Study On Segmentation Algorithm Of Three-dimensional Vessel Tree

In clinical medical diagnosis,medical images are important auxiliary information for doctors in diagnosis and surgery.The human body contains rich vascular tissue.The use of computer technology to segment the vascular tree from the medical image makes it easier to perform qualitative and quantitative analysis of organs and diseases,and helps doctors to develop more detailed and reasonable surgical plans.So,it has important clinical significance.In recent years,with the improvement of computing power of computers and the existence of massive data sets,it has been widely studied and favored in the academia by training deep convolutional neural networks for the segmentation of medical images.However,the use of a deep fully convolutional neural network to segment three dimensional images of hepatic vessels presents several difficulties: 1)Although abdominal spiral CT images are readily available,the annotation of 3D hepatic vessel data sets is time-consuming and labor-intensive.There are few 3D blood vessel images with labeled datasets;2)In order to separate the hepatic veins from the portal veins,conventional methods based on graph analysis and image registration are not robust and complex.In the segmentation results of the fully convolutional neural network,there are many misclassification resultsIn response to the above problems,this paper proposes two methods to solve them separately.Firstly,an intrahepatic blood vessel segmentation method based on statistical morphology and Hessian matrix is proposed.Experts can correct the segmentation result to obtain accurate annotation.As a result,the speed and accuracy of hepatic vascular data labeling can be significantly improved.A full convolutional neural network algorithm framework with iterative connected-domain analysis was proposed to segment the hepatic and portal vein vasculature from abdominal helical CT images.The main work of this article is as follows:(1)A semi-automatic algorithm based on statistical morphology and Hessian matrix is proposed for the segmentation of the blood vessels within the liver.Firstly,the original CT image is denoised and smoothed with improved three-dimensional median filter.Secondly,in order to obtain better Hessian enhancement effect,a top hat operation method based on statistical information is proposed for the problem of uneven gray levels of background pixel values.Then,in order to enhance intermittent blood vessels,the image is grayscale transformed and subjected to multi-scale Hessian enhancement;Finally,the Hessian-enhanced image is smoothed using anisotropic diffusion filtering,and using the two-stage segmentation algorithm including regional growth and active contour model to obtain the complete segmentation result of the vessel tree.(2)An algorithm framework combining fully convolutional neural network and iterative connected domain analysis is proposed for the segmentation of hepatic veins and portal veins.Firstly,on the basis of the semi-automatic algorithm segmentation results,manual correction is performed to obtain complete hepatic vein and portal vein annotation data sets;subsequently,use the rotation,translation and deformation methods to perform data amplification on the training data set.Two-dimensional and three-dimensional fully convolutional neural network models were trained using the augmented data set,and by using a weighted loss function to improve the accuracy of the 3D model segmentation.During the test,the results of the segmentation of the two fully convolutional neural network models are analyzed in an iterative three-dimensional connected-domain analysis to obtain a consistent semantic segmentation result of the vessel tree.The qualitative and quantitative test experiments on several sets of spiral CT datasets verify the validity of the two proposed algorithms.The contrast experimental results with normal region growth algorithm show that the proposed semi-automatic intrahepatic blood vessel segmentation algorithm not only achieves better visual effects for the segmented hepatic vascular tree,but also has a higher quantitative index than the ordinary region growing algorithm.The multi-class vascular segmentation algorithm framework based on the full convolutional neural network has a high degree of automation and can effectively separate hepatic veins and portal veins,providing an effective aid for clinical diagnosis and surgical planning.