Research On Liver Tumor Segmentation Method Based On Deep Learning

The task of liver tumor segmentation is specifically to segment the tumor area of the liver from the CT image.However,the task is difficult due to the inconspicuous contrast between the liver and adjacent organs,the extensive tumor tissue morphology,and high noise.With the continuous improvement of deep learning algorithms,the continuous improvement of GPU computing power and the continuous expansion of databases in recent years,deep learning methods,especially U-Net networks,have been widely used in the field of medical image processing.However,the U-Net model may not perform well when the liver tumor size is different and the image contrast is weak.In order to improve the segmentation effect of U-Net on liver and tumor,this paper proposes an improved network model based on Res U-Net.Res U-Net is a UNet network with residual module added.Optimize the jump connection of Res U-Net through the attention gate module to better combine the deep features with the shallow features;optimize the short-range connection through the squeeze-incentive mechanism to make the network pay more attention to the liver and tumor parts;through the design bottleneck convolution Structure to reduce model parameters and speed up model convergence;by designing a dynamic mixed loss function to improve the stability of the model training process.Experimental results show that the above improvements can effectively improve the segmentation effect of liver and tumor.In order to improve the segmentation effect of tumors of different sizes,this paper optimizes the design for tumors of different sizes.Through the window optimization setting module(WSO),the network automatically selects the window width and window position,thereby improving the contrast between the tumor and the background;By introducing multi-scale cavity convolution and pyramid pooling modules,the model’s ability to extract context information and image spatial information is increased,thereby improving the model’s ability to segment tumors at different scales.Through experimental verification,the improved Res U-Net model proposed in this paper has significantly improved the segmentation effect of liver and tumor.Optimization for tumors of different sizes can segment tumors of different sizes very well,and has good practicability.