Representing Anatomical Knowledge: A Head & Neck Atlas of Multiple Coupled Structures

The large amount of acquired data - especially in case of radiological 3D data - renders the extraction of anatomical information from medical image data by means of manual delineation impracticable. There, computer-based automatic approaches can bring a benefit. A number of anatomical regions are hardly recognizable due to their low contrast to surrounding structures. However, radiologists are capable to annotate also those regions in the image data. It is a challenge for automatic algorithms to represent this anatomical knowledge of the clinical expert in a way that it can be used by a computer. For the head & neck region, we have developed such a model which contains a set of coupled anatomical structures. There, they are represented with regard to their shape and size but also to their relative position to each other. The model distinguishes between active and passive structures. The former represent regions which can be clearly distinguished from their surroundings. The latter can only be identified due to their relative position with respect to other organs. The novel modelling approach is more flexible than classical atlas-based segmentation techniques. Furthermore, it is based on a generic framework that can easily be adapted to other clinical applications beyond the head & neck region.