Based On Demons Algorithms2D、3D Multi-modal Medical Image Non-rigid Registration Research

Along with the continues development of medical imaging technology and information processing technology,the medical imaging offers a variety of images of the model. Multimodal medical image non-rigid registration, which can obtain a different pattern of rich information, adapt to the different soft tissues and organs, has very important clinical application value. This paper mainly studies the Demons algorithm based2D and3D multimodal medical image non-rigid registration.Firstly, introduce the significance of medical image registration and purpose, and its development history. Introduced the classification and methods of medical image non-rigid registration, and expounds the definition and framework of non-rigid registration.Secondly, discuss non-rigid registration of pretreatment, image interpolation, the feature space and the search space, multi-parameter optimization of basic principle and advantages and disadvantages of various methods, and the non-rigid registration of subjective and objective assessment methods in detail.Thirdly,put forward a kind of local structure tensor of conservation and Demons algorithm with the combination of improved algorithm, using the local structure tensor is not affected by light, keep moving direction information and image morphological information,to construct a new energy function dealing with multimodal registration, make up for the original Demons algorithm based on gray level assumes that the cause of conservation is not suitable for multimodal image processing and image gradient information missing, etc, through the array of different image experiment proves that this algorithm for two-dimensional multimodal medical image non-rigid registration accuracy and feasibility.Finally,combining the local structure tensor and Demons algorithm of3D data of the expanding of improved algorithm, through pretreatment, interpolation, gray-scale compression processing, solve the problem of complex3D data, and optimize the energy function based on the L-BFGS algorithm, multi-scale refinement of layered thought and solution to3D data registration. Experiment shows that algorithm can handle more than three-dimensional modal registration and large deformation problems.