in addition to the oxygen concentration. On the other hand, it has been shown that Electron Paramagnetic Resonance (EPR) can produce precise in vivo oxygen concentration measurements in animal tumor models [7] but the technology is presently not clinically available. The above observations have motivated our efforts in developing a preclinical PET/EPR hybrid scanner to provide a research platform for studying and improving PET-based hypoxia imaging using quantitative oxygen maps produced by EPR as the ground truth for validation. The resulting quantitative PET-based hypoxia imaging techniques can be readily translated for clinical use to improve cancer treatment. Previously, we built an exploratory system [8] for the purpose of demonstrating the technological feasibility of hybrid PET/EPR imaging modality. The EPR subsystem of this technology demonstrator consisted of a 25mT permanent magnet, three planar orthogonal gradient coils (capable of a maximum gradient of 30mT/m), and a loop-gap radio-frequency (RF) resonator running at 725 MHz. Its PET subsystem contained 14 detector modules (DM) arranged on a ring that was sufficiently compact to be installed between the EPR permanent magnets. The DM contained an 8×4 array of individually coupled LYSO/silicon-multiplier (SiPM) detector pixels having a 3.2mm pitch. The two subsystems were aligned so that the centers of their imaging FOVs coincided. Figure 1 shows a photo of this demonstrator and the simultaneously acquired FMISO-PET and EPR images of a tumor implemented on the flank of a mouse that it produced. The imaging protocol and quantitative image analysis results obtained from the earlier 10–16 mice images by using the PET/EPR were reported [8–10]. Although successful as a technology demonstrator, the 25.4mm axial field-of-view (AFOV) and 1.6mm image resolution of its PET subsystem were too limited for many mice imaging applications. The short AFOV also leads to low sensitivity.

Figure 1:

(a) The present PET/EPR technology demonstrator has a limited 25.4mm AFOV and 1.6mm image resolution for mouse imaging. The spacing between the EPR magnets into which the PET subsystem is fitted is about 12cm. (b) The PET subsystem before integration into the field-of-view of the EPR. (c) Simultaneously acquired FMISO-PET and EPR-pO₂ images of a tumor implemented on the flank of a mouse. The pink curves show the tumor boundaries delineated on a co-registered MR T2-weighted image.

To address the deficiencies of the technology demonstrator in the AFOV, resolution, and sensitivity in PET imaging, we are developing a new PET subsystem that has a four-times extended AFOV of 104mm and uses three times smaller LYSO crystals of 1×1×10mm³ to support a sub-millimeter image resolution. The extended AFOV enables full-body mouse imaging and hence allows one to study spatial and temporal correlations between the PET and EPR images more thoroughly. By Monte-Carlo calculations, the new PET subsystem has a 8.3% sensitivity when accepting ≥400keV events and assuming a 15% energy resolution. Due to the extension of the AFOV and the use of small crystals, compared to the technology demonstrator the number of detection channels has greatly increased with this new subsystem. Recently, we reported the successful development of a longer and higher-resolution DM [11] that contains as many as 1,152 LYSO crystals and 128 SiPMs but maintains the needed geometrical compactness and electronic simplicity for integration into the field-of-view of the EPR system by introducing a newly devised two-tiered multiplexing readout to produce only six outputs for each DM.

We have built four such DMs and employed them to develop a partial prototype of the new PET subsystem. In this paper, we present performance measurement results obtained for these DMs that demonstrate that the design can yield uniformly superior performance properties. We also show initial imaging results with the partial prototype for a ²²Na source and a ¹⁸F-filled U-shaped tube source, and report the assessed spatial resolution of these images.

2. Material and Methods

2.1. The next-generation PET subsystem and the detector module

The designs of the new PET subsystem and its DM have been recently described in [11]. For completeness, they are briefly summarized below.

