Brain structures deformation to obtain anatomical averages

Anatomic and functional atlases are very useful tools in stereotactic neurosurgery planning. This work presents a first approximation to the construction of an anatomical human brain atlas, which can be considered as a representative model of the local population. The atlas is built up from MRI stacks, obtained with a 1.5 T imager which is able to produce a 1.5 mm resolution. The proposed approach allows the authors to obtain an anatomical average, keeping the different brain structures. The anatomical deformations are generated by a modified version of the non-linear Marquardt-Levenberg optimization method and a bidimensional affine transformation using four parameters. The process is carried out in a local way by subdividing the image in regions to provide anatomical deformations using a rigid transformation. It is assumed that the brain structures have been previously segmented and rigidly registered. In order to reduce the discontinuities appearing when reconstructing the resulting image, a cubic B-spline interpolation was used. Preliminary results show a promising method to estimate an anatomical average of brain structures in a bidimensional way.