Study On Vascular Segmentation And Registration For Multimodality Medical Images

Multimodality vascular images are of great importance to diagnosis of vascular diseases,navigation of computer-aided surgery as well as observation after surgery.Segmentation and registration of multimodal medical images are two important branches of medical image processing.Since image contrast is low and vascular structure is complex,accurate extraction of blood vessels in two-dimensional and three-dimensional medical images has always been a problem not to be solved properly.In addition,the registration between intraoperative twodimensional vascular images and preoperative three-dimensional vascular image can effectively assist the vascular interventional surgery to improve the efficiency of surgical operations and reduce the surgical risk.This article focuses on multi-modal vascular segmentation registration.The main contents of this thesis include:(1)An algorithm based on quadrature filter to automatedly extract vessel in different scales is proposed.Local phase response of the image at different scales is first extracted by using the quadrature filter.Secondly,the enhancement of the linear structure is achieved by blurring and shifting operations.Subsequently,the responses of the filters at different scales are fused to obtain the best blood vessel enhancement results.Finally,based on the enhancement of blood vessels,thresholding method is used to segment the vessels.(2)A three-dimensional vascular segmentation method based on convolution neural network is proposed which is used to extract vessel in Time of Flight Magnetic Resonance Angiography(TOF MRA).Firstly,the surface contour of the blood vessel is reconstructed from the vascular centerline.Then the training sample of the MRA data is established by sampling.In this paper,a convolutional neural network based on residual learning is constructed.The training set is used to train the samples until the network converges.Finally,the network model obtained by training is used to extract the vascular structure.(3)A 3D-2D registration method based on weighted Gaussian mixture model is proposed.Firstly,based on the result of vessel segmentation,a minimum path algorithm is used to extracts the vessel centerlines.In addition,minimum spanning tree is use to build the3 D vascular tree model.Secondly,gaussian mixture model of vessel point cloud is built.After that,orientation and radius is calculated by centerline.Then,a weighted gaussian mixture model is used to calculate the similarity between 3D and 2D vessel point.With the help of optimizer,the registration task is finally accomplished.