Figure 2(a) shows a drawing of the new PET subsystem. It has an inner diameter (ID) of 60 mm, an outer diameter (OD) of 115 mm, and an AFOV of 104 mm. It consists of 14 DMs, each of which is a linear array of eight detector units (DUs). A DU in turn contains a 12×12 array of 1×1×10mm³ LYSO crystals (JTC, China) read by a 4×4 array of SiPMs (Hamamatsu 14161–3050HS-04). Except for that coupled to SiPM, all sides of the LYSO crystal are covered with BaSO₄ to reduce light loss, resulting in a pitch of 1.05 mm. The SiPM pixel has an active detection area of 3.0×3.0mm² and a pitch of 3.2 mm. Table 1 summarily compares the key design parameters of the PET subsystems of the technology demonstrator and the next-generation system. Compared to the former, the latter has significantly increased numbers of LYSO crystals and SiPMs, and hence a greatly increased complexity.

Figure 2:

(a) A drawing of the PET subsystem of the next-generation PET/EPR system. For reference, the blue-colored segment indicates the 25.4mm PET AFOV of the present technology demonstrator. (b) A schematic showing the highly multiplexing readout that produces six outputs (A-D for event energy and crystal identification, and SL1&SL2 for event timing and DU identification) for the new DM consisting of eight DUs (S1-S8). (c) The RN circuit used for a 4×4 SiPM array.

Table 1:

Key design parameters of the PET subsystems.

	technology demonstrator	next-generation system
AFOV	25.4 mm	104 mm
LYSO crystal size	3×3×10mm3	1×1×10mm3
no. of LYSOs	448	16,128
no. of SiPMs	448	1,792

Figure 2(b) depicts the newly devised two-tiered multiplexing readout for efficient handling of the greatly increased number of detection channels. For each DM, the four outputs of the resistive network (RN) [12] applied to read the anode outputs of the 4×4 SiPMs of a DU are wired together to produce four outputs (A, B, C, and D). These outputs are processed following the conventional Anger-logic expression to produce the pulse height (hence, the event energy) and 2-d event positions (hence, the ID of the hit crystal in a DU). On the other hand, the 16 cathode outputs of the SiPM array of a DU are wired together to feed into a strip-line (SL) [13] to produce two outputs (SL1 and SL2). The average of the signal occurring times observed at these outputs yields the event time while their difference is used for identifying the hit DU in the DM. In this way, the readout produces only six outputs for 1,152 crystals in a DM from which all necessary event information can be derived.

Figure 3(a) shows one of the developed DMs, implemented on a 30×185mm² readout printed circuit board (PCB) with RN and SL circuits marked and the DU blocks numbered. Two DUs are installed on a smaller PCB that distributes a bias voltage from the readout PCB to all the SiPMs and connects the SiPM outputs to the readout PCB. A common bias voltage is applied to all DUs in the DM. The DM uses micro-miniature coaxial (MMCX) connectors and cables for connecting its six outputs to the data acquisition (DAQ) electronics. Figure 3(b) shows a zoom-in view of the SL circuit. A serpentine 25 mm-length trace between adjacent injection points on the SL can be observed. It is introduced to increase the difference between the signal arriving times at the two outputs of the SL so that the hit DU can be accurately identified by using time-to-digital (TDC) converters that are available off-the-shelf or can be implemented using the field programmable gate arrays.

Figure 3:

(a) A photo of the DM and its components. (b) A close-up of the SL circuit.

2.2. The partial prototype and experimental setup

Figures 4(a)&(b) shows the partial prototype of the new PET subsystems by employing four newly developed DMs, installed in a 190mm long cylindrical supporting frame having a 115mm OD and a 60mm ID. The four DMs were placed with a 90° separation between neighbors, forming two opposing coincidence-detection pairs: DM1 with DM3 and DM2 with DM4. For energy and position measurements, a 19cm long ⁶⁸Ge line source, having approximately 3.0 MBq (80 μCi) of activity, was used to provide broad irradiation to all DUs of the DMs. Imaging data were acquired by placing a ²²Na point source, having approximately 0.7 MBq (20 μCi) of activity, at z = ±5mm from the axial center. To demonstrate its long AFOV imaging capability, a 40cm length PE-50 tube (OD = 0.96mm and ID = 0.58mm) filled with ¹⁸F solution (6.5 MBq of activity) was also imaged. As shown in Fig. 4(c), the tube was looped in the middle to form a U-shape and inserted into two guiding holes (ID = 1.4mm) of a 3D-printed tube holder so that the two legs of the U-shape were parallel to the z axis. The two guiding holes were at 2.5mm from the z axis so that the distance between their centers was 5.0 mm. This partial prototype did not provide sufficient sampling of the 5cm-diameter transverse field-of-view. Therefore, when performing imaging the source was rotated 22.5° four times in order to obtain sufficient sampling.

