High performance medical image processing in client/server-environments

As 3D scanning devices like computer tomography (CT) or magnetic resonance imaging (MRI) become more widespread, there is also an increasing need for powerful computers that can handle the enormous amounts of data with accept able response limes. We describe an approach to parallelize some of the mor e frequently used image processing operators on distributed memory architec tures. It is desirable to make such specialized machines accessible on a ne twork, in order to save costs by sharing resources. We present a client/ser ver approach that is specifically tailored to the interactive work with vol ume data. Our image processing server implements a volume visualization met hod that allows the user to assess the segmentation of anatomical structure s. We can enhance the presentation by combining the volume visualizations o n a viewing station with additional graphical elements, which can be manipu lated in real-time. The methods presented were verified on two applications for different domains. (C) 1999 Elsevier Science Ireland Ltd. All rights r eserved.