Research On Three Dimensional Reconstruction For Real Head Model Based On Support Vector Machine

Highly evolutional brain is the centrum which predominates all the activities of human being. The task of brain science is to open out the profound mystery of brain function. It has great significance in life science. From it, research on the analysis of three dimensional cerebral electric field of is the hot area recently. Nevertheless, the precondition for this study is to construct the real head model of human. In this paper, using Magnetic Resonance Imaging (MRI) image sequence, the real models of 7 kinds of encephalic tissues has constructed successfully based on Support Vector Machine (SVM) theory. The main works are illuminated as follows:(1) MRI image filtering: Filtering process can obtain high quality MRI image in order to build solid foundation for image segmentation and image three dimensional reconstruction. The most important characteristic of MRI image is the complicated change of gray level. So, traditional filtering methods can not fit for this kind of complicated image. In this paper, Adaptive Template Filtering Method (ATFM) is adopted to filter MRI image. This algorithm not only effectively suppresses noise, but best preserves the useful information to the largest degree as well for MRI image. However, selecting threshold for ATFM is a complicated problem which directly affects the filtering result. Combined with immune algorithm, Immune ATFM (IATFM) was proposed to effectively deal with the problem of threshold selection and perfected ATFM. The experimental results demonstrate that IATFM yields satisfactory filtering result. Both filtering performance and visual effect of IATFM surpasses traditional filtering methods. So it is a filtering method fitting for complicated medical image.(2) MRI image segmentation: Since the generalization ability of SVM is severely depend on selecting feature vectors and their spatial distribution, in this paper, a series of local texture measures and statistical vectors of gray are extracted for each pixel as feature vectors. Then Principal Component Analysis (PCA) is used to reduce dimension. PCA can not only eliminate redundant information which is linearly interrelated among the feature vectors, greatly reduce data quantity, enhance image information, but make the sample space prepared for SVM more compact and reasonable as well. It is essential reason to combine PCA with SVM to improve generalization ability.In head MRI image, the boundary of each encephalic tissue is highly irregular and non-linear. It is real challenge to traditional segmentation algorithms. The advantage of SVM is to deal with the problem of high dimension, non-linearity and irregularity. In this paper, SVM was used to segment MRI image. Through deeply studying on Artificial Immune System (AIS) theory and SVM theory, combined with multi-knowledge, a whole set of theory and algorithm of Dimensional Weighted Immune SVM (DWISVM) was firstly proposed. DWISVM ends the phase which selecting parameter for SVM has been depending on heuristic experiment and manual adjustment and lacking theoretical direction. At the same time, DWISVM break through the traditional idea of SVM which assume that each dimension of sample is equal for the contribution to the final classification accuracy. The new idea proposes the concept of dimensional weight. And combined with artificial immune theory, DWISVM can deal with linkage optimization problem of SVM parameters and each dimensional weight. DWISVM perfects SVM theory. Using DWISVM, 7 kinds of encephalic tissues are finally segmented only from T2 weighted MRI image, and reach satisfactory generalization accuracy. This 2D segmentation result is the solid foundation for highly accurate 3D reconstruction. From the literature we can search, non scholar can make it to this degree. And no correlative report is found to classify gray matter and white matter from T2 weighted MRI image.(3) MRI image reconstruction: The highly non-linear 2D boundary of encephalic tissue must results in highly irregular 3D curve surface which mainly represents in the remarkable and frequent change of curvature and the uncontinuity of curve surface. It is very difficult to reconstruct this kind of highly irregular 3D curve surface in image 3D reconstruction area. It is real challenge to traditional segmentation algorithms. The advantage of SVM is to deal with the problem of irregularity. Through detailedly analysis problem and deeply studying on SVM theory, Sphere-Shaped SVM (SSSVM) was firstly introduce to medical image 3D reconstruction area. It can transform the irregular tissue in three dimensional space to a regular hyper-sphere in high dimensional space. The function which describes the surface of hyper-sphere in high dimensional space is the mathematical model of 3D encephalic tissue. In order to deal with the problem of selecting parameters for SSSVM, SSSVM was combined with immune algorithm. So Immune SSSVM (ISSSVM) was proposed to reconstruct MRI image and reach satisfactory modeling accuracy. Finally, 7 kinds of encephalic tissues were processed for 3D visualization.