Maximization Of Mutual Information Based Medical Image Registration

The purposes of this paper are to explore the Medical Image Registration using the combined methods of computer analysis and digital image processing, and to improve the performance of original registration algorithms based on Maximum of Mutual Information (MMI) by proposing new methods, for the sake of assisting physicians' diagnosis.Owing to its property of applying multi-modality imaging information into the clinical usage has the Medical Image Registration been the research focus. According to different emphases, the Medical Image Registration can be divided into different kinds, and owns many algorithms to realize it, such as Point Method, Face Method, Correlation Method, Moment and Principal Axis Method, Atlas Method, Fourier Method etc. In all of these, the registration method of MMI has been appreciated by researchers as its advantages, such as high accuracy, advanced stability, non-preprocessing and auto-registration. The standard MMI method makes use of the statistical property of image voxels and pixels and arrives perfect results of registration, up to sub-voxel level. However, it has some drawbacks and the performance of registration will be deteriorated, especially in conditions when the images lack mutual information, when the gray level relationship is not clear after images have spatial transformation, or when the sampling number is too small to judge the mutual information. Scholars have proposed many new methods to ameliorate it in order to counter its drawbacks. The combined method of MMI and Edge Correlative Deviation as well as the combined method of MMI and Mutual Edge Distance, two new methods given by this paper, bears many good qualifications in aspects of countering drastic noise and the lack of intensity values and having more accurate parameter curves as well as strong robustness and accuracy, because they take advantage of the statistical property of mutual information and the relationship of voxels' positions.After analyzing the standard MMI, This paper introduces the principles, methods, divisions and applications of medical image registration, and proposes two improved methods—The combined method of MMI and Edge Correlative Deviation as well as the combined method of MMI and Mutual Edge Distance. This paper gives thorough analysis of the two methods' principle...