A Real-Time Scintillation Crystal Identification Method and Its FPGA Implementation

Abstract

At the expense of increasing the complexity, Zernike Moment (ZM) was used in Crystal Identification (CI) to solve the parallax error with the highest identification performance among recent methods that are based on Digital signal processing (DSP) algorithms. In Positron Emission Tomography (PET) scanners, the CI method must fulfill the real-time event rate processing that requires low complexity and fast algorithms. The main idea of this paper is the complexity reduction of the ZM-based CI method and using a Support Vector Machine (SVM) classifier to comply with the rate of real-time PET scanners. To achieve this goal, 1-D form of ZM and fast computations of cascaded ZM and SVM stages are proposed to minimize the complexity while preserving the highest performance among compared methods. The proposed fast ZM-SVM CI method is implemented using both MATLAB and FPGA and verified by identification of ~ 100 000 pulses of LSO-LuYAP scintillation crystals. The FPGA realization of the fast ZM-SVM CI method offers an identification rate of 90.6% at a rate of 6.2 Mevents/s using 16 samples per event which is sufficient for real time small animal PET scanners.