Abdominal CT Sequence Image Segmentation Based On Convolutional Neural Network

Medical image segmentation,an important research direction in medical image analysis,is of value in both theoretical research and application on pathological analysis,clinical diagnosis,dynamic surgical planning,computer-aided medical treatment and so on.While Liver cancer is known as the "king of cancer" for the high mortality and malignancy of itself,in which,the Primary HCC is a malignant tumor with high incidence and great harm in China.Therefore,it is of great significance to do research on accurate automatic liver segmentation diagnosis.In addition,the pancreas with a very small volume is also one of the most important abdominal organs in the human body,of which,the status has great significance to human health.Meanwhile,pancreatic cancer with its own characteristics,strong invasion,early metastasis,high malignancy,rapid development and poor prognosis,has become a serious threat to human health and poses a huge challenge to clinical medicine for the low five-year survival rate less than 5% and the high mortality rate.Pancreatic cancer has become a serious threat to human health and poses a huge challenge to clinical medicine.Hence,the research of automatic pancreas segmentation and recognition has important reference value for doctors to diagnose and determine the surgical plan.Computed Tomography(CT)as one of the main medical imaging technologies has been widely adopted in clinical examination and diagnosis because of the advantages of high spatial resolution and high signal-to-noise ratio and so on.Based on above,the segmentation of liver and pancreas in abdominal CT images is studied in this thesis with deep learning method,convolutional neural network.The main research contents are as follows:1)Proposing a liver CT image segmentation method based on LiBlockNet,aiming at improving the problems of large individual difference of liver,low gray contrast with adjacent organs,inaccurate segmentation boundary caused by blurred boundary with adjacent organs,and loss of segmentation results of non-connected liver region.The algorithm first preprocesses the image,and then uses the LiBlockNet model composed of convolution blocks which uses different convolution kernels to extract different input image scales features fused as the output of the next layer to achieve segmentation.to realize liver segmentation.2)Proposing a pancreatic CT sequence image segmentation method based on dynamic ROI region and VGGU-Net,aiming at improving the problems of irregular shape,small volume and obscure boundary of pancreas,such as low segmentation accuracy and complicated segmentation methods.The method consists of two steps: location and segmentation.In the localization part,the relative position of the liver,kidney,spleen and pancreas is located according to the relative position of the four organs which means finding the three centers of the liver,spleen and kidney in the whole sequence image,then using these three centers to locate the ROI of the pancreas.In the segmentation part,the pancreas region in the ROI is segmented by the proposed VGU-Net model and then realizing the whole algorithm.