ROI Segmentation And Visualization Methods For Medical Serial Images

The study focused on segmentation of serial medical images and their 3D visualization methods.Medical image segmentation is a challenging problem ,yet there is not a perfect and universal approach to solve this problem. In this thesis we proposed a new approach that regard medical image ROI region segmentation as non-linear separation of multi-dimension vector spaces. Try to find an effective interface to separate the vectors composed of the image element multi-dimension features, and then to separate the image elements which belong to ROI region from those which belong to non-ROI region. In order to find such interface, we proposed a novel approach based on wavelet decomposition and hybrid artificial neural network.. In the approach, we firstly extract multi-dimension features of each image element by using wavelet decomposition, and use them as the input vectors for the neural network, then we use self-organizing neural network to optimize the input vectors. Finally a modified BP network is used to learn from the optimized vectors and the interface is obtained which can be used to separate the ROI regions from the image background. The experimental tests of medical images indicated many advantages of this approach , such as its strong ability of extensive fitness and its effectiveness of segmentation. 3D rereconstruction results of the segmented serial images are also presented , which showed the effectiveness of the proposed approach.In microsurgery of peripheral nerve recovery, the knowledge of the spatial location ,distribution and adjacency relation of the fascicular groups in the nerve trunk is of great importance. Correct joining of the nerve fascicular groups is the key factor in functional recovery after the operation. Aiming at the study of human ulnar nerve trunk, we got ulnar nerve trunks from freshcadavers(male adult) and took section slides in every 2cm. All slides were stained by AchE histochemical method and the photoes of the serial slides were inputed into a computer. With the support of the OpenGL graphic library, contour-based 3D surface rereconstruction techniques were used to create the 3D visualization of every fascicular groups in the ulnar nerve trunk. The results of the 3D visualization can show the integral situation of the fascicular groups more directly and more effectively. Through arbitrary 3D rotation and virtual cutting on the computer, the spatial location distribution and adjacency relation of the fascicular groups in the ulnar nerve trunk can be observed from arbitrary point of view.It is important for 3D surface reconstruction from medical volumne data field (VDF)to render 3D anatomical structure details with an effect of high fidelity and reality.To achieve fidelity and reality, most existing approaches focus on increasing the spatial sampling density of VDF. The fact is that increasing the spatial sampling density of VDF can not be accepted in clinical practice. We presented a 3D iso-surface vector field smoothing(VFS) method to approach the effect of fidelity and reality without increasing the spatial sampling density of VDF. From the point of view of vector space, this VFS method regarded the extracted iso-surface as a vector field in 3D Euclid space. The smoothing neighborhood is defined as a set of normal vectors of vertices of the surface triangle meshes which adjoin the currently considered vertex. Based on the principle that properties in a neighborhood of each vertex of the surface triangle meshes space would have some coherence and not change abruptly, We substitute the mean of the normal vectors of its neighborhood for its current normal vector. When one step of VFS process finished, a next step of VFS process can be performed on the result of the last step of VFS process until the rendered results of surface reconstruction reach satisfaction.The effectiveness is proved by the experimental results which are also presented in the thesis.This paper also presented a method of 3D surface reconstruction and visualization for PACS workstation. Gues...