Research Of Algorithms For CT Image Segmentation Based On Sequence Images Analysis

Human abdomen is one of the parts of most tissue/organs,most complex structure and relatively high morbidity,and its symptoms are various.As the development and popularization of the application of computer in every walk of life,aided diseases diagnosis technology has made great progress.The location and extraction of the profiles of tissue/organ is an important prerequisite of computer-aided liver diagnosis and surgery.However,images obtained from abdomen's CT-scan-sequenced images are subject to bulk effect tissue movement,noise and the low resolution ratio,etc.,hence make great challenge to separating the tissue/organs in the CT-scan-sequenced image.For the extraction of liver and aortic aneurysm in abdomen's CT-scan-sequenced images,this thesis proposes two effective segmentation algorithms separately.The main idea of them is to utilize the features of accordance,similarity and mini-variance of organs/tissue area in the CT-scan-sequenced image sequence.It can be detailed as follows:(1)For the liver segmentation of CT-scan-sequenced images,an approach fusing prior constraints among multi-section sequences is proposed.It considers the relevance of image sequences with the idea of accordance,and general method can be divided into two parts: the first one proposes the liver-segmentation method based on prior constraint among CT image sequences;and the second one fuses the input CT image section,vertical plane and liver-segmentation profile by voting,obtaining the modified result of 3-dimensional liver segmentation.The superiority of the proposed liver-segmentation method is the no requirement of any known training sample for constructing the archive of liver shape models,which has strong generalization ability,can be extended to the application of the segmentation of non-labeled data,and can provide training samples for cold boot method requiring train data.(2)For the segmentation of aortic aneurysm in CT images,firstly,an algorithm based on the threshold segmentation method extracting aortic lumen and extracting aortic profiles by means of graph cuts model that based on shape.Then,the aortic profiles are excluded from aortic lumen areas to obtain the result of hollow aortic aneurysm.And finally,3-dimensional level set algorithm is introduced to modify the over-flat phenomena of aorta hump part segmentation caused by 2-dimensional segmentation profile sequences.To combine with clinical applications,the surface rendering algorithm based on mobile cube is used to reconstruct the 3-dimensional hollow model of aortic aneurysmal wall,which makes it easier for the later 3-dimensional model print and the application of cardiac surgery extracorporeal simulating experiment.