Liver Segmentation Based On Deep Learning

The liver is the largest organ in the human body.At present,liver cancer is the second largest cancer after lung cancer in the world,which seriously affects human life and health.For the treatment of liver cancer,liver tumor resection is still the most effective.However,the internal tissues of the liver are intricate and complicated,which has brought great difficulties to tumor resection.Therefore,accurate segmentation of liver parenchyma,tumors,and blood vessels before surgery is of great significance for the formulation of surgical plans.In recent years,artificial intelligence technology has developed rapidly,so it is also widely used in the field of medical image segmentation.This paper is mainly based on the method of deep learning to segment liver parenchyma,liver tumors and hepatic vein blood vessels.The research work is as follows:1.For the liver parenchyma segmentation task,this task is relatively easy,mainly based on 2D convolutional neural network implementation.In this paper,V-FCN full convolutional neural network and Unet network based on coding and decoding structure are respectively implemented for this task.Then,ResUnet network is proposed for network degradation,and the optimal results are obtained.In order to remove noise,more accurate results are obtained by post-processing of the largest connected component.2.For the tumor segmentation task,in order to solve the problem that the 2D network can not extract the features in the CT slice,this paper implements the 3D Unet,DenseUnet,and Vnet networks.The 3D network can focus on learning the association information between the CT layers and get more effective features.Aiming at the problem of segmentation of small target areas,based on Vnet,long connection and short connection and multi-scale feature extraction modules are introduced to take into account global features and local features to implement Vnet++ network,and finally get better segmentation results than Vnet.3.For the task of venous segmentation,this paper uses two algorithms.One is a traditional algorithm based on region growth,which manually selects seed points and sets growth criteria to achieve blood vessel segmentation.The other is a method based on deep learning.Aiming at the disadvantages of 2D and 3D networks,a more efficient 2/3D Unet network is implemented.Then,the SE_Vnet++network that can expand the importance of the receptive field area and the importance of the automatic learning channel is realized,and a more accurate segmentation result is obtained than Vnet ++.In this paper,multiple groups of abdominal CT image data are used as test data to segment liver parenchyma,tumors,and venous blood vessels and calculate DICE indicators.Finally,the DICE indexes of liver parenchyma,tumors,and venous blood vessels have reached 0.9602,0.7971 and 0.7531 respectively.The results prove the feasibility of the proposed algorithm.