Research On Automatic Liver Segmentation For 3D CT Images

Liver surgery is one of the major treatments for common liver diseases,the accurate separation of liver tissue from CT images is an important step in computer-aided diagnosis and computer-assisted liver surgery,and usually manual operated by experienced medical experts,which is inefficient and limited by experience.Therefore,it is of great practical significance to study the automatic liver segmentation method.This thesis is based on previous studies,using several steps for liver segmentation:first get preliminary segmentation by atlas registration,then build the classifier in the border area of the liver and finally classify the voxels thus obtain final segmentation.Meanwhile,using effective preprocessing to remove irrelevant interference and achieve automatic segmentation.Specifically,this thesis do the relevant research and experiment in these aspects:preprocessing the CT images,average atlas segmentation,convolution neural network。The main work of this thesis is as follows:In the preprocessing step,a series of effective preprocessing operations are designed by using the anatomical prior knowledge,such as regional growth and threshold filtering,including normalizing the cross section deflection angle of the image,searching for the range of the liver,threshold filtering of the image.Preprocessing help to determine the region of interest for subsequent segmentation,align the position of the liver in the image and remove some irrelevant information of the image,reduce the error and of improve the quality of the liver segmentation.Using atlas segmentation to obtain the primary segmentation,registration between any two atlas using B-spline transform to get deformation fields and then obtain the average atlas in the training set.The primary segmentation results of the liver were obtained by affine registration and B-spline registration,and the segmentation results were analyzed and evaluated.Based on the results of average atlas segmentation,we proposed using the deep convolution neural network to classify voxels in border area of the liver,he corresponding deep convolution neural network is designed.The training data is extracted by the sliding window,and train the network by the training data,the final results can be obtained by morphological post-processing and smoothing operations.The results of segmentation are analyzed and evaluated,and compared with other algorithms in the literature,which shows that the algorithm has good accuracy and robustness.