Medical Image Registration Method Based On Mutual Information

Medical imaging has aided clinical diagnosis and many clinical applications by providing multi-modality medical images. As different imaging methods provide relatively complementary information of the same anatomy location of the patient, it is commonly desirable to combine the information provided by different modality images. The aim of medical image registration is to normalize images from different imaging time or different modality into a common coordinate, and further to make the individual information represent in the same coordinate for direct comparison and analysis.This thesis mainly discusses the registration method based on mutual information, which includes both the basic theory and implementation of this method. The application of this method is explained in two projects: the first one is the registration of plant slice sequence, and the other one is the validation of automatic diagnosis results of cerebral ischemic by registration. Finally the evaluation methods of registration results are discussed and two of them is applied to the results of our experiment.It mainly discusses:(1) The basic theory of mutual information, consisting of its definition, the meaning as a similarity measure, its calculation details and the effect of different intensity scale to it. The spatial transformation and interpolation is also discussed, and the refined Powell optimization algorithm is explained by its theory and implementation steps;(2) Application to the registration of plant slices sequence with its registration result and 3-D reconstruction result. Application to multi-modality cerebral ischemic patients, and their spatial normalization by registration to evaluate the results of automatic diagnosis. The results is compared with the method based on the correlation ratio with analysis;(3) The major evaluation methods for registration results, and introduce two evaluation values: slice-slice overlay segmented region ratio and volume-volume overlay segmented region ratio. Comparison and analysis of these two evaluation values is explained.