Compton sequence estimation based on the deep learning method

Abstract

Determining the sequence of Compton scattering and photoelectric absorption events for a Compton camera system through timing information is difficult due to the finite timing resolution of radiation detectors. The conventional method compares the energies of two sequential events and determines the order of these events. The deep learning method can estimate the sequence of Compton scattering followed by the photoelectric effect better than the conventional method because it determines the sequence based on both energy and positional information of the radiation interaction. The initial information of the deep learning models is the position and energy information, and the input data are then processed in the nodes of the hidden layers. In this study, the performance of deep learning models for Compton sequence estimation and the effect of position information on these methods were investigated. The accuracies of the deep learning method and the conventional comparison method were compared. The weights connecting each node were analyzed to evaluate the effects of position and energy information to determine the Compton sequence.