Research On The Technique Of Segmentation And Separation Of Vertebra Based On CT Image

For simulating the displacement and rotation of the vertebrae which nail has been implanted on, reconstructing a 3-D model of patient's spine is first step in virtual spinal diorthosis surgery. However, it is difficult to segment vertebrae image individually from CT image because of complicated structure and spatial position of spine. So we can reconstruct a whole 3-D spine model only after segmenting by using traditional methods. And we can't simulate the spinal diorthosis surgery because displacement between the adjacent vertebras is disabled in this whole 3-D spine model. For solving this problem, we could assemble a new 3-D spine model with individual vertebra which is separated from whole 3-D spine by using virtual assemblage technology. In this new 3-D spine model, displacement between adjacent vertebras is enabled. So simulating of spinal diorthosis surgery could be done.The researches of this thesis is root in the system of virtual spinal diorthosis surgery, and include CT image denoise and interpolation, 2-D segment of spine, 3-D reconstruction of segmented spinal image, and separated 3-D vertebra image from reconstructed spinal image. The description relate to these researches of this thesis are carried out as fellow:1. We bring out a denoise method of adaptive wavelet threshold denoise based on medial filter according to the characteristic of CT image, which mostly include impulse noise and white gauss noise. By using this method, mixed noise of CT image can be denoised and contour information of vertebra of CT image can be protected very well.2. CT image interpolation carried out by using the method of matching pairs of pixels for solving the problem of excessively great space between slices of CT.3. We improve on the energy function of Active Contours Model (Snake model). A alluring point be set on the edge of concave contour for luring the point of snake. And the Euclidean distance transformation between the lure point and snake point being appended to the traditional energy function of snake model. By this way, snake can approach the concave contour quickly and exactly.4. Volume rendering of segmented spinal image be implemented by using ray casting method.5. A method of separating 3-D vertebra from whole 3-D spinal individually is proposed. First of all, conjoint pixels of 3-D spinal between the neighboring vertebras are disconnected by 3-D thinning algorithm. Whereafter, a virtual balloon initialized between the space of neighboring vertebras, and inflated for detecting the gap between the neighboring vertebras. Then, some points produced alone the gap of neighboring vertebras in the sagittal images. Curve fitting with these points is done superveniently, and we called these curves are dissection lines. Afterwards, a closed 3D-plane is calculated based on the placement of the dissection lines. This 3D-plane dissects different parts of spine. Finally we can get individual 3-D vertebra image.