Research On Automatic Segmentation Method Of Liver CT Images Based On Deep Learning

With the improvement of people’s living standards,the number of patients associated with liver organ lesions is rising every year,causing incalculable damage to patients and their families.Computed tomography technology is a common means of liver examination,and segmentation of liver and liver tumors from liver CT images is clinically important to enable physicians to quickly locate complex liver and liver tumor regions and improve diagnostic efficiency.For the past few years,deep learning algorithms have been superior in segmentation of images,and more and more researchers have applied deep learning to the field of medical image segmentation.Due to the morphological diversity of liver and liver tumor,the automatic liver CT images segmentation method based on deep learning has become a hotspot and difficulty in current research.In this thesis,based on deep learning technology,we have studied two aspects of fast segmentation of liver region and accurate segmentation of liver and liver tumor in liver CT images,and also designed a liver diagnosis system software for improving the diagnosis efficiency of liver patients.The details are as follows:(1)To address the problem that most of the deep learning algorithms for liver region segmentation in liver CT images run slowly,this thesis designs a lightweight automatic liver segmentation method based on PPLC-UNet,which can achieve the segmentation of liver regions in liver CT images and label the liver.PPLC-UNet is an improvement of U-Net model with encoding and decoding structure.The encoding part adopts the backbone network of PP-LCNet for feature encoding,the decoding part incorporates the improved SE module and residual structure,and the hybrid loss function Mix Loss combining BCE Loss and Dice Loss is adopted for model training to achieve fast segmentation of liver regions in liver CT images.(2)Aiming at the problem of low accuracy in the segmentation of liver and liver tumor from liver CT images at present,this thesis designed an automatic segmentation method of liver and liver tumor based on multi-scale semantic feature network,which could segment liver and liver tumor from liver CT images and label the contour of liver and liver tumor.The multi-scale semantic feature network adopts Res2 Net backbone network to extract multi-scale semantic features,introduces RFB module to improve the extraction ability of semantic features,introduces SA module for semantic segmentation to enhance semantic features,and integrates feature information of different scales through transposed convolution after adjusting semantic features at different scales.The precise segmentation of liver and liver tumor in liver CT images was realized by using the combined Loss function CG Loss training network which included cross entropy loss function and generalized Dice loss function.(3)In order to facilitate the automatic segmentation of liver CT images by computer vision technology,this thesis designs a set of liver diagnosis system software based on Py Qt5 software framework.The software is developed using Python language,and the data is stored in the My SQL database.Two liver CT image segmentation methods proposed in this thesis are added,aiming to improve the treatment efficiency of patients with liver diseases and the service quality of the medical system.