The Research Of Brain Extraction Based On Scale-Invariant Keypoint Matching

With the rapid development of medical imaging technology, and the wide application of medical imaging equipment in clinical, such as CT and Magnetic Resonance Imaging, the analysis of neuroimaging techniques plays a more and more important role and has laid a solid theoretical foundation for medical clinical diagnosis. In the past decade, magnetic resonance imaging technology, avoiding radiation on human cells and wound on human-body, has developed quickly and found far-ranging applications. At present, according to the different degree of human’s participation, brain tissue extraction technology can be divided into three kinds of methods, which are manual, semi-automatic and automatic. The manual method can achieve high accuracy, but it is time consuming, has higher requirement of the operator in professional skill, and has influence of subjectivity; the automatic method are often hybrid and can achieve higher accuracy and stability and speed, but it is always sensitive to image scale and rotation, and needs to adjust parameters which are quite a lot and always not the same.In order to solve the automatic extraction algorithm’s sensibility of image scale and rotation, this paper proposes a SIFT feature matching method based on distance constraint. First, it extracts the profile lattice and the SIFT descriptors for template image and the subject image; then it matches in both vector and coordinate distance, and makes the results in ascending order, selects the first five points for the results of vector and coordinate distance, if there are same point, it is the match point and the matching is successful; finally, a two-way matching method is adopted to ensure the accuracy of the method.In order to solve the problem of tuning parameter, this paper presents a algorithm of brain extraction based on scale-invariant keypoint matching. It uses the SIFT feature matching method based on distance constraint, adding a loop when it is processing subject image. For the subject image, first, it needs to get the discrete lattice of the image by the improved BET algorithm; then, it needs to get the feature descriptors for the discrete lattice and match with the discrete lattice of the template image. Automatically adjust the parameters according to the matching results and continue the contour evolution with new parameters until reached the maximum iteration limit; finally, it can get the accurate brain contour.This paper presents a method to extract the brain tissues which uses the improved algorithm based on BET and the SIFT algorithm comprehensively. This method achieves the purpose that the algorithm is insensitive to image scale and rotation, and can adjust the parameters automatically. We make quantitative and qualitative analysis by a lot of experiments. The analysis result shows that the algorithm can get a good extraction effects.