| With the help of technologies of computer image processing and analysis, computer graphics, virtual reality, computer networks and so on, medical image processing and analysis has improved the level of diagnosis and treat greatly, and has been paid much more attention to by people. It has become a research and application hotspot both at home and abroad, and formed a new crossover subject with rapid development.This thesis included contents of two main aspects: The first was the study of medical image processing software platform. I took part in the work of construction of integrated-platform for multi-modality medical image information computing and processing (IMICP) and design of 3DTool, which was a medical image processing and analysis software based on IMICP.The second was the research on image registration algorithms. New nonrigid image registration algorithms based on implicit shape representation and multi-resolution gird FFD with edge information fusion were proposed, which emphasized on innovation and their applicatins.IMICP was established on ITK and VTK, and integrated algorithms including filtering, segmentation and registration. IMICP implemented a medical image software platform with uniform interface, abundant algorithms, all-around functions, reasonable hierarchy and uniform style. It offered a good cooperating and communicating space for medical image researchers.3DTool, the medical image processing and analysis software based on IMICP, had clear and reasonable framework, abundant algorithms and convenient and friendly GUI. It offered some activity for users and supplied tremendous expandable functions by Plugin software structure. On the one hand, 3DTool provided for doctors a medical treatment and diagnose assistant tool which was intuitionistic and easy to use. On the other hand, it validated the feasibility of IMICP platform framework and design consideration, and provided an example for other developers.As for the research on registration algorithms, new nonrigid image registration algorithms based on implicit shape representation and multi-resolution gird FFD with edge information fusion were presented, which used a hierarchical model including both global and local transformations to recover the whole transform field and resolve the problem of deformation image registration. The global transform took advantages of both implicit shape representation and mutual information measurement, and implemented a global registration framework that was invariant to translation, rotation and scaling to align the extern contours of the images. Local registration under the base of global registration can enhance speed and avoid to trap in local extremum. In local registration, B-spline based FFD was chosen to modeling local deformations, and the method of multi-resolution grid was used to implement a coarse-to-fine hierarchical registration, which improved the optimizing speed and algorithmrobustness. We developed three models for different applications: (1) In registrations of mono-modality images, to take full advantage of edge information, we fused the images with their edge information before registration, which emphasized the distribution of image edge information in registration. (2) Edge registration measurement was imported to the MSD criterion to improve the image similarity measurement. (3)B-spline based multi-resolution grid FFD transform model and mutual information similarity measurement were combined to realize the local deformation registration for multi-modality images. Furthermore, an optional interactive registration was suggested for images with huge local deformations in some parts, which fits in with all the three models represented here.We realized and verified the algorithms described above, and got satisfactory experiment results. These algorithms were encapsulated as DLL and added to the Plugin depository of 3DTool. We make innovation in registration algorithms and at the same time implemented the engineering application value of these algorithms. |
PDF Full Text Request