Real time Method based on SVM for Depth-of-Interaction Identification in PET Scanners

Positron Emission Tomography (PET) scanners are based on a ring of scintillation detectors where each detector is comprised of an array of crystals usually coupled to a photo sensor. The parallax error is the common error that degrades the spatial resolution of the reconstructed image in the PET scanners. This error resulted from the mis identification of the exact depth-of-interaction (DOI). Phoswich based detector has been used as a solution to accurately identifying the DOI. The pulses emitted from the scintillation crystal layer of this detector are discriminated based on the Pulse Shape Discrimination (PSD) method. In this paper, a high rate merged PSD method which depends on only half-length of the scintillation pulse is proposed to enhance the performance of PET scanners. The frequency transform (discrete cosine transform or discrete sine transform) is merged with the support vector machine (SVM) classifier. The PSD method is verified by discriminating 100 000 scintillation pulses of both LSO and LuYAP crystal. The event rate was 24.5 M events/sec. Therefore, the proposed method realizes a high real-time event rate which nominates it for a Clear PET small animal PET scanner. The discrimination efficiency is kept as 91.36% by merging DST4 with the linear kernel of SVM