Development and initial results of a prototype brain PET scanner using a time-based digitizer and an FPGA-based real-time coincidence processor

Objectives: We are developing a brain-dedicated PET scanner that can be inserted into an ultra-high field (7T) MRI scanner. As the first step of this research, we developed a prototype brain PET scanner and an FPGA-based DAQ system. The brain PET scanner was designed to obtain both DOI and TOF information. The FPGA-based DAQ system consisting of two time-based digitizers and one real-time coincidence processor was developed to receive a large number of signals from the PET scanner and process the data at high count rates.

Methods: A brain PET scanner with a ring diameter of 254 mm and an axial FOV of 52 mm consisted of 14 sectors. Each sector supported up to 2×2 detector blocks, each consisting of a dual-layer crystal array and an 8x8 SiPM array. The dual-layer LSO crystal array consisted of 13×13 upper crystals and 14×14 lower crystals with the dimensions of 1.78×1.78×8 mm³ and 1.78×1.78×12 mm³, respectively. The outputs of the 8x8 SiPM array were multiplexed to yield four position-encoding signals and one timing signal for each detector block. The present brain PET scanner is equipped with a single block ring with an axial FOV of 26 mm. The signals of 14 sectors were input to two time-based digitizers. Each time-based digitizer consisted of 132 energy channels and 33 timing channels, and received signals from up to 33 detector blocks (8 sectors). The energy channel received the position-encoding energy signal using a charge-to-time converter (QTC) that linearly converted the energy into the pulse width. The pulse width was measured using a binary counter with a resolution of 625 ps implemented in a Kintex-7 FPGA (KC705, Xilinx). The timing channel was a time-to-digital converter (TDC) with a 10-ps resolution implemented in the same FPGA. The single event data including the position, energy, and timestamp were sent to the coincidence processor using a 2.5-Gbps transreceiver. The coincidence processor implemented in a Virtex-6 FPGA (ML605, Xilinx) compared the timestamps of single event data and generated prompt and delayed coincidence data. The coincidence data were sent to the DAQ computer using 1-Gbps Ethernet. In addition, the in-house clock distributor was developed and used to synchronize the clock signals in the DAQ system. A 2D Hoffman brain phantom filled with ¹⁸F-FDG was imaged using a developed brain PET scanner and compared with that obtained using a Siemens mCT scanner. The phantom image was reconstructed using the 3D-OSEM with 4 subsets and 32 iterations. The normalization, attenuation correction, and random correction from the delayed coincidence data generated by the coincidence processor were applied. Results: Almost all crystals in both layers were clearly resolved in the flood histograms. The energy resolutions of the upper and lower crystals were 10.7±0.6% and 11.1±1.5%, respectively. In addition, there was no data loss at a high count rate of 8 Mcps. The developed brain PET scanner represented the more detailed brain structures than mCT with PSF and TOF information. Conclusion: We demonstrated the high-resolution phantom image obtained using the prototype brain PET scanner and the FPGA-based DAQ system. The TOF and PSF information will be measured and incorporated in the future studies.