Figure 4:

(a&b) A front and side view of the partial prototype. (c) The setup for imaging a plastic tube filled with ¹⁸F solution.

Figure 5(a) shows a diagram illustrating the experimental setup. For all experiments, the prototype was placed inside a light-tight box. The 24 output signals produced by the four DMs were connected to RF amplifiers (Mini-Circuits ZFL-100LNB+) outside the light-tight box and individually amplified. Then, they were digitized by using two CAEN DT5742 waveform digitizes, operating at the sampling rate of 0.7 GSPS (giga samples per second). For energy and position measurements with the ⁶⁸Ge line source, 80 million(M) events were collected. For imaging studies, a total of 80M and 40M events were acquired for the ²²Na point source and ¹⁸F-filled tube, respectively. The acquired data (~4.4TB total) are stored on the hard drive of the DAQ computer for off-line analysis. Figure 5(b) shows the noise level of the SL and RN output signals produced by one of the DMs when varying the SiPM bias voltage. The noise level was computed as the standard deviation of the first 100 waveform samples of the acquired events (lasting ~145 ns) that by setup could not contain any part of a true signal. As the bias voltage increases, the noise level remains rather constant initially but increases sharply when it raises above 43.5V. Therefore, in all the experiments conducted, a single SiPM bias voltage of 42.5V was used for all four DMs to yield a high gain without drastically increased noise. For reference, the nominal breakdown voltage of the SiPMs used is about 38.5V.

Figure 5:

(a) A diagram showing the experimental setup. (b) The noise level of RN and SL outputs of one of the DMs as a function of the SiPM bias voltage.

2.3. Data analysis

The CAEN DT5742 waveform digitizer produces 1,024 samples for a signal. By using the 0.7 GSPS sampling rate, the sampling duration was 1.46 μsec. Figures 6(a)&(b) show exemplary results of the acquired SL and RN signals.

Figure 6:

Exemplary waveforms of (a) the two SL outputs and (b) the four RN outputs.

Event information is extracted from the sampled waveform data as previously described in [11]. For completeness, they are summarized below.

• Calculation of pulse heights (PH): Integrating the sampled waveforms obtained at the RN outputs A, B, C, and D over the 1.46 μsec sampling duration yields the PHs A, B, C, and D. The event pulse height Z is calculated by

  $$Z=A+B+C+D$$ (1)
• Identification of the hit DU: Let t_SL1 and t_SL2 be the signal occurring times at SL1 and SL2. The SL differential time is calculated by

  $${\delta t}_{SL}=t_{SL1}-t_{SL2}.$$ (2)

  The hit DU is determined based on discriminating δt_SL.

• Identification of the hit crystal: The X and Y positional values of an event are calculated from the PHs using the Anger logic:

  $$X=(B+C)/Z,\hspace{1em}Y=(A+B)/Z.$$ (3)

  The hit crystal is determined based on discriminating (X,Y ).

• Calculation of the event time: The event time is calculated by

  $$t_{\mathit{\text{event}}}=(t_{SL1}+t_{SL2})/2.$$ (4)

  The time pick-off methods for obtaining t_SL1 and t_SL2 in Equations 2 and 4 can be different.

3. Results

3.1. Performance in DU identification

Figure 7 shows the δt_SL histograms obtained for all four DMs using only the photopeak events whose energy is between 420 keV and 600 keV (see section 3.3 for determination of event energy) and by employing digital constant-fraction discrimination (CFD) with a threshold fraction of 0.35. As discussed in [11], for calculating δt_SL for identification of DU we found that CFD is superior to leading-edge discrimination (LED) for time pick-off because the two SL signals can have slightly different amplitudes as they travel different SL path lengths. All four histograms show eight distinct peaks that correspond to the eight DUs in a DM. The average peak-to-valley ratios (P/V) are also calculated and shown. All P/V values are high, exceeding 26, but DM1 and DM3 (especially DM1) have lower values than DM2 and DM4. Since the DUs in the DM1 and DM3 were built first by using LYSO and SiPM arrays purchased earlier, we speculate that the higher P/V values of DM2 and DM4 reflect improvements in the quality of the LYSO and SiPM arrays purchased more recently and in our build of the DUs as we gain experience.

