Reconstruction of Compton-camera images using artificial neural networks

We report on a new and fast approach for the three-dimensional reconstruction of X-ray source distributions monitored by a Compton-camera consisting of two layers of pixel-detectors. We use multi-layer feedforward artificial neural networks trained by a standard backpropagation algorithm to determine the probability that a given Compton scattered photon originated from a certain point in the reconstruction space. Summing up this probability for all measured photons at each volume element (voxel) of the discretized reconstruction space gives a good estimate of the real X-ray source distribution. We reconstruct and discuss the point spread function obtained with different materials (germanium, silicon) used for the scatter detector and incident photons having an energy of 122 keV and 512 keV.