Researches On Image Registration Based On Mutual Information

Medical imaging has provided more and more kinds of medical images as the development of biomedical engineering and computer technology, such as X-ray, MRI(Magnetic Resonance Imaging), fMRI(functional MRI), SPECT(Single Photon Emission Computed Tomography), PET(Positron Emission Tomography), DSA(Digital Subtraction Artery), and so on. Different imaging modalities could provide different information of the human tissue and organ. For example, CT has high resolution and of high value in locating the point of the pathological changes, while MRI is good at soft tissue imaging which is good for locating the range of pathological changes. PET and SPECT has low resolution, however they could the functional and metabolic information of organ. Therefore, it is desirous strongly from clinical doctors to integrate the different information form different images, which could serve for clinical diagnosis better. Nevertheless, images of diverse modalities have dissimilar imaging theories, different resolution and different imaging parameter. As a result of that correction of image location is a pre-requisite step before image fusion. Therefore, the subject of "medical image registration" is brought in. This paper has done much research on the core technology and algorithms on image registration, as well as the evaluation and validation of registration results. The work in this paper is as follows.(1) The local features of region of interests (ROI) such as bone, has been introduced into mutual information based registration as the step of "sparse match" , whose transform matrix could be calculated directly. The step of bringing the features could decrease the computation cost of mutual information for registration, increase the time efficiency and avoid the mismatch to some extent.(2) An objective validation method based on distance has been proposed, and five forms of error have been given out.At last, an image registration system including the registration validation algorithm is implemented.