Image based mammographie ontology learning

Understanding the content of an image is one of the challenges in the image processing field. Recently, the Content Based Image Retrieval (CBIR) and especially Semantic Content Based Image Retrieval (SCBIR) are the main goal of many research works. In medical field, understanding the content of an image is very helpful in the automatic decision making. In fact, analyzing the semantic information in an image support can assist the doctor to make the adequate diagnosis. This paper presents a new method for mammographic ontology learning from a set of mammographic images. The approach is based on four main modules: (1) the mammography segmentation, (2) the features extraction (3) the local ontology modeling and (4) the global ontology construction basing on merging the local ones. The first module allows detecting the pathological regions in the represented breast. The second module consists on extracting the most important features from the pathological zones. The third module allows modeling a local ontology by representing the pertinent entities (conceptual entities) as well as their correspondent features (shape, size, form, etc.) discovered in the previous step. The last module consists on merging the local ontologies extracted from a set of mammographies in order to obtain a global and exhaustive one. Our approach attempts to fully describe the semantic content of mammographic images in order to perform the domain knowledge modeling.