Electronics for monolithic scintillator PET detector modules based on neural network position estimators

Abstract

We are currently developing a prototype of monolithic scintillator PET detector module based on neural network position estimators. The detector module comprises of a 25.5mm×25.5mm×10mm LYSO crystal coupled to Hamamatsu 64 channels multi-anode PMT H7546B. Comparing with classical pixelated detectors, the prominent drawbacks of neural network based detector module are lower signal to noise ratio and more complicated signal readout scheme and data processing. The former needs low noise multi-channel readout front-end electronics with high dynamic range and high resolution signal digitization; meanwhile the latter can resort to modern FPGA technology. This paper describes our design and implementation of electronics for single block crystal PET detector modules which we are building. We choose to digitize 16 readout signals compressed from the 64 PMT channels after the simulating optimization of several possible signal readout geometries. Benefiting from the capacity and the flexibility of resource allocation of modern general class of FPGA, we implement all necessary digital nuclear signal processing techniques into a FPGA. Specially, the neural network position calculation is realized also in the FPGA. The validation and performance tests show that the system functions well with 15.3M events on-line processing throughput. The modular detector prototypes based on this electronics system can be a good platform for future physical experiments.