Anatomical object recognition using deformable geometric models

摘要

This work addresses the problem of identifying specific objects within three-dimensional data sets. In the specific example chosen we are trying to locate part of the brainstem, and associated structures of the upper spinal cord and mesencephalon. and determine its size, shape and orientation. The approach is in two parts: firstly, to use a controlled, deformable model, based on superquadric geometric primitives as an initial estimate and apply genetic algorithms as the technique for solving the complex optimization problem of defining an approximate encompassing envelope within which the object will be found. The second step implements a segmentation technique, based on image features, to refine the tentative object into a more complex, and realistic, shape suitable for subsequent visualization or volumetric measurement.