| Summary: | 碩士 === 國立中正大學 === 電機工程研究所 === 89 === In this thesis, we are aimed at the disease hydrocephalus and atrophy to build a content-based retrieval system of brain CT medical image database. Our system is divided into two schemes : the image segmentation and feature extraction, and the database indexing and searching schemes. In image segmentation, the graylevel distance transform(GDT) region growing, gradient vector flow(GVF) region growing and Otsu thresholding plus GVF snake are applied to the lateral ventricular boundary segmentation. The ventricular features are then extracted from the ventricular boundary. In database indexing, we utilize the samples of ventricular feature vectors to build a simulative testing dataset. Then the Linde, Buzo, and Gray (LBG) algorithm is used to cluster the feature vectors to build the database indexing structure. In database searching, we propose a fast matching algorithm and use the Euclidean distance to measure the similarity. The query result is the 10 most similar feature vectors in the testing database.
In accordance with the content-based retrieval system of brain CT medical image database, the accurate ventricular features will assist the diagnosis of physician. In the database system, the similar pathological CT images will be obtained. The physician can utilize these similar images to get some advisable information. Finally, the database system can be used to the education of medicine. The students will be guided to learn about the similar pathology.
|
|---|
