Medical Image Retrieval

With the development of multimodality medical imaging equipments, the medical images have become indispensable to modern clinical diagnosis and medical research. A great number of images are generated in a hospital everyday. It becomes an urgent problem to be solved how to organize, manage and index medical images in so large scale. The traditional information retrieval techniques, which are based on text comparison, are not satisfied to retrieve large scale medical image databases. New image retrieval methods should be explored. It is a very promising idea to introduce content-based image retrieval (CBIR) technique into indexing medical image databases, so we studied the algorithms and techniques of content-based medical image retrieval in the thesis.One of the key points of the technique of CBIR is to select and extract proper features to describe the information of images. Based on the fully review of general feature-extraction methods, combining the imaging mechanism and special characters of medical images, we focused mainly on the feature-extraction algorithms suitable for medical images. Many features, such as statistical values of gray levels, texture features (including gray co-occurrence matrix, gray-primitive co-occurrence matrix), shape features (focus on texture features extracting from image of edge direction angle) and the features extract from feature points of a medical image. Every result of features above were shown. And we discussed the effectivity of the results by using the precision percentage and recall percentage.One kind feature can not show the image totally. So we discussed severalretrieval methods, such as retrieving by combining the features above and two grade retrieval. Then we discussed the method of determining the retrieval features by the retrieval image. To decrease the gap between high level concepts in human minds and low level features computed by computers, relevance feedback technique is performed in CBIR systems according to users'feedback, which tries to establish the link between high-level concepts and low-level features. We proposed a new formula to get the optimal query, which is based on the relevance feedback technique and the positions of the retrieved images in the result set are considered. It accelerates the convergence speed.