Training of artificial neural network for tomographic reconstruction of time varying object

This paper presents a study investigating the potential of neural networks for the reconstruction of X-ray computer tomographic images of time-varying object. To obtain a good image of a time-varying object without motion artifact one requires a large number of consistent projections equally spaced in angle. A set of projections are consistent if all the projections relate to the same X-ray absorptivity distribution. This requires that all projections are measured during the same phase of the time-varying object. We introduce a new image reconstruction method based on a priori knowledge of the projections to achieve this objective. This method is implemented on a novel neural network and a new training algorithm is proposed. Computer simulation shows that this training algorithm is valid and minimizes the reconstruction mean square error