A neural network system for menisci surface estimation in MR discrete 3D scenes

Menisci pathologies are more easily assessed if images represented on planes parallel to the menisci major section are examined. In the case of magnetic resonance imaging (MR), this procedure requires the physician to locate the plane where the two menisci approximately lie. Unfortunately, this operation is time-consuming and it is often accomplished by successive refinements. In this paper we propose a system for the automated location of such plane. The system works on sets of MR sagittal tomograms and includes three modules, two of which, those performing the image understanding tasks, are based on neural networks. Satisfactory results have been obtained with short computation time.