Research On Liver Vascular Segmentation Algorithm Based On Deep Learnin

Blood vessel segmentation of CT image is an important part of computer-aided treatment system.Accurate segmentation of liver vessels is an important prerequisite for the creation of tumor surgical planning tools and medical visualization applications,and the shape and location of intrahepatic vessels are of great significance for liver surgery.Because the vascular tree in the liver is complex and highly intertwined,accurate segmentation of liver vessels has always been a challenging task.For the task of liver blood vessel segmentation,two improved models are proposed:(1)Liver blood vessel segmentation algorithm based on distracted 2.5D UNET.Firstly,based on UNET,multiple adjacent slices are input at the same time to simulate the threedimensional mode and enhance the utilization of the information between adjacent slices by the two-dimensional model;For the complex and intertwined vascular structure,the resnest module is added to improve the model fitting ability to improve the segmentation effect;Finally,based on the improved model,the target is divided into two levels by using cascade method.The first level divides the liver region,and the second level divides the blood vessels in the liver region,so as to reduce the proportion of false positive in the segmentation results.(2)Liver blood vessel segmentation algorithm based on feature recombination and recalibration 3D UNET.Firstly,reduce the number of down sampling layers of 3D UNET to reduce the loss of vascular details caused by down sampling;Secondly,the feature recombination and recalibration module is added to avoid the decline of network fitting ability caused by the reduction of lower sampling layer,and the weights of blood vessels and background are reset at the same time in channel dimension and spatial dimension to improve the dissemination of blood vessel information in the network;Finally,the attention mechanism is added to the expansion path to make the overall constraint on the feature map,so that the focus area of the model is focused on the blood vessel.Through experiments,the proposed model is compared with the two-dimensional and three-dimensional models with good segmentation effect in the near future.Based on the distracted 2.5D UNET model,compared with UNET + + and other models,the dice index is improved by about 1% ～ 2%.Feature reconstruction and recalibration 3D UNET model has a dice index of 64.80%,which is higher than the three-dimensional segmentation models of medical images such as c2 fnas panc and UMCT.In conclusion,the model proposed in this paper can better segment liver blood vessels,which has very important practical significance.