A Fast Point Cloud Registration Algorithm For Neurosurgery Navigation

With the increasing improvement of living standard,more attention is paid to the health condition,which makes the medical technology is also facing higher requirements.The neurosurgical navigation system is a new medical treatment method based on computer technology,point cloud processing technology and other advanced science and technology.The application of point cloud registration technology was first proposed in medical diagnosis.It can transform the source point cloud to the same coordinate system of the target point cloud.This algorithm can effectively help doctors to make medical diagnosis,more conducive to the evaluation of patients' affected parts.Point cloud processing technology plays a key role in the assisted navigation of neurosurgery and is also the focus of this thesis.In this thesis,the point cloud simplification and point cloud registration are studied accurately and carefully with the background of neurosurgical navigation.The specific work mainly includes the following three aspects:(1)When point cloud data simplification is carried out,the commonly used methods such as grid method and curvature usually result in incomplete boundary reservation,easy loss of feature points,oversimplification of plane and non-feature areas and the formation of void phenomena.Therefore,this thesis proposes a spatial subdivision point cloud simplification algorithm based on K nearest neighbor.In this method,the data of the nearest neighbor of the sampling point is firstly obtained as the local surface,and the least square method is used to fit the micro-tangent plane of the nearest neighbor of the sampling point.The sampling point is projected onto the micro-tangent plane,and the boundary features of scattered point cloud are identified according to the maximum included Angle of the line between the sampling point and the projection point corresponding to the K nearest neighbor.Then the Gaussian curvature of the point cloud data after removing the boundary features is calculated and compared with the given value to distinguish the characteristic and non-characteristic regions.For the feature region,a curvature-based simplification method is used to simplify the feature region.For the non-characteristic region,the simplification method based on the center of gravity is adopted.Finally,in the addition stage of boundary points,characteristic regions and non-characteristic regions,the distance threshold is set to calculate the Euclidean distance between two points of clouds and filter the redundant and repeated parts.(2)Aming at the problem of traditional iterative closest point algorithm(ICP)original position of the demand is higher and the computational cost of finding the nearest corresponding point is large,based on principal component analysis(PCA)registration method of two steps is proposed in this thesis.Firstly,the PCA was selected to carry out the rough registration of point cloud data.For the point cloud data after rough registration,feature points were extracted;for the point cloud data after feature points were extracted,ICP algorithm was used to carry out the fine registration operation.Compared with the original ICP algorithm,the speed and precision of the two-step registration method based on PCA are improved effectively.(3)Based on the above research and combined with the specific neurosurgical navigation system,the constructed point cloud data simplification method and point cloud registration method were applied to the neurosurgical navigation system.The registration procedure was completed before surgery.The completed registration steps have effectively improved the accuracy and precision of intraoperative navigation,and the completed point cloud data simplification steps have effectively improved the registration efficiency of the overall navigation system.