Research And Application On Semantic Annotation Technologies Of Medical Images

The key technologies and algorithms of the medical image semantic annotation are thoroughly studied in this thesis. The semantic annotation and retrieval system of medical images is designed and achieved. It mines semantic information from the medical image content and its associated text, and can provide doctors with medical image semantic annotation and retrieval services. Working as one part of The National Natural Science Foundation-funded Project "Data Mining Technology Based on the Medical Images (60372072)", the main research work are as follows:(1) The medical image classification and annotation based on feature selection. This thesis proposes the feature selection method based on mutual information (MI) and its improved Mi-based greedy optimization algorithm; achieves the classification and annotation of the support vector machines (SVMs) based on MI feature selection, and finally establishes the mapping from low-level visual features to the high-level semantic features for medical image classification and semantic annotation ,which solves the problem of the content-based medical image classification included and the problem of the large-scale computing when all the characteristics participated in the classification. The experimental results show the system's good practical effects.(2) The medical image semantic annotation based on Hybrid Model. This article proposes the medical image semantic annotation based on Hybrid Model which synthesizes three methods: the SVMs of classification and annotation based on MI feature selection, the improved annotation method based on correlated vocabularies of the training set, and the CUMLS-based semantic mining method of medical images. The new model combines the advantages of each algorithm, studies and improves the technologies and algorithms of the semantic annotation of medical images. Furthermore, this thesis provides the improved realization algorithm of annotation method which is based on correlated vocabulary of the training set, and proposes the CUMLS-based Semantic mining method of medical images, which mines semantic information from medical image-associated text.(3) The visual feature extraction technology of medical image. This article proposes the improved median filtering algorithm ;studies the enhanced algorithm based on the rough set; and analyzes the feature extraction based on the canny operator,moment invariants and gray-level co-occurrence matrix. (4) Design and implementation of semantic annotation and retrieval model. It designs the framework and the structure of semantic annotation, forms the combination of the auto-annotation and the semantic retrieval of medical images; moreover, by researching the basic needs of the medical imaging data informationization construction in hospital, it completes the R&D of the semantic annotation and retrieval system of medical images...