The Research Of Brain Extraction Based On SIFT Feature Matching

With the rapid development of technology,medical imaging equipment is widely used in clinical physical examination and medical research.Compared with other medical imaging technologies,MRI provides clear brain imaging with high resolution of soft tissue,avoiding radiation on human cells and wound on human-body.Accurate,stable and efficient brain extraction in MRI has an important impact on the clinical diagnosis and brain functional analysis.At present,brain extraction from MRI can be manual or automatic manner.Manual extraction can reach high accuracy,but it takes a lots of time and has influence of subjectivity;Automatic extraction can achieve higher speed,accuracy and stability.But it needs to multiply manually adjust a lot of unfixed parameters to get better extraction results.In order to solve this problem,this paper proposes a brain extraction method based on SIFT feature matching to adjust active contour.The methed first initializes a active contour at the vicinity boundary of the brain.Then,using improved BET algorithm evolves the active contour,pushed closer to the brain boundary.In the evolution process of BET,vertices in the active contour are used as the key points calculated their descriptors by SIFT.The Euclidean distance of the descriptors and the spatial distance of the keypoints are used as the metrics showing the similarity of the key points between the target image and template image,to automatically adjust the parameter of BET and renew active contour.Through multiple contour evolutions,using Graphcuts method refines the active contour to obtain more precise brain boundary.Because,the SIFT algorithm is frequently used in this method,spending lots of time in every iteration.This paper optimizes SIFT in parallel based on the CUDA architecture to accelerate its speed.Through the division of grid,each block calculates the main direction and feature descriptor of each vertex on the active contour,and completes vertex matching.Experimental results show that the parallel optimizing SIFT spends greatly shorter time,lending to fast matching between vertices.This paper,proposes an accurate,stable brain extractation combineing improved BET algorithm and the image matching method,to automatically adjust the parameter.At the same time,using CUDA accelerates the speed of SIFT.Through many experiments,we make quantitative and qualitative analyses,which show that the method can get better extraction results.