Research On Multi-modal Medical Image Registration And Fusion

With the continual development of modern medical imaging and radiology and constantly emerging of a new generation of imaging equipment, such as CT, MRI, PET, SPECT, the image processing and analysis techniques based on computer technology play an increasingly important role. Especially the registration and fusion of the multi-modal medical images has become one hot spot. In clinical applications, to improve medical diagnosis and treatment, usually by analyzing comprehensively the results of the same patient's multiple modality images, with more comprehensive access to information about various forms and functions of human body. But because of orientation differences, different imaging parameter settings (resolution, contrast), etc. when imaging the human body, it's very difficult for a doctor to align many images just by imagining. Therefore, searching for the best geometric transformation method of space between different images, to make them accordance both in spatial location and anatomical structure, then fusing images' information that after registration in a new image and displaying it, eventually anatomical and functional information can combine to complete the fusion. This provides an accurate basis for the timely and efficient diagnosis of the disease or an appropriate treatment plan.This paper outlines the multi-modal medical image registration and fusion from theory, methods and other aspects, focusing on the mutual information-based multi-modality medical image registration of the registration methods based on gray-scale. For the CT, MR, PET image data of head, it gives good image fusion results of brain CT-MR, PET-CT, PET-MR. The first step is to extract DICOM format image information, and set the parameters of the initial registration. Then with the mutual information as similarity measure using the optimized algorithm to search the registration parameters obtained when the mutual information is maximum. Finally the best registration is made. The optimal iterative search process includes the selection of the parameter optimization algorithm, the gray interpolation calculation during the process of coordinate transformation. With less manual intervention the method doesn't need image feature point extraction and pre-processing of the classification of organizations and only rely on the image information.In contrast with the traditional mutual information image registration algorithm, this paper increases two rough and fast registration algorithms prior to the detailed mutual information image registration. One is multi-level resolution for the registration, which makes a coarse registration for images under the low-resolution giving the corresponding registration parameters, and then a fine registration follows. The other is the compression of gray-scale images including linear compression with worse registration results and classification according to histogram peak, which uses the image histogram information to divide gray space into several parts, and then rough registration, with the result as the initial conditions of mutual information registration algorithm. Except linear compression, the other two algorithms can reduce the time of mutual information registration algorithm and avoid the local minimum optimization algorithms arise.Finally the pixel-based image fusion is applied. And the registration algorithm is evaluated.