Figure 7:

Histograms of the SL differential time δt_SL obtained for all developed DMs by using photo-peak events whose energy is between 420 keV and 600 keV.

As marked by the vertical arrows on the horizontal axis in Figure 7(d), the valleys between peaks were identified and their values were used to define the discrimination boundaries for identifying the hit DU for a given δt_SL.

3.2. Performance in crystal identification

The hit LYSO crystal within the identified hit DU was determined based on the distances between the event (X,Y ) value calculated by using Equation 3 and the 12×12 peaks identified on the (X,Y ) histogram (flood map) obtained for that DU. Figure 8(a) shows an exemplary flood map obtained. The peaks in the map correspond to the 12×12 crystals in a DU and an event is assigned to the crystal whose associated peak is nearest the (X,Y ) value of the event. The algorithm described in [11] was successfully applied to identify all the peaks in the flood maps of all DUs in all four developed DMs. For assessing the quality of crystal identification, following Du. et al [14], we calculated the distance-to-width ratio (DWR) that, as illustrated in Figure 8(b), was defined as the ratio of the distance between two adjacent peaks in a flood map to the average width of these peaks. Thus, a greater DWR means a better separability of the peaks. In computing DWR, the position and width of a peak were estimated from the Gaussian fit to the peak. Also, we compute a mean DWR for a DU by taking the average of the DWRs of all adjacent peaks in the flood map of the DU. Figure 8(c) shows the mean DWRs obtained for all DUs in all four DMs. We observe that all the mean DWRs exceed 2.7, and DUs in the same DM have very similar mean DWRs. Also, DMs built in the same batch (DM1-DM3 group versus DM2-DM4 group) have similar mean DWRs.

Figure 8:

(a) An exemplary flood map (histogram of (X,Y )) of one of the DUs showing 12×12 peaks of which the 8×8 central peaks are well separated and those at the edges remain distinguishable. (b) A profile in Y -direction for the two peaks inside the region marked in (a). DWR was calculated as the ratio of the distance between the peaks to their average width $(9.45/(\frac{2.3+2.2}{2})=4.2)$. (c) Mean DWRs calculated for all DUs in all four DMs.

3.3. Performance in event-energy measurement

The PH spectrum for a crystal was obtained by histogramming the Z values of the events that were assigned to that crystal. From the PH spectrum, the PH value of the photopeak was identified. Figure 9(a) is an exemplary image showing the normalized photopeak PH obtained for the 12×12 crystals in a DU. The normalization was done by linearly scaling the photopeak PH such that the mean of the scaled photopeak PH in a DU equaled 100. The exemplary image is rather uniform for the central 8×8 crystals and the photopeak PHs of the edge crystals are lower, especially for the four corner crystals. It is possible that the drop is because scintillation lights in edge crystals are more likely to escape before detection.

Figure 9:

(a) An exemplary image showing the normalized photopeak PHs for the 12×12 crystals in a DU. (b) Means and standard deviations of the photopeak PHs of crystals that share the same X index in a DU, separately shown for each DM. (c) Means and standard deviations of the energy resolutions of crystals that share the same X index in a DU, separately shown for each DM.

