Computer aided detection of pulmonary embolism in computed tomography angiography images

PE, one of the circulatory system problems, is the clinical result at the end of that the lung artery or one or a few branches of them are bunged up by clot of blood. Besides it is a hard disease to diagnose, early diagnosis and treatment of it can save lives. The detection by CAD of PEs which can be diagnosed from CTA Images has been carried out at this thesis work. There are two different approaches to detect PEs by CAD in literature. The first one is to detect PEs by indicating lung vessel tree. The second one is to detect PEs by tobogganing method without indicating lung vessel tree. The lung vessels firstly have been segmented with high intensity values at the works done with the first method. However, because the vessel structure will be damaged in PE regions, secondly to continue the vessel structure smoothly, PEs have been distinguished from the tissues with the same low intensity like them in their regions according to their sizes and shapes. Then, vessels and PE have been connected by tracking method or by looking at region, shape and volume properties of the vessels. However, there is no a constant shape of PEs. It can be anywhere in the vessel tree regionally. It isn?t be considered as PE under a certain size, though it can be in different size above it. Besides, if the vessels are bunged completely, it will be hard to do tracking so that it seems unconnected in CTA images. So the performances of the methods in previous works are lower. A new method has been developed at this thesis work. firstly, lung segmentation has been carried out. Secondly, by making use of the unchangeable properties of anatomical structure in CTA images, some reference points (RP) have been detected in order to use at vessel segmentation. Due to these RPs, truncus, left-right pulmonary artery, lobar segment vessels have been segmented with the PE. RP and PE have been distinguished from the other tissues having the same intensity and from the lung and heart disease tissues except for PE. Subsegment vessels have been detected with PE by looking at the differences of intensities in lung region. Lastly, lung vessel tree has been detected by connecting all of the vessels. Thirdly, PEs have been detected from the vessel tree, by examining the intensity differences in a way of inside regions of vessel and removing the components under a certain size from the image. The results at the end of these processes have been analyzed by the performance measures which are used in medical image evaluation and compared to the previous works. It has been seen that the performance of the method for PE detection at this thesis is better than the ones in literature and it gives rather enough results to the surgeries medically.