Research On Medical Image Segmentation And Visualization

With the progress of computer technology and medical imaging technology,the medical imaging data available to doctors is becoming more and more abundant,and the dependence of modern medical diagnosis on medical imaging is increasing,and medical imaging postprocessing plays an increasingly important role in the clinical application and research of modern medicine.Medical image segmentation and visualization is an important part of medical image post-processing and a research hotspot in the field of medical imaging.Medical image segmentation and visualization is also the basic function of medical image application software,and is the basis of modern medical diagnosis,surgical guidance and so on.Medical image data is large,manual segmentation is time-consuming and subjective,accurate and fast medical image post-processing technology can better play the diagnosis role of medical image and improve the efficiency of doctors.This thesis mainly focuses on the segmentation methods of human brain magnetic resonance(MR)image and cardiac computed tomography angiography(CTA)image,and studies the curved planar reformation(CPR)visualization method for coronary artery based on CT image.In the segmentation of brain MRI images,the main research is to implement a skull stripping algorithm and use it for segmentation of the perivascular spaces(PVS).Firstly,the classic BET skull stripping algorithm is studied.Aiming at the problem that BET can not effectively remove the skull of the experimental data,a new skull stripping method based on affine registration is implemented.Then on the basis of this method,a segmentation method based on tubular structure enhancement for perivascular spaces is proposed,which leads to the automatic segmentation of the perivascular space in the white matter area of the top brain.In the segmentation study of cardiac CTA images,based on the coarse segmentation of the coronary arteries and the centerlines calculated from them,a precise lumen segmentation method based on the cylinder model is proposed.This method can get accurate vessel lumens and obviously improves the problem of lumen over-segmentation caused by the coarse segmentation process,so that the segmented vessels can be used to quantize and assess the lumen stenosis.In view of the visualization of coronary arteries,the thesis mainly studied a variety of curved planar reformation(CPR)algorithms and elaborated on their principles.First,the author studied traditional CPR algorithms including stretched CPR and straightened CPR,so that the whole blood vessel can be displayed on one image.Secondly,the author improved and implemented a spherical CPR algorithm.By using the maximum density projection,the method can display the entire coronary tree on the tree’s envelope surface,as well as in the derived spherical image and planar image,which enhances the visualization intuitiveness.In the generation of the envelope surface image,a more accurate coronary tree envelope surface is obtained by introducing the heart surface,which makes the display of blood vessels more accurate;in generating the derived images,a appropriate rotation matrix is designed to reduce the deformation of the blood vessels.Real system verification shows that these three CPR methods are very practical.