| Summary: | 碩士 === 國立臺灣海洋大學 === 通訊與導航工程學系 === 99 === Currently, many brain lesions or diseases are diagnosed by observing the magnetic resonance (MR) images. Especially, the volume of gray matter and white matter tissue in brain is changed with different diseases. In contrast to the brain diseases diagnosed according the the physician’s knowledge and experience, the development of the brain tissue classification algorithm could provide a valuable medical help in prevention and treatment of diseases. However, the grayscale value of the various parts of brain tissue in MR images is close to and difficult to be separated effectively. The study is mainly on the segmentation of the gray and white matter tissue in brain by using MR images.
The thesis developed a two-stage classification of brain tissues by using multispectral image processing. The classification procedure in the first stage is non-supervised, which applies independent component
analysis (ICA) in enhancing the image features, and brain tissues are categorized through K-means clustering method. The classification results in the first stage show that the gray or white matter tissues within brain are recognized as the same class as other brain skull organization, respectively. In order the promote classification accuracy, it is essential to remove the brain skull organizations from MR image before the classification of gray or white matter tissues. In the second stage, a supervised classification is
proposed to further categorize the classes which recognized as gray or white matter from the classification results in the first stage.
Simulation results demonstrate that gray matter or white matter tissues within brain can be separated effectively form brain shell organizations. The proposed two-stage classification of brain tissues verified by the experimental data, can increase around 20% TI values of
accuracy in the gray matter and white matter tissue classification compared the those results using only the first stage in classification. Furthermore, the developed classification method in this research can provide auxiliary in medical diagnosis.
Keywords: MR image, independent component analysis, K-means.
|
|---|
