CT Images Segmentation Of The Liver And Tumor Based On Deep Learning And It’s 3D Model Reconstruction

Nowadays,the most effective treatment for liver cancer is surgical resection.Thus,it is very necessary to analyze and evaluate the patient’s liver condition before surgery.CT scan is currently the most common method for diagnosis of liver cancer.But how to accurately extract the liver and tumor areas from the abdominal CT image and precisely display the segmentation results in the form of a three-dimensional model has always been a difficult problem.Excellent segmentation and reconstruction results can allow doctors to have a clear,intuitive and comprehensive cognition of the patient’s liver condition before surgery,so as to help doctors formulate better surgical plans.All of these have become one of the urgent needs of modern doctors.With the rapid development of artificial intelligence in recent years,the application of deep learning in the field of automatic segmentation of medical images and the generation of three-dimensional models based on the segmentation results of two-dimensional slice have become popular research topics.Aiming at segmentation of liver and tumor areas and reconstruction of liver 3D models in CT images,this paper designed a set of complete segmentation method based on deep learning,and improved the MC(Marching Cube)3D model reconstruction algorithm.The specific contents are as follows:Firstly,this paper analyzes the advantages and disadvantages of the commonly used CT image preprocessing methods through experiments,and designs a set of liver CT image preprocessing scheme according to the characteristics of CT images and segmentation targets,which mainly covers filtering,edge extraction and gray histogram equalization.The liver and tumor areas are more obvious and the boundary information is more prominent when the images were processed by this set of methods,which lays a solid foundation for the subsequent accurate segmentation.Secondly,in view of the false positive segmentation caused by direct input of CT images without target regions into the network,the scheme adopted in this paper is to design a classification neural network based on DenseNet.By introducing group convolution and improving the Dense module,the CT images with target regions can be screened out.Then the CT images can be input to the appropriate segmentation network to improve the segmentation accuracy.Then this paper proposes a cascade segmentation method for liver and tumor segmentation.To begin with,the liver area segmentation model PcA U-Net can be designed based on U-Net,including the design of the multiscale residual convolution module instead of ordinary convolution,and the addition of a multi-feature attention module to the bottom of the network.The experiments prove that the network greatly increases the accuracy of liver segmentation.Then,the tumor area segmentation model,LPcA U-NET,can be designed based on PcA U-NET,which further improves the multi-feature attention module,improves the jump connection structure,and introduces the boundary monitoring module.The experiments show that the network has a good effect on tumor segmentation.Finally,the interpolation method can be improved based on MC 3D model reconstruction algorithm,the smoothness parameters can be introduced,and the 3D liver model can be reconstructed by using the segmentation results of liver and tumor areas.