Implementation of Self-organizing Map Based Positioning Scheme on FPGA

For continuous crystal based PET detector, we develop a Self-organizing Map (SOM) Neural Network Based Positioning Scheme which can achieve 2.07mm average resolution and is feasible for implementation on Field Programmable Gate Array (FPGA). In this paper, we propose the FPGA design of SOM scheme and apply it to our experiment data. Taking advantage of the pipelined and parallel structure, the implementation of this algorithm is able to process 5M events per second with system clock running at 200M Hz. The test results show that the FPGA solution has almost the equal performance with software platform. Considering the potentiality of DOI determination using SOM scheme, it is promising to realize real-time 3D position estimation in the future.