The segmentation of CT images of the brain by fusion

Contouring of important anatomical structures within CT image sets of the brain is a difficult and time-consuming process. For stereotactic radiotherapy treatment planning and for the pre-planning of neurosurgical procedures, this is generally thought to be essential. Inaccuracies may result in irreparable mechanical and radiation induced damage during treatment. Precise delineation using manual techniques is often made difficult, however, by poor visibility of structures within the CT data set. In addition, the number of image planes and contours to be considered can prove prohibitively time-consuming to examine. In automated techniques, the identification of features without knowledge of their location and approximate dimensions is unsatisfactory and the fitting of parametric shapes for this purpose is limited by being structure specific. The aim of this paper is to describe a new approach to automated segmentation of intracranial features. Positional information for the various features is obtained by warping a ''template'' brain with the new CT data set. The geometric transformation can be implemented with respect to the relative position of the skulls. Image processing techniques can then be applied and the warp improved iteratively. The warping process leads to the features in the ''standard'' brain becoming identified with equivalent features in the brain of interest, so allowing them to be labelled. Examples are given for volumes of different complexity and further developments are suggested.