Event-by-event random and scatter estimator based on support vector machine using multi-anode outputs

In PET, the incident angle of gamma rays is estimated from coincidence information, but coincidence events are contaminated with random and scatter components. The mean contribution to the image from these components can be measured or estimated, but the noise resulting from the statistical variations in the detected events still remains and decreases noise equivalent count rates (NECR). Theoretically, incident angle to detectors or other related information can be used to discriminate random and scatter events from true events for increasing the NECR. This information can be delineated from spatial distributions of deposited energies on multi-anode photo multiplier tubes (PMTs), which arise from inter-crystal scattering and vary with the coincidence event type (true or random/scatter). In this work, a novel method for random and scatter subtraction has been developed using the support vector machine (SVM), a recently developed and widely used statistical pattern recognition scheme. SVM input data is a pair of spatial distributions of 256 outputs of multi-anode PMTs from coincidence detectors. The SVM is trained by coincidence events generated from a detector simulator using Monte Carlo calculation. True and random coincidence events are simulated and the results show that the proposed method is applicable for event-by-event estimation of random coincidence.