Volume Image Segmentation Based On Vector Quantization And Its Applications On MRI

Image segmentation is an important technique of image process, and the segmentation on medical image is a popular application. In recent years, segmentation on volume image has been developed gradually, especially on the application of medical volume image. Nowadays, the research of volume image segmentation still has some defections at the utilization of neighborhood structure information, automatic computation of segmentation number, and spatial feature segmentation. Medical image segmentation is still a challenge for researchers, and has significant theoretical and practical value.This thesis conducts research on medical MRI image segmentation, and our main work is as follows:(1) The thesis conducts research on image segmentation using Vector Quantization (VQ), and the local regions (sub-blocks) of the image are classified during VQ process. During the process, not only the gray value information of pixels is utilized, but also the neighborhood structure information is used, similar with the cognition process of human vision. During the VQ process of sub-blocks, the codebook design is realized by SOM neural network, the optimal size of codebook is obtained by minimizing the ratio of within-class scatter and between-class scatter, and accordingly the segmentation number of image is determined adaptively.(2) By developing the proposed VQ-based image segmentation method into higher dimension, the thesis conducts further research on the segmentation of MRI volume image. A volume image segmentation method based on VQ is proposed, which takes the cubes of volume image as basic units during VQ-based 3D data segmentation. Based on the features of MRI volume images, the volume interpolation algorithm is designed, the edge pattern detection method for cubes is developed, the slice by slice segmentation manner and integral segmentation manner are presented during SOM-based codebook obtaining process, and finally the vectors constructed by cubes are segmented adaptively by VQ process. The proposed volume image segmentation method is applied to the segmentation of human brain MRI volume images, and the experiment and result analysis are conducted on IBSR and Brain Web database respectively.(3) Based on the volume segmentation result of value domain, the thesis proposes a spatial domain segmentation method for volume image by analyzing the spatial features of volume image. Firstly, the spatial connectivity of value domain segmentation result is computed by connectivity detection, and the volume image is further segmented accordingly. Then, the quantitative description on volume image segmentation is obtained by computing the spatial geometrical parameters of volume segmentation result, and the segmentation of MRI volume image is accomplished. The experiment is conducted on the result of value domain segmentation, the validation of spatial segmentation is confirmed, and the quantitative description of brain components and the spatial geometrical parameters of the segmented tumor are applied to practical treatment of clinical research.