Real time depth of interaction determination based on Fourier Transform and Support Vector Machine

Abstract

The resolution of the PET scanners is limited by the parallax error due to the missing knowledge about Depth of Interaction (DoI) of the incident photons inside the scintillation crystal. In real time, a DoI calculator must fulfill a suitable event processing rate, which requires low complexity DoI algorithms. The DoI is obtained when the crystal within a phoswich detector is identified based on the shape of the scintillation light distribution. The main idea of this paper is to enhance the performance of crystal identification by using a Support Vector Machine (SVM) classifier and Discrete Fourier Transform (DFT) feature extractor. Besides, the paper introduces a complexity reduction method by merging the SVM classification phase and DFT in order to comply with the real-time rate. A real-time FPGA-based merged DFT-SVM DoI method has been implemented and validated to discriminate pulse-shapes of LSO-LuYAP scintillation crystals events. A correct assignment of 92.2% at rate of 6.2 Mevents/s is achieved for a sample of ∼100 000 pulses from LSO-LuYAP crystals read out using vacuum photomultipliers. Compared with recent DoI methods, the proposed method provides the highest performance and fulfills the required real-time rate of PET scanners