Abstract

We propose in this paper to deal with the notion of medical image registration: both physicians and surgeons need in the patient room or in the operation theater images of the medical target to reach and treat; these images can be very precise and come from a pre-operative acquisition device (NMR, CT-scanner,…) or from a per-operative set of sensors (active or passive infra-red sensors, laser acquisition, numerical 2D X rays, 3D echography,…). The main problem is then how to compare these multi-modal images having different resolutions and extracting different features (anatomic or functional) from the patient medical reality ? In a first Section, we present the hard side with essentially optical sensing methods ; after, in a second Section, we present an example of matching algorithm built in order to compare and super-impose pre- and per-operative images. This algorithm computes a confinement tree thanks to the watershed lines and we explain under which conditions the confinement tree can be an invariant parameter of the image for some topological transformations. We propose a simplification algorithm for this tree. Then, for two topologically closed images, for which the simplified trees are the same, we show how to match it, and how it is possible to consider those two matched trees as an initial condition for matching the remaining nodes of the initial trees. Then, a spline interpolation can be used in order to obtain the continuous transformation from the first image to the second one, using very few hypotheses. Finally, we describe several medical actions or surgical operations for which a physician or a surgeon needs such registration tools.

© 2000 SPIE