Abstract
This paper describes an approach to establish the correspondence between a magnetic resonance (MR) image of the brain and a slice through a 3D anatomical model. The model is of voxel structure that symbolically labels primary tissue types such as grey matter, white matter, CSF, etc. In this approach a slice is first searched for in the model to achieve the best general match with the brain MR image in question. The operation involves a minimization of parameters such as position, rotation, slant, tilt and enlargement. Having thus found a globally good registration between the image and the model, local matches that link every pixel in the image through to the model slice are then searched for. This pixel-by-pixel match is expressed within a pair of maps, one for the vertical deformation and the other for the horizontal one. The matching algorithm consists of a series of octave separated blurring convolutions combined with exhaustive grey-valued correlation. Because every pixel in the model slice is labelled in terms of its tissue type, and because every pixel in the image has been matched directly to the model, every pixel in the image is now classified. This classification is used directly to perform segmentation which serves as a basis for the computation of medically relevant indices.
Key Words: