2.1 Design of DOI detector

The scintillator block is a home-made DLO-LYSO array with a crystal size of 2 mm×2 mm×7 mm (SIPAT, China). All surfaces of the crystals are polished. The top layer is a 15×15 array of 31.6 mm×31.6 mm, while the bottom layer is a 16×16 array of 33.6 mm×33.6 mm. The top layer is placed half crystal offset in two dimensions on the bottom layer (Fig. 1a). Due to the staggered design, the detector has better sampling density than common designs and can achieve better image performance for a whole system ^[31]. Between the crystals are ~ 0.1 mm-thick enhanced spectral reflector films (3M).

Fig.1.Full-screen View

scheme of the DLO LYSO-SiPM detector module (a), and a home-made SiPM board with two 8×8 arrays (b).

As shown in Fig. 1(b), two 8×8 SiPM arrays were fabricated on the same board based on the MicroFJ-30035-TSV SiPM (SensL Technologies Ltd., Ireland), which is an industry-leading low dark noise, MR-compatible and high PDE sensor through silicon via package ^[20]. Each individual SiPM has 5676 microcells (35 μm×35 μm) with a package size of 3.16 mm ×3.16 mm. The 8×8 array has an overall dimension of 33.7 mm× 33.7mm with 1-mm gaps between adjacent SiPM pixels. In this study, only one of the two SiPM arrays was used.

2.2 Data acquisition system

The output channels of the 8×8 SiPM array including 64 standard outputs and 64 fast outputs (the derivative of the internal fast switching of the microcell) are directly fed to a 128-channel ASIC called EXYT ^[32] on back side of the SiPM board. The standard output channels are multiplexed by on-chip resistor networks to generate 3 analog outputs (E, X and Y) for energy and positions decoding. The fast output channels are used to generate signal T for timing information. Each fast output channel has a 5-bit current DAC for local threshold adjustment to reduce non-uniformity of the threshold of timing discriminators. The threshold for each fast channel is scanned in advance.

The pulses of E, X and Y are continually sampled by a free-running ADC chip with 80 MHz sample rate (12-bit AD9634, Analog Devices), and processed by programmable logic circuits in FPGA (Spartan-6, Xillinx). Energy is calculated in FPGA by a digital integration mode. At last, the data are transmitted to a PC through Gigabit Ethernet. Fig. 2 shows a printed circuit board with two ASIC chips and the readout electronics for the detector module.

Fig. 2.Full-screen View

The two ASIC chips for two 8×8 SiPM arrays (a) and readout electronics for the detector module (b).

2.3 Experiments

2.3.2 Timing performance evaluation

Fig.3 illustrates the experimental setup for timing performance evaluation with a single crystal coincidental to a detector module. The ²²Na source was placed close to the single crystal to irradiate all LORS from each crystal in the detector module. The other experimental conditions were the same as the flood source experiment.

Fig. 3.Full-screen View

Coincidence measurement setup for the detector module.

2.4 Data processing

A crystal position map (CPM) was generated from the flood histogram of ²²Na measurement by using an automatic segmentation program ^[34] we developed based on a stratified peak tracking method. In the program, the peaks of the top layer were identified first with the following procedures:

1) Creating a square image of the original flood histogram to increase the distinction between the top and bottom layers;

2) Generating a template image of the top layer by using a SVD(singular value decomposition)-based method;

3) Determining the initial peaks of the top layer using the X and Y dimensional projections of the template image; and

4) Locating the accurate peaks of the top layer by using a mean-shift algorithm.

Then, responses of the top layer generated using the located top peaks and a Gaussian kernel were subtracted from the original flood histogram. The peaks of the bottom layer were tracked using the similar procedures. At last, a CPM was created by classifying each pixel to its nearest peak and assigning it with the crystal index of that peak.

