The Research On The Key Technology For Segmentation On CT Images Of Organs

Image segmentation techniques are the basic operations of many advanced imaging processes, mainly focusing on how to extract interest areas from input original images. It has significant impact bearing in the imaging process areas. This thesis with related researches conducted under it is mainly set to solve two key medical-engineering image segmentations for liver entities and coronary arteries.This thesis is based on current widely used segmentation techniques to summarize their advantages and disadvantages, such as region growing, morphological erosion and expansion, graph theory, mean-value drifting, active contours. By comparing the advantages and shortages of these methods, we conclude that it’s impossible for us to use a single or simple combination of these methods to segment images with complex foregrounds and backgrounds like human liver organs and coronary arteries we target in this thesis. Therefore, this thesis organically and specifically analyze the immanent properties of images and liver organs, combine imaging operational flow and related algorithms and get a more advanced systematic segmentation method to segment and extract CT images.There still exist great challenges due to the inherent complex properties owned by organ CT images. In order to solve these problematic challenges, this thesis introduces two segmentation techniques to segment liver entities and to extract coronary arteries which are based on the correlations between adjacent images of CT series and connected domain respectively.The specific target of our work is to extract liver entities and coronary arteries from series of abdomen CT images. Compared with traditional methods aforementioned which are only feasible in processing a single image every time, our technique introduced in this thesis can extract or segment interest areas, including both 2D and 3D, from a panorama or portions of a whole images series. The reason we are able to make it from a 3D scale is that the priori knowledge of human liver is used as a guiding model throughout our research. According to careful observation, any two adjacent liver CT images concretely have very small variations in appearance change of liver areas, which take charge of most portions of corresponding images and have almost identical pixel values. By centering on this, a semi-automatic segmentation method with more advanced features is realized based on the correlations between any adjacent CT images. In order to tackle the mutations caused by liver foregrounds, this semi-auto technique makes use of the prior knowledge in the similarities between up-down adjacent images to combine region growing, level set and some manual interventions. The experimental results show that this method can reliably and accurately segment liver entities and extract coronary arteries from points of medical standards and time costs, where the accuracy can reach as high as 92%~98% compared with standard manual entities by physicians and the time consumed is in controlling effective range.Besides the liver segmentation, vascular extractions or vascular segmentations are also important basic imaging processes. What this thesis target at about vessels in liver organs include extracting coronary arteries, skeletonize extracted arteries, measurements of tubular radii of arteries and spotting soft plagues and hard plagues. For the convenience of following research work, this thesis combine automatic threshold technique and connected domains to wipe out dross encircling artery tubes and the coronary open point at aorta, which then is used as a basic start point to cut off other organs and tissues having negative impacts on final segmentation. After these processes, an independent coronary aorta and related organizations are extracted. The experiment illustrates that this method can extract good coronary arteries by exactly pinpointing the coronary artery point and effectively eliminating other redundant connected tissues. And experimental results show our method is effective and time-saving.The methods introduce by this thesis has been verified as highly effective and robust after amount of experiments and statistics based on real data, indicating they are possessing significantly practical value.