The Design And Implementation Of Skull And Ventricle Segmentation Algorithms Based On MR Brain Image

MR (Magnetic Resonance) image has been widely used in medical image processing and analysis for the higher resolusion and soft tissue contrast. MR brain image segmentation is an important part of medical image processing and analysis, which is not only the basis of other brain medical image processing and analysis such as brain diseases diagnosis and 3D resconstruction, but also improving the real-time and reliability of diagnosis and treatment of brain diseases effectively.This thesis generally introduces MR imaging and MR brain image features firstly, and summarizes the classifications of brain image preprocessing and segmentation methods separatly. On this basis, sub-linear gray transform, SUSAN (Smith Univalue Segment Assimilating Nucleus) and PM (Perona and Malik) algorithms have been designed and implemented considering the fruzzy MR brain images. The experiments show that SUSAN and PM of the three algorithms can preserve edge details while image filtering.For enabling accurate measurement of brain structures, this thesis focuses on the studying and implementation of the stripping skull algorithm based on BET (Brain Extraction Tool) and the ventricle segmentation algorithm based on GVF (Gradient Vector Field) Snake. BET is an automated, highly efficient segmentation algorithm of stripping skull. Automated brain center initialization of BET can't finish the stripping skull tasks accurately for the images that contain non-head tissues (containing neck data), so this paper has designed and implemented BET segmentation algorithm of the manual brain center initialization. For the ventricle segmentation of the low quality MR brain images, SUSAN filter is used before ventricle segmentation using GVF Snake algorithm to get the final results.The two above algorithms have been carried on under the Microsoft Visual Studio 2005 enviroment, and integrated into the Unimed of NSR. The experiments show that the both segmentation algorithms separately finish the task successfully, and the results have been recognized by the clinicians. To some extent, they make fuction in computer-aided diagnosis.