Automatic recognition of brain regions from magnetic resonance images

An application of active contour models to the problem of atlas to image matching for magnetic resonance (MR) brain images is demonstrated. A B-spline approximation to the original snake model of M. Kass et al. (1988) is used, along with a multigrid method to overcome some of the problems of the snake model. In the particular example shown, the multigrid method yields an accurate warping of the atlas curves onto the segmented image. Overall, the single-level snakes performed well; however, an instance in which the single-level algorithm could not correct a region mismatch is shown. The procedure can be extended to the 3-D problem by replacing the B-spline curve with a B-spline surface. The snake energy is also modified to reflect mechanical properties of the surface.<
>