Research Of Bone Registration Based On MR Image

Detect bone lesions magnetic resonance images, as well as their morphologicaldescription, has become a widely used by bone diagnosis and treatment. Testing thebone and calibration, contrast different periods of bone body position changes, youcan directly diagnose the severity of bone, and so be related to treatment, expandthe assessment of treatment effects.In order to achieve the communication among different imaging devices, thecross-regional, multi-modal medical images, the images generally use the DICOMstandard files to be stored. By analyze medical image and the process of DICOMfiles, based on the traditional PDE methods and evolved, use the C-V modelsegmentation method to extract the pig hind magnetic resonance image, and thenextracted from the different periods of NMR resonance image by registrationresearch.The traditional iterative closest point method, existence of local convergence,convergence is slow and poor, robustness and iterative end threshold difficult to set,based on the traditional ICP and on the basis of the above problems, gives acorresponding improvement: initialization, based on the principal axis, theregistration combined with feature points initialize method; convergence rate, usek-D tree to search, greatly improving efficiency; robustness, with the truncated-point method, the threshold point of law, will be excluded; iterative thresholdsetting, given a set threshold, the iterative method of termination. Finally, thismethod is summarized as improved iterative closest point algorithm.Registration process in this thesis is divided into a coarse registration andaccurate registration of two steps, first principal axis registration combined withthe feature point method to initialize the condition, and then on the basis of therough registration using improved iteration point algorithm to accurate register, usethe known simulated data and leg magnetic resonance image registration methodvalidation and analyze the results.