Research Of Algorithms For Hemorrhagic Stroke MR Image Processing

Strock has become the third reason for human death in the twenty-fist century. Therefore, we need to find effective treatment methods for this kind of devastating disease. With the improvement of medical imaging, MRI is used in the tests and diagnosis for stroke widely, and obtained the rapidly development, however, it took a lot of image data, and brought huge workload for doctors. The development of computer technology makes possibility for taking computers into the medical image processing and analysis field, and brings a revolutionary leap for imaging diagnostic. In this thesis, according to characteristics of the hemorrhagic stroke MR image, researched the algorithm for MR image processing in the following aspects mainly:(1) For the image preprocessing, start with the mathematical model of the image noise, research the common denoising model, Present an adaptive denoising algorithm based on non-local thought. In this algorithm the image space information will be used, and have a better denoising result. This algorithm gets a noise variance estimates through the Laplace transform of the noise image, and then obtain the smooth weight for denoising algorithm. Finally we have an adaptive denoising algorithm based on non-local thought. The algorithm has bigger superiority not only from the visual effect but also in peak signal-to-noise ratio after processing.(2) Nowadays, Fuzzy c-means algorithm is more popular in image segmentation, In this thesis, we have an improvement for the robust FCM algorithm, the new algorithm embedded the non-local weights in its penalty term, and eliminated the noise effects using the image redundant information. Experimental investigation shows that this algorithm is more suitable for the noise image segmentation, and has very good inhibitory effect for image noise. (3) For lesions extraction, this thesis researches the active contour model based on partial differential equation. On the analysis of parametic cative contour and geometric active contour, this paper focused on the non-initializational LiChunMing model based on the level set mothods. According to the characteristic that the initial contour will have an influence for the segmentation result, the thesis presents an initial contour determination mothod using fuzzy membership. Through setting a threshold of FCM algorithm in the category of lesions fuzzy membership, we can obtain the initial contour curve of the level set directly. Experiment shows that this algorithm has very big enhancement in the segmentation accuracy and segmentation time.