Neural network model and algorithm for real-time multicriterion image reconstruction from projections

Formulates the image reconstruction problem in terms of a multiobjective entropy optimization based neural network model and investigates its performance. The value of a neural network as a computational device is shown. The authors also demonstrate the efficiency of their implementation of neural net processing and its advantage over the traditional entropy optimization algorithm (MART) based on computer generated noisy projections. The method presented has the particular ability to solve severely ill-posed reconstruction problems. The algorithm described has been implemented on a serial computer but is ideally suited for parallel processing.