Abstract:
One important goal in neuroscience is to find a correlation between the function
of the brain and the localisation of the involved brain regions, especially on
the surface of the brain. Today, the activation of the brain can be visualized.
Nevertheless statistical evaluation works only when analyzing large groups of
patients or volunteers.
This paper is about computer aided, inter-individual analysis of human brain's
surface. The problem can be generalized onto the problem of classifying convoluted
surfaces of the same class. First, techniques for automatic segmentation of the
cerebrum out of neurologic MR data was developed. This was necessary to get
the large number of surfaces. A completely new kind of visualization of the
cortex` surface through the use of depth maps was developed. These maps show
the distance to a surrounding sphere, defined by a bezier area wrapped around
the surface. This representation makes the brains surface accessible for manual
or automatic classification. In the next step a modified method for hierarchical
computing of motion fields was developed. These motion fields result out of a
pixel-wise correspondence between the different surfaces, and the computation
leaded to the classification of the structures. Finally, this classification has
been prototyped. This was done by computing new, previously unknown brain
surfaces out of linear combinations of a limited number, randomly chosen
protoypes of surfaces.
Through these steps a method was developed, gathering the cortex of the human
brain automatically for the first time and to compute pixel-wise
correspondences between the brain surfaces of different individuals. The
resolution using this technique is significantly better than the widely used
Talairach grid system, which is based upon global transformations.