Research On Brain MR Image Segmentation Method Based On Fuzzy C-Means

Magnetic resonance images have many advantages, such as high contrast, high resolution, multi-azimuth and so on. It is widely used in various types of brain disease research, but due to magnetic resonance imaging equipment and other objective factors, the clinical collection of brain tissue MR image has being noise, uniform gray, partial volume effect and other unfavorable factors, and brain tissue shape, boundary and topology is more complex, then how to quickly and accurately segment brain tissue MR images is the focus of today's research.Fuzzy c-means (FCM) clustering algorithm can well describe the complexity of MR images and blurred the phenomenon of organizational boundaries, and it can meet the needs of unsupervised segmentation, so in recent years FCM algorithm become more popular medical image segmentation method, by domestic and foreign scholars extensive research. In this thesis, the FCM-based brain MR image segmentation method be studied and improved, and then the DARTEL method is used to register the brain image after segmentation. And the data of Alzheimer's disease and mild cognitive impairment were used to verify the effectiveness of the method.The main work of this thesis is as follows: 1. The method of edge detection is used to remove non-brain tissue from NIFTI format images, the effects of non-brain tissue components on the segmentation results were excluded, the brain tissue was extracted. And then improve the accuracy of edge detection. 2. From the fuzzy C-means algorithm, the advantages and disadvantages of the algorithm are analyzed, the objective function of the fuzzy C-means algorithm is based on the Euclidean distance, but the Euclidean distance is not clear to distinguish the small differences between the data points, and the Mahalanobis distance can adjust the geometric distribution of the data adaptively, so that the distance of similar data points is small. Therefore, the distance function of the algorithm is improved by using the Mahalanobis distance instead of the Euclidean distance. The algorithm is used to extract the image of the brain tissue. The two types of tissue types were divided into three groups: gray matter and white matter. The results were comparative analysis by volume calculation and machine learning classification. 3. The segmented images were registered with DARTEL registration method to the unified template, the brain atrophy regional comparison.The improved method and the original method were compared by volume calculation,classification result analysis and regional contrast of brain atrophy. It was found that the improved method had a good result in the localization of the lesion area and the correctness of the classification result, it can effectively improve the accuracy of brain MR image segmentation, Alzheimer's disease and mild cognitive impairment in the brain MR image segmentation has a better effect and higher accuracy.