Research On Key Segmentation Algorithm Of Tissues From CT/MRI Images

The development of medical imaging and modern computer technology brings great opportunity to computer-assistant diagnosis.For clinical doctors,Computer Tomography(CT)and Magnetic Resonance Imaging(MRI)provide effective assistant function in disease diagnosis,surgery and postoperative evaluation.Early diagnosis can effectively detect diseases in hidden stage and reduce the treatment cost for patients.Medical image segmentation techniques can be employed for clinicians to extract regions of interest as well as 3D reconstruction and visualization,which is important for quantitative analysis and preoperative surgery evaluation.Therefore,how to obtain fast and accurate segmentation results becomes main problem in modern hospital.The thesis focuses on four types of diseases which cause serious damage to human body(atherosclerosis?lung cancer?multiple sclerosis?alzheimer's disease).The thesis proposed fast and accurate segmentation methods with the aim of satisfying the clinical diagnosis of surgical planning,early prediction of diseases,even the postoperative recovery treatment.The thesis presented a fast and efficient segmentation algorithms to satisfy the clinical diagnosis of the surgery planning,and received good results.The thesis consists of the following:(1)A vessel intensity prior information-based level set method for carotid artery segmentation is proposed.Carotid artery includes common carotid artery(CCA),internal carotid artery(ICA),external carotid artery(ECA)and the carotid bifurcation(CA).Carotid artery segmentation is challenging due to the low contrast between carotid artery and surrounding soft tissues,the variation in artery size and shape and image noisy.We propose a vessel intensity prior information-based carotid artery segmentation method.The aorta is first segmented by a region growing-based approach.The information of the aorta is used to construct an intensity-based model which can be taken as a constrained term of our level set method.The carotid artery are segmented by evolving the level set contour.(2)In this thesis,an fractional differentiation enhancement based method is proposed for lung vessel segmentation.Lung vessel segmentation is challenging due to the low contrast between lung vessel and other lung field region tissues,and the variation in vessel size and shape.We propose an fractional differentiation enhancement based method for lung vessel segmentation.Firstly,the lung field is automatically extracted using mathematical morphology and region growing method.Secondly,small vessels are enhanced by the fractional order differential operators.The final vessel tree are extracted by a local threshold segmentation.This method is evaluated using the VESSEL12 framework available online,twenty chest CT scans are used for evaluation purposes.We compare our method against other methods,the results demonstrated that the proposed method is fast,robust and efficient.(3)A robust energy minimization algorithm for MS-lesion segmentation is proposed.Multiple Sclerosis is a chronic and inflammatory disease,fast and accurate MS lesion detection is very important for radiologists to provide effective treatment.In this thesis,we propose a robust energy minimization algorithm for MS-lesion segmentation.The energy minimization is proposed for seeking the optimal segmentation result of lesions and white matter.Then post processing operation is used to extract the lesion regions.(4)A multi-atlas registration-based level set fusion method for hippocampus segmentation is proposed.Hippocampus segmentation is challenging due to shape variability,low contrast between hippocampus and its surrounding tissues.We propose a multi-atlas registration-based level set method.This method consists of two parts:image registration and level set evolution.Our work focus on the latter.Labels are obtained by multi-atlas registration,the obtained labels are merged into our fusion term.Label fusion is achieved by combing the three terms:label fusion term,image data term,and regularization term to acquire the final hippocampus.We evaluated our method on the data from MICCIA 2012 Multi-Atlas Labeling Challenge.The dice of our method is 0.85,which is higher than other fusion methods.The surface of our results is more regular than others.