Temporal-Spatial Consistent Registration On Four Dimensional Lung CT Image Based On Tracking And Optimizing The Trajectories Of Landmarks

With the rapid development of medical image technology, not only static twodimensional and three-dimensional image, but also dynamical four-dimensional image that includes temporal dimension can be obtained. Since containing more useful information, 4D image has been attached more attention by researchers in recent years. Meanwhile, 4D medical image registration has been at the leading edge of medical image research. Because 4D medical image registration is an important part of 4D medical image research. Image registration has many uses in medicine, such as the detection of disease development and treatment, the location of operation focus, and radiotherapy. Landmark-based image registration is the main branch of image registration, which owns advantages of lower computational cost and time consuming over other image registration. We mainly focus on research landmark-based image registration in this paper.Firstly, spatial information enhancement structure tensor(SEST) based object tracking algorithm is proposed to address the issue that structure tensor can’t analyze vessel structure in lung CT image in this paper. SEST is improved on the basis of original structure tensor object tracking algorithm, which introduces the analysis ability of structure tensor and utilizes the prior knowledge of lung respiratory motion to shrink search area. Some experiments verify the tracking accuracy and robustness of SEST.Secondly, to address the issue of the majority of landmarks inaccurate tracking and several landmarks obvious deviation, the temporal-spatial constraint optimization model is proposed on the basis of the researching work of our group. The temporal-spatial constraint optimization model restricts transformation to keep smooth in time and space. Besides, in order to reduce computation, the cost function is solved in local area instead of whole image. The efficiency and feasibility of the above model is validated by some experiments.Lastly, temporal-spatial consistent optimization model is proposed in the paper. It adds consistent constraint based on the above model to solve the issue of big consistent error existing in temporal-spatial constraint optimization registration results. In order to achieve more reasonable registration results, a multi-hierarchical registration strategy is employed. This registration strategy hierarchically appends the landmarks of big local consistent error into the initial landmarks set to launch a new registration until convergence. Better registration results can be achieved in some experiments.