Towards interactive vision edge-region integration in MRI

It is pragmatic to look for ways in which the knowledge of an expert can interact with data driven processes and such high level knowledge as can be represented usefully in a system. Central problems such as image segmentation do not go away but systems can be judged somewhat differently, not as successes or failures, but on the basis of the amount of work that they save a human expert. This argument is illustrated by two projects: (1) 3D reconstruction from biplanar angiograms; and (2) Segmentation of cranial MR images.<
>