Abstract
One of the most serious errors in PET systems is the parallax error, which occurs due to spatial resolution limitation in PET. This error can be solved by using multilayer scintillation detector (phoswish detector) where each layer has a different decay constant. Each phoswish detector is optically coupled to a single PMT. Hence, by applying pulse shape discrimination (PSD) methods to identify the scintillated layer, the parallax error can be reduced. This paper proposes a PSD method relied on a Fractional Discrete Cosine Transform (FrDCT) and a Support Vector Machine (SVM) classifier. The proposed method can be used to differentiate between two arbitrary types of scintillation pulses where each type has a different decay constant. For demonstration of its performance, the proposed FrDCT-based method is applied to a data set of scintillation pulses of LSO and LuYAP crystals. Furthermore, FrDCT with different factors as well as different kernels of SVM; such as Linear, Quadratic, and Radial Basis Function (RBF) kernels, are used to study their effects on the discrimination efficiency. The highest achieved efficiency is equal to 95.9% for the 0.9 factor of FrDCT using a quadratic kernel SVM. On the other hand, the linear kernel consumes the lowest execution time at the expense of slightly reducing the discrimination efficiency.