Research On Liver Tumor CT Image Segmentation Based On Deep Learnin

Liver tumor lesions have a high fatality rate.Early segmentation of liver and liver tumor from abdominal CT images can effectively assist physicians in making a diagnosis of the patient.However,manual segmentation is time-consuming and inefficient.In addition,CT images are characterized by low contrast,relatively similar gray values between organs and tissues,and variable shapes and sizes of liver tumors,making it still a challenging task to segment liver and liver tumors quickly and accurately.With deep learning techniques excelling in computer vision tasks,it has been widely studied and applied in the field of medical image segmentation.Therefore,deep learning technology is used in this paper to achieve efficient automatic segmentation of liver and liver tumors.The main research contents are as follows:(1)Segmentation of liver region in abdominal CT image.A Mutil-scale Semantic Feature Attention-Net(MSFA-Net)based on multi-scale semantic feature fusion attention mechanism is proposed for the problem that the gray values of liver and adjacent organ tissues in CT images are relatively similar,which makes some small detailed features not easy to be noticed.The network uses Dilated Residual Convolution(DRC)to capture multi-scale features.Then,MSFA module is used to combine the low-level features and high-level features of adjacent feature extraction layer with attention mechanism to obtain multi-scale features and pay attention to the important features.In addition,the feature transmission is enhanced by Deep Supervise(DS)at each layer of the network decoder.Experiments show that the proposed network can effectively focus on key features of different scales and obtain more accurate segmentation results.(2)Segmentation of liver tumor region in abdominal CT image.To solve the problems of small proportion of liver tumor in CT images,complex structure and low contrast with normal tissue,a liver tumor segmentation network based on cascade structure and Feature Fusion W-Net(FFW-Net)was proposed.First,the network adopts a cascade of two 3D U-Net structures to improve the ability of information extraction,and Side-output Feature Fusion Attention(SFFA)module is used to fuse the features of different levels and focus on important information in combination with Attention mechanisms.Then,the multi-scale semantic features are extracted using Atrous Spatial Pyramid Pooling Attention(ASPPA)module.Finally,the improved Deep Supervise(DS)module fully fuse these features to improve the segmentation performance of the model.Experiments show that the proposed network achieves better segmentation results than other methods of the same type.(3)Liver tumor CT image segmentation system.This system is implemented by using Python language,B/S architecture,Vue front-end technology and Flask back-end technology,which is convenient for physicians to use for segmentation prediction of liver and liver tumors.The system provides users with the function of automatic segmentation of liver and liver tumor from abdominal CT images,and realizes visualization of segmentation results.