Support Vector Machine Applications In Medical Image Processing

Brain functional imaging is a way of directly and repeatedly detecting human brain which is a non-invasive measurement technique. In the research of advanced brain functional imaging, data processing is a key part. With the development of brain functional imaging technology, such as PET and fMRI, people can obtain abundance information which are able to use in locating brain function area, detect new function areas, research the corresponding relationship of brain function areas and so on. Therefore, the objective of the thesis is how to further expand the potential of these technologies, and effectively extract the information in the activities of brain function from functional magnetic resonance imaging data. It is of great significance in the research of the activities of brain function, and clinical diagnosis and treatment.The thesis is mainly concerned with the multi-variable and high dimensions characteristic of fMRI data, then using SVM to classify the brain function states availably, which can probe brain functional activities. The main works as follows:First, my work is to solve the problem about how to combine the fMRI data with SVM, which is programming to process fMRI data using SVM. It is proved that the validity of program by simulating data.Two different methods, PCA and temporal compression, are proposed to deal with the fMRI data which are both validly probing the brain area about visual activities and two-hand movement, for instance, the brain active areas of visual experiment are focus on Occipital Lobe and the brain active areas of two-hand movement experiment are focus on a part of the frontal cortex. They are helpful in conducting the clinic diagnosis and treatment. It is proved that SVM will be a new way to find out the different modes of brain function by the actual fMRI data's analysis.In summary, based on the processing of functional magnetic resonance signals, this thesis applies several methods to localize the brain area during visual activities and two-hand movement. At last the results from our simulation and processing real data are compatible with that from physiology and pathology, and demonstrate the effect and research value of those methods.