The Study Of Multi Modality Medical Image Registration

With the development of medical imaging physics, different forms of imaging devices produced multi-modality medical images, which can provide varieties of medical information such as anatomical and functional information. Each has its advantages as well as shortcomings. If these information from different resources are integrated by some way, so they can provide more detailed information for medical diagnosis and researching the function and structure of human being.Medical image registration has been widely applied to clinical research. This paper discusses image registration technique in details. Based on some similarity measure rule between images, we registered multi-modality medical images involve MR, CT, PET. At the beginning, we simply narrated the basic medical imaging physics and the criterion of Digital Imaging and Communications in Medicine. Further discussed the method of medical image registration, and analyzed its detail process which including transformation, interpolator, similarity measure rule, optimizer, multi-resolution. Implement the registration program with Powell optimizer and mutual information metric. Analyzed the transform model and assign different weight to each component of it. Compared the performance of some interpolation algorithm in common used. Make choice of bilinear interpolation algorithm as our interpolator. Powell optimization algorithm needn't calculate the derivative of target function, effectively eliminates the influence of local extreme. Via aligning the centroid and principal axis of two images to pre-registration, we have increased the accuracy and speed of registration. Use only portion information of images, can largely speed up registration process, but it maybe lose the accuracy.Undeniable, some shortage existed in our paper yet. Firstly, we do the registration in the logical space, predigest the registration modal. Secondly, rigid transformation has not considered the elastic distortion of human body. Lastly, we couldn't estimate the registration accuracy perfectly.