The Study Of Core Image Processing Technology On Coronary Artery Stenosis Computer-Aided Diagnosis

Inducing some symptoms such as myocardial ischemia and low blood oxygen saturation, coronary artery stenosis lesions are considered as major cause of myocardial infarction and even death. Using computer-aided diagnosis system based on medical images, the arteriostenosis can be identified in a more efficient, autonomous and rubust way, contributing to provide reliable evidence for early diagnosis and reduce manipulation strength by manual operation, misdiagnosic and missed diagnostic rate. As a result, the reseaches on related technology have aroused widely attention and become hotspots.Through studying on research status, it finds that the diagnosis systems using traditional algorithm have achieved significant progress in this field. However, there exsit some unsettled issues on the critical processing steps, such as coronary segmentation, skeleton network extraction, stenosis lesion grading and classification diagnosis. These problems to be solved are listed as follow:(1) Due to the difference of contrast agent filling situation and imagery environment, severe intensity inhomogeneity and blur phenomenon appear on the vessel lumen. While most modeled coronary artery segmentation methods only depend on intensity features without necessary priori geometrical information, leading to the difficulty on distinguish vessel and non-vessel points accurately.(2) During the calculation of traditional medialness measuring function based on Gaussian convolution, neighborhood information spreading and interference appear due to overestimate scale and low-pass smoothing features, deteriorating numerical accuracy.(3) Due to the influence of individual difference and cardiac motion, the topology configuration of coronary artery has characteristic of complicated variety, which brings great challenge to algorithm topological adaptability. Moreover, some problems, such as low pixel-wise skeleton extraction precision and poor skeleton network topology homotopy retention ability, are significant to be solved for coronary artery stenosis measurement.(4) The current used quantitative indicators of stenosis degree, such as NASCET and ECST, cannot reflect local property of stenosis lesions. Furthermore, the most existing researches only concentrate upon stenosis grading while very few prospective studies focus on stenosis lesion classification.Through the research and analysis on aforementioned problems, this thesis puts forward some novel solutions and algorithms. The main innovations are listed below:(1) Based on connected component analysis and shape identification, improved local shape analysis is proposed. the novel method can effectively refuse non-vessel structures with block and blob shape, vessel-like structures introduced by image noise and tiny vessel structures, ensuring high segmentation accuracy and anti-noise performance.(2) Considering the coronary artery tree as research object, this thesis contributes to study on topology adaptive skeleton network extraction. The medialness measuring function based Gaussian affinity voting is proposed to eliminate interference between adjacent vessels; The skeleton initialization based on level-set graph is proposed to initialize multi skeleton curves; The vessel skeleton tracking based on layered multiple hypothesis is proposed to improve accuracy on vessel bifurcation, achieving simultaneous extraction of skeleton location and vessel radius; The coronary artery skeleton network extraction based on stretching active contour is proposed to export subpixel-wise skeleton structure. Moreover, topology check and network reconfiguration strategy are proposed to keep homotopic between extracted skeleton network and original vessel shape.(3) Referencing NASCET and ECST indicator, the novel quantitative indicator of stenosis lesions is proposed, realizing quantitative analysis and grading diagnosis. Furthermore, though multi morphology index such as lumen circularity, eccentricity and tortuosity measurement, the diagnostic criteria of stenosis grading and classification is proposed.