Non-rigid Registration And Motion Field Computation Of The Lung Model

Lung cancer is a severe disease that harms human health,and its mortality rate ranks first among all cancers.Multi-modal medical image fusion method can realize early diagnosis and treatment of lung cancer,which is beneficial to increase the patients' 5-year survival rate and improve their quality of life.Due to the existence of complex curved surface,severe deformation and sliding motion,the non-rigid multi-modal medical image registration,lung field modeling,and accurate calculation of lung motion parameters are the key problems to achieve multimodal medical image fusion and tumor tracking.In this paper,lung point cloud simplification method is studied,and a new similarity measure for lung non-rigid image registration is proposed.The lung motion field estimation,the surface reconstruction and parameter calculation are realized.The main research contents of the thesis are as follows:A point cloud simplification method for lung model is proposed.Spatial hierarchical clustering and point cloud refinement method are used to obtain the initial values,then the distribution of the simplified point cloud is optimized using the Voronoi diagram-based partitioning method.A new method to choose the representative point of a cluster is defined.The point set in a cluster is triangulated and the point closest to the center of gravity is selected as the representative point.The scheme can improve the problem that the representative point may be biased towards the high-density point cloud area when the distribution of the point cloud is uneven.Experiments show that the proposed point cloud simplification method can improve the shape retention and distribution uniformity.A similarity measure for non-rigid medical image registration is proposed.The new metric discretizes mutual information to each pixel and adds a weight term consisting of gradients.the influence of the direction and the magnitude of the gradient has been taken into account.The new metric provides a strong constraint on the gradient direction consistency.The gradient direction of the corresponding pixels of the reference image and the floating image should be the same or the opposite.The characteristics can improve the problem that mutual information has large error in non-rigid registration in low contrast regions.A lung motion field modeling method based on non-rigid point cloud registration is proposed.The tangent-plane distance is used to represent the distance term,and the objective function is expressed as the Frobenius norm formation of the matrix.The optimal local affine transformation parameters can be calculated by matrix derivation and gradient descent method.The experimental results show that the tangent-plane distance reduces the constraint of the motion along the tangent direction.The proposed method does not require the assumption that the the source point cloud and the target point cloud should be in one-to-one corresponding relationship,which improves the problem that the registration would be trapped in local minima during the optimization process and thus enhances the accuracy of the lung field estimation.The reconstruction of the lung surface and the calculation of the lung motion field parameters are studied.The NURBS curve fitting method based on cylindrical coordinates is used to reconstruct the surface of the lung,and it achieves higher accuracy than that based on the Cartesian coordinate system.A calculation method of tidal volume is proposed.The lung model surface is discretized into a triangular mesh and the sweeping volume of the triangular patch is calculated by the integral method.The method can be used when the model surface is not closed.The average breathing vector and the motion vector component of the lung motion field are calculated,and they can be used to evaluate the damage and health condition of the lung.LungMed,a digital processing software platform for lung images,is developed.The platform has functions such as image visualization,image segmentation,point cloud reduction,lung field modeling,and motion field parameter calculation.Experiments using 4DCT images from DIR data are carried out to verify the effectiveness of the proposed non-rigid registration and motion field calculation methods.