Analysis Based On Statistical Methods, Magnetic Resonance Imaging Of The Human Brain Image Segmentation And Three-dimensional Data

With development of medical diagnostic equipment, the high resolution magnetic resonance (MR) imaging is a powerful non-invasive tool for medical diagnosis, it is plays an increasingly important role in the quantitative analysis of anatomical structures. This paper focuses on two topics of the MR brain image analysis. One is tissue segmentation and the other is three dimensional shapes of the brain description and quantitative analysis.Tissue segmentation is a crucial processing in MR brain image analysis. For the MR image segmentation, we need to overcome the ambiguity of the image to improve the accuracy of segmentation, and find more accurate segmentation threshold, so a multi-threshold MR image segmentation method based on fuzzy Markov Random Field (MRF), clustering and Gaussian curves is proposed in this paper. In the approach, since the degradation of MRI, the neighborhood prior information was inducted by the energy of the cliques. First of all, Membership function of pixels was re-defined, which not only depended on gray features, but also had taken the type of variance and neighborhood information into consideration, and fuzzy MRF was developed; Second, the most simplified energy function of posterior probability was obtained with Bayesian, meanwhile, maximum a posteriori (MAP) was used as the statistical clustering criteria, and sought the solution of MAP by iterated conditional modes(ICM), then every class center was updated by the centroid of the fuzzy class. Third, two-dimensional histogram of image was produced, and projected into one-dimensional histogram by the optimum projection theorem. Fourth, the cluster center points obtained in fuzzy MRF clustering method were introduced into projection to define the gray scope of Gaussian curves, in which the projection histogram was fit with Gaussian curves, and several intersections were acquired at the point where two adjacent Gaussian curves meet. Finally, the lines identified by the intersections were found in the two-dimensional histogram to divide image as threshold values. Experiments show that the algorithm greatly improves the anti-blur and noise immunity compared to one-dimensional Otsu method and two-dimensional Otsu method, and it overcomes the problem of poor partition connectivity to a certain extent, thus an optimized segmentation result is obtained for MR image. Preliminary study was completed on three-dimensional shape of the brain tissue in this paper. Parameterized surface model of three dimensional objects is an important representation method for the shape analysis and studies. However, the traditional parameterization methods only attempt to preserve the areas or angles in the mapping, which does not guarantee the correspondence between the surfaces. In the paper, we propose a shape-character-derived parameterization method for statistical shape analysis of banana like tree dimensional objects structure. In this algorithm, firstly we extracted the shape character and determined the two-dimensional edge of the hippocampus according to tree 2-D images of the sagittal, coronal and transverse of the brain MR images. Secondly, MR images were reconstructed in the tree dimensional areas by Mimics10.01 software, and the three-dimensional surface of the hippocampus was mapped onto the unit sphere, and determined their latitude circle in according with the edge of the hippocampus in the two-dimensional surface. Thirdly, the body axis of the hippocampus was got by the barycenter of a series of latitude circles. Fourthly, a dateline is extracted on the surface, which is used to get the longitude for each latitude circle. Lastly, this series of latitude and longitude lines and their coordinates formed parameterized shape surface model to represent the hippocampus. The result shows that our method preserves the correspondence between the parameterized surfaces.In the disease related analysis of the hippocampus, this paper calculated the standardization volume of the hippocampus and measure the distance from the surface to the center axis, and using t statistics test the level of significance p of the two indicators above in Alzheimer's dementia hippocampus, which obtain meaningful statistical conclusions.