Uncertain reasoning based upon directional topographic features for medical image analysis

An uncertain reasoning system, based upon domain-specific knowledge and extracted directional topographic features (DTFs), is proposed to handle uncertainties and variations in the analysis of medical images. DTFs of a two-dimensional intensity image are the features, such as ridges and valleys, labeled from the underlying intensity surface in a chosen scanning direction. By using domain-specific knowledge, the system reasons about significant anatomic features based upon properties of, and spatial relationships among, the detected DTFs. These deduced anatomic features are then used to infer the location of the object of interest for other analyses. It is shown how to make use of partial and locally ambiguous information at different levels of abstraction to achieve reliable interpretation. Results of some applications are given to demonstrate the merits of the system