Applying SOFM and Its FPGA Implementation on Event Processing of PET Block Detector

This chapter introduces an application case of self-organizing feature map (SOFM) on event processing of Positron Emission Tomagraph (PET) detector. The Siemens Inveon Dedicated PET scanner uses an array of 64 20x20 segmented LSO detectors in a tomographic arrangement to allow PET imaging of small animals. A detector consists of 20x20 LSO crystal pixels coupled to a Position Sensitive Photomultiplier Tube (PSPMT) through tapered light guide (Fig. 1 (a)). The 511 kev gamma ray enters LSO pixel, deposits its energy and emits visible light through scintillation process. The light is guided through light guide and detected by PSPMT. The position of scintillation (X, Y) is calculated by Anger logic. The flood image (position profile) of an Inveon PET detector is shown in Fig. 1 (b). The white clouds indicate the accumulation of scintillation events. The matching and calibration of the system electronics to the detector arrays involves three stages. The first stage identifies the individual crystal elements (pixel) from the raw X and Y values. That is to build a mapping relation between the X, Y value to the crystal index. The current implementation of crystal identification uses a hill-climbing algorithm to find the exact position of every crystal by locating the peaks in position profile. Crystal Lookup Tables (CLTs) are generated as the results of the crystal identification process. Although this method achieves around 95% accuracy, it involves intense human interactions for crystal look-up table corrections. The complex algorithm can only be implemented on a PC, which makes the crystal identification an off-line procedure.