Integrated skeleton and boundary shape representation for medical image interpretation

Abstract

We propose a method of extracting and describing the shape of features from medical images which provides both a skeleton and boundary representation. This method does not require complete closed boundaries nor regularly sampled edge points. Lines between edge points are connected into boundary sections using a measure of proximity. Alternatively, or in addition, known connectivity between points (such as that available from traditional edge detectors) can be incorporated if known. The resultant descriptions are objectcentred and hierarchical in nature with an unambiguous mapping between skeleton and boundary sections.