Energy values (E) were histogramed using their positions and the CPM to generate the energy spectra for both the ²²Na and ¹⁷⁶Lu intrinsic radiation experiments. The ¹⁷⁶Lu spectra were subtracted from the ²²Na spectra to generate calibrated the energy spectra. The 511keV peaks in the calibrated energy spectra were identified and fitted by Gaussian functions N(μ, σ). The energy resolution (η) of each crystal was obtained from the fitted result of η= 2.355 σ/μ.

For the timing performance study, a 450–650 keV energy window was applied to the single crystal and each crystal of the detector module. The coincidence times were histogramed for each crystal and then the coincidence resolving times (CRTs), i.e., full width half maximum (FWHM), were derived by fitting Gaussian functions to the histograms.

3.1 Crystal identification and segmentation

Fig. 4(a) shows the flood histogram of the ²²Na source. Almost all crystal responses are clearly distinguished except some at the corner. Fig. 4(b) shows the summed profile of the red box area in Fig 4(a). The average peak-to-valley ratio is 8, indicating that very small decoding error is expected for this detector. The valley counts in the summed profile are mostly contributed by the events with multiple Compton scatters in different crystals. These events cannot be distinguished in a flood histogram of a traditional PMT-based detector ^[35,36], however, due to the excellent decoding ability, the events with multiple Compton scatters in different crystals are separated from the events deposited full energy in a same crystal including photoelectric interactions and multiple Compton scatters. The reconstruction algorithm will be improved in the future by applying different response models for different types of events.

Fig. 4.Full-screen View

CPM with the ²²Na source placed on top of the detector module (a), the summed profile (b) of the red box area in the flood histogram, peaks of the top(c) and bottom(d) layers, and the segmentation result (e)of the crystal position map.

As shown in Figs.4(c) and 4(d), peaks of the top and bottom layers were exactly located. As shown in Fig.4(e), the flood histogram was segmented to 481(16×16+15×15) zones. Each zone has indices of layer, row and column. These indices were automatically generated in the procedures. Based on the segmentation results, a crystal position map was generated.

3.2 Energy resolution

Figs.5(a) and 5(b) show energy spectra of the ²²Na source and the intrinsic¹⁷⁶Lu in a typical crystal with, respectively. The calibrated energy spectrum is shown in Fig. 5(c). The energy resolutions and energy channel numbers of 511 keV peaks of the bottom layer are shown in Figs. 5(d) and 5(e). The average energy resolution (mean value of all crystals) and the average channel number of the 16×16 crystals are (14.0±1.6)% and 419±34, respectively. The energy resolutions and energy channel numbers of the 511 keV peaks of the top layer are shown in Figs. 5(f) and 5(g). The average energy resolution and the average channel number of the 15×15 crystals are (12.7±0.6)% and 417±26, respectively. The slightly better energy resolution of the top crystals is counterintuitive, as they are expected to have a higher light loss due to longer travel path to the SiPM. One possible reason is the depth variation of the scintillation emission sites in the top crystals is smaller.

Fig. 5Full-screen View

Energy spectra of the ²²Na source (a) and intrinsic ¹⁷⁶Lu (b) the calibrated energy spectra (c), energy resolutions (d) and channel numbers(e) of the bottom layer, and energy resolutions (f) and channel numbers (g) of the top layer.

3.4 Coincidence resolving time

Fig. 6 shows the coincidence resolving times for the single crystal with the top/bottom layer crystals of the detector module. The average CRTs at the top and bottom layer are 565±66 ps and 635±114 ps, respectively. Similar to the energy resolution performance, the top layer has better timing performance than the bottom layer. Meanwhile, the crystals at the edge of both layers have better timing performance, though their light collection is lower. It may be because that the optical photon collections in various depths of a crystal are different due to the divergent light propagation, which introduces jitter to time determination, and the crystals at the center have higher difference than the crystals at the edge.

Fig. 6Full-screen View

The coincidence resolving time for the single crystal with the top layer (a) and bottom layer (b) crystals.