We also computed the mean and standard deviation of the photopeak PHs (un-normalized) obtained for crystals sharing the same X index in all DUs of a DM. Figure 9(b) plots the resulting means against the crystal X index for all four DMs, with the error bars indicating the standard deviation. Again, we observe that the edge crystals (X index = 1 or 12) have lower mean photopeak PHs than the central crystals. Also, the DM1-DM3 group has a lower average photopeak PH than the DM2-DM4 group (163.8 versus 188.3). This possibly reflects that, compared to the DM1-DM3 group, SiPMs of the DM2-DM4 group have lower breakdown voltages and hence yield higher gains when the same bias voltage is used for all four DMs. Energy resolution at 511keV was also calculated for each crystal as the ratio of the full-width-at-half-maximum (FWHM) to the PH value of the photopeak. Figure 9(c) similarly shows the mean and the standard deviation of the energy resolutions of crystals having the same X index. Despite the difference in the photopeak PH between the DM1&DM3 and DM2&DM4 groups, all four DMs have almost the same energy-resolution profile that shows a rather uniform value of about 14% for the central crystals and degraded values of about 15–17% for the edge crystals. The energy resolution averaged over all 4,608 (12×12×8×4) crystals in all four DMs is 14.8±2.0%.

The PH of an event identified for a crystal was converted to energy by multiplying with the factor 511keV/Z_peak where Z_peak is the photopeak PH identified for the crystal.

3.4. Performing in event timing

Coincidence time resolution (CTR) was measured for two opposing DM pairs, i.e., between DM1 and DM3 and between DM2 and DM4. As discussed in [13], when computing event time using Equation 4 for coincidence detection, time pick-off was done by using digital LED with a 4 mV threshold. Figures 10(a)&(b) show the histograms of the event time difference measured for the DM1-DM3 and DM2-DM4 pairs by using photo-peak events whose energy is between 420 keV and 600 keV. CTR was given by the FHWM of the Gaussian fit on the histogram. The same 1.8ns CTR was obtained for both DM pairs.

Figure 10:

(a) A histogram of the event time difference (black curve) obtained between DM1 and DM3, whose Gaussian fit (red curve) has a 1.8ns FWHM. (b) The histogram and FWHM obtained between DM2 and DM4.

3.5. Preliminary imaging results and assessment of image resolution

From the acquired events, coincidence pairs were obtained by accepting 420–600 keV photopeak events and using a 10ns coincidence time window. Reconstruction of the coincidence data was performed by using the maximum-likelihood expectation-maximization (MLEM) algorithm [15] implemented in the CASTOR package [16] that calculates the system matrix using Siddon’s ray tracing algorithm [17]. Identical detectors were assumed. The resulting image consisted of 144×144×208 0.5mm-side cubic voxels. There were no corrections for attenuation, randoms, and scatter. Twenty (20) iterations were used.

For the ²²Na point source, from the acquired ~80M events, approximately 8M coincidence pairs were derived. Figure 11(a) shows the transverse slice through the peak-intensity voxel of the reconstructed image when the source was placed at z = 5mm. Figures 11(b)&(c) show, respectively, the intensity profiles in the x and y directions through the peak-intensity voxel. By fitting with Gaussian functions, the FWHMs of these profiles were measured to be 1.1mm. Figure 12(a) shows the coronal image in the x = 0 plane of the sum of the two images of the ²²Na source when it was placed at z = ±5mm. Figure 12(b) shows the intensity profile of the sum image along the z axis. The distance between the centers of the two sources seen in the image was measured to be 9.4 mm, which is close to the known 10mm physical separation between the two source positions. The average FWHM of the z-profiles of the two sources was estimated to be 1.2mm, which is comparable with the above 1.1mm FWHM transverse resolution. However, the full-width-at-tenth-maximum (FWTMs) of these z-profiles are about two times as large as that in the transverse directions (3.9mm versus 2.0mm). We speculate that this is caused by the depth-of-interaction (DOI) blurring associated with many axially oblique lines of response produced with the 104mm AFOV of the prototype.

Figure 11:

(a) The transverse slice of the image of a ²²Na point source placed at z = 5mm through the peak-intensity voxel. (b)&(c) Intensity profiles in the x and y directions through the peak-intensity voxel. Also shown are the Gaussian fits (red curves) of the profiles and their FWHM and FWTM values.

Figure 12:

(a) A coronal view in the x = 0 plane of the sum of the two images of the ²²Na source placed at x = ±5 mm. (b) The intensity profile of the sum image along the z axis.

