Research On Liver And Tumor Segmentation Method Based On Improved U-Net Network

Liver cancer is a malignant tumor with extremely high morbidity and mortality in the world.It is a serious threat to people’s health.Early screening and treatment can effectively reduce the morbidity and mortality of cancer.After obtaining the CT image of the patient’s abdomen,the doctor needs to further determine the location,size,volume and other information of the liver tumor.It is necessary to mark the liver tumor area in the image.However,manual segmentation requires a lot of time and effort.In order to improve the accuracy and efficiency of liver tumor segmentation,this paper proposes a study of liver and tumor CT image segmentation methods based on U-Net network.(1)In view of the shortcomings of traditional gray-scale,edge,and texture-based segmentation methods,this paper proposes a liver and tumor image segmentation method based on an improved 2D U-Net network.The 2D U-Net network is used as the basic network.The convolutional layer adds a residual learning structure to prevent over-fitting during network training;in view of the small proportion of liver tumors in CT images,the loss function is easy to fall into the local minimum problem,the cross-entropy loss function and Dice The loss function composes the mixed loss function,which effectively solves the problem of category imbalance in the network training process.(2)In view of the problem that 2D U-Net does not learn image spatial information,this paper proposes a 3D U-Net network.In order to prevent the gradient from disappearing during network training and improve the convergence performance of the network,the residual module is introduced,and the space is introduced at the same time.The location attention mechanism and the channel attention mechanism associate the spatial dimension with the semantic information of the channel dimension,and combine the global and local features of the image to enhance the expressive ability of the features.calculation is very large,the storage requirements are high,and the end-to-end training is limited by GPU memory and data size.A 2.5D U-Net network is proposed.The introduction of hole convolution between the encoding layer and the decoding layer reduces the loss of features,increases the field of vision of the liver and its tumors,improves the accuracy of the detection and classification of the liver and its tumors,and introduces the spatial position attention mechanism and channels.The attention mechanism enhances the ability to express features.The improved U-Net network is used to segment the liver and its tumors.The experimental results show that the improved network model has a better optimization effect.It can accurately segment the liver and its tumors from the abdominal CT image,which has certain clinical references value.