Three-dimensional head vessel segmentation based on CT and CTA registration

• Main Authors: CHEN,YOU-JUN, 陳宥均
• Other Authors: LIN,WEI-YANG
• Format: Others
• Language: zh-TW
• Published: 2019
• Online Access: http://ndltd.ncl.edu.tw/handle/3aqsb4

碩士 === 國立中正大學 === 資訊工程研究所 === 107 === In this paper, we present a three-dimensional head vessels segmentation based on CT and CTA registration to solve the problem of registration failure resulting in removed vessels. Our proposed method consists of two major parts, including seeds screening and vessel tracing. For seeds screening , we use the original CTA image and its registration result to screen the seeds on the vessels. As for vessel tracing, we use the screened seeds as the origin of each vessel and two-dimensional cross-section analysis for segmentation of the connected vessels. In our experiments, we collected two different datasets from Dr. Shao-Jer Chen, Department of Dalin Tzu Chi hospital. In conclusion, the accuracy of seeds screening is 90%. Most non-vessels seeds could be screened out. The average Recall of our vessel segmentation is 0.8936. Although in some of our experiments, if there are residual seeds on the bone, a small piece of bone will grow, resulting in decrease precision. But compared with the method of direct subtraction, our method is still better, which can be improved by up to 17%.