For the ¹⁸F-filled tube, 2.5M coincidence pairs were obtained after energy qualification and coincidence filtering. The coronal view in the x = 0 plane of the resulting image in Figure 13(a) shows the U-shaped tube extending to the edge of the AFOV, demonstrating that successful imaging could be achieved over the entire AFOV. Figure 13(b) shows an intensity profile along the vertical line marked by the yellow arrows in Figure 13(a). The distance between the centers of the two peaks was estimated to be 4.9mm, which is very close to the known physical separation of 5.0mm of the two legs of the U-shaped tube. The FWHM of the intensity profiles were 1.2mm, which is consistent with the above results obtained by using the ²²Na source. It is noted that this FWHM includes the contribution by the size of the tube source. As a crude estimate, if the contributions of the source size and of the resolution of the system adds in quadrature, given the 0.96mm OD of the tube the image resolution of the prototype is estimated to be approximately 0.72mm.

Figure 13:

(a) A coronal view of the image of the ¹⁸F-filled tube. (b) The intensity profile along the vertical line identified by the yellow arrows in (a).

4. Conclusion and the next step

We are developing a new PET subsystem for a preclinical PET/EPR hybrid scanner that has a 104mm-length AFOV for supporting full-body mouse imaging and achieving a high system sensitivity and employs 1.0mm-width crystals for supporting sub-millimeter image resolution that is often needed in mouse imaging applications. To achieve electronic simplicity and geometric compactness desired for the subsystem, a two-tiered highly-multiplexing readout that combines the use of RN and SL readouts was introduced. Previously, we reported in [11] successful development of the compact DM employing the new readout for the subsystem. Recently, we have built four such DMs and employed them to develop a partial prototype of the proposed PET subsystem. In this paper, we reported measurement results that demonstrated the developed DMs had uniformly superior performance properties, indicating the robustness and repeatability o the DM design. Successful imaging results of point and U-shaped tube sources were achieved with the partial prototype. Close to the axis of the prototype, images of a ²²Na source showed a resolution of approximately 1.1mm and 1.2mm FWHM in the transverse and axial directions, respectively. In comparison, the FWTMs were about 2.0mm and 3.9mm in the transverse and axial directions. The much degraded axial FWTM was possibly due to elevated DOI blurrings in the axial direction with a long-AFOV system. The tube imaging experiment demonstrated that successful imaging could be achieved for the entire 104mm AFOV. Also, after correcting the tube size, the result suggested that the prototype could achieve a sub-millimeter image resolution.

Based on the successful results reported above for the partial prototype, we have now developed fourteen (14) new DMs that employ the same design as the current DM but the PCB size is further reduced to 175mm×20mm and the input bias voltages and output signals are carried by a thin bundle using the FCF8 micro-coaxial cable assembly (Samtec, US). Figures 14(a)&(b) show the full PET subsystem assembled by installing these DMs onto a supporting frame. In addition, we are evaluating using the BASP-10011 boards (Brightonix Imaging, Korea) for the subsystem. A BASP-10011 board, shown in Figure 14(c), is a small board that provides 16 free-running ADCs (12bit, 80MSPS) and 16 TDCs implemented in a field-programmable-gate-array (FPGA) chip (Xilinx Artix-7) for real-time event processing. Four such boards provide 64 ADC (64 TDC) channels that are sufficient for handling the 56 RN outputs (28 SL outputs) of the PET subsystem. As shown in Figure 14(d), the board is measured to have a 42ps FWHM intrinsic timing resolution, which is more than adequate for processing the SL outputs (approximately 200ps FWHM resolution is needed for producing the δt_SL histograms in Figure 7). Our initial tests with the board for the new PET subsystem have been successful and we are optimizing this BASP-10011 based practical DAQ solution.

Figure 14:

(a) A front view of the newly assembled PET subsystem made of 14 new DMs onto a supporting frame that is 185mm long and has a 60mm ID and a 115mm OD. (b) A side view of the PET subsystem. Thin input/output cable bundles for the DMs can be observed on top. (c) A 50×100mm² BASP-10011 board providing 16 TDC and 16 ADC channels. (d) A time-difference histogram obtained by injecting test pulses to two TDC channels of BASP-10011, showing a 42ps FWHM intrinsic timing resolution.