S-Space College of Engineering/Engineering Practice School (공과대학/대학원) Program in Bioengineering (협동과정-바이오엔지니어링전공) Theses (Ph.D. / Sc.D._협동과정-바이오엔지니어링전공)
Advanced segmentation algorithms using numerical model and graph theory-Application to the lung images-
수치 모델과 그래프 이론을 이용한 향상된 영상 분할 연구 -폐 영상에 응용-
- 공과대학 협동과정 바이오엔지니어링전공
- Issue Date
- 서울대학교 대학원
- chronic obstructive pulmonary disease (COPD); computed tomography; multi-organ segmentation; thoracic cavity; pulmonary artery and vein decomposition; two level minimum spanning tree constructions
- 학위논문 (박사)-- 서울대학교 대학원 : 공과대학 협동과정 바이오엔지니어링전공, 2016. 2. 김희찬.
- This dissertation presents a thoracic cavity segmentation algorithm and a method of pulmonary artery and vein decomposition from volumetric chest CT, and evaluates their performances. The main contribution of this research is to develop an automated algorithm for segmentation of the clinically meaningful organ. Although there are several methods to improve the organ segmentation accuracy such as the morphological method based on threshold algorithm or the object selection method based on the connectivity information our novel algorithm uses numerical algorithms and graph theory which came from the computer engineering field. This dissertation presents a new method through the following two examples and evaluates the results of the method.
The first study aimed at the thoracic cavity segmentation. The thoracic cavity is the organ enclosed by the thoracic wall and the diaphragm surface. The thoracic wall has no clear boundary. Moreover since the diaphragm is the thin surface, this organ might have lost parts of its surface in the chest CT. As the previous researches, a method which found the mediastinum on the 2D axial view was reported, and a thoracic wall extraction method and several diaphragm segmentation methods were also informed independently. But the thoracic cavity volume segmentation method was proposed in this thesis for the first time. In terms of thoracic cavity volumetry, the mean±SD volumetric overlap ratio (VOR), false positive ratio on VOR (FPRV), and false negative ratio on VOR (FNRV) of the proposed method were 98.17±0.84%, 0.49±0.23%, and 1.34±0.83%, respectively. The proposed semi-automatic thoracic cavity segmentation method, which extracts multiple organs (namely, the rib, thoracic wall, diaphragm, and heart), performed with high accuracy and may be useful for clinical purposes.
The second study proposed a method to decompose the pulmonary vessel into vessel subtrees for separation of the artery and vein. The volume images of the separated artery and vein could be used for a simulation support data in the lung cancer. Although a clinician could perform the separation in his imagination, and separate the vessel into the artery and vein in the manual, an automatic separation method is the better method than other methods. In the previous semi-automatic method, root marking of 30 to 40 points was needed while tracing vessels under 2D slice view, and this procedure needed approximately an hour and a half. After optimization of the feature value set, the accuracy of the arterial and venous decomposition was 89.71 ± 3.76% in comparison with the gold standard. This framework could be clinically useful for studies on the effects of the pulmonary arteries and veins on lung diseases.