Accelerated Diffeomorphic Non-rigid Image Registration with CUDA Based on Demons Algorithm

摘要：

Image registration is an indispensable process in thedetection of brain structural and anatomical abnormities.Inverse-consistency, topology preserving and real timeapplication are essential to provide accurate deformation fieldsfor statistical analysis of brain variability. Unfortunately, theprevious algorithms lacked of these features. We present aregistration method by adapting the optimization procedure on aLie pseudo-group so that the generated deformations are smoothwith low energy of the deformation. In order to speed up theperformance of registration method, we have implemented it onNvidia 8800GTX GPU with Computer Unified DeviceArchitecture (CUDA) platform. Experimental results indicatethat large deformation compatible and topology preservingdemonstrated through experiments on synthetic data and realbrain CT images. Additionally, a faster runtime among theavailable GPU-based registration implementations againststraight forward CPU version was achieved.

作者: Yufeng HUANG Tong TONG Wei LIU Ya FAN Huanqing FENG Chuanfu LI

作者单位: Department of Electronic Science and Technology University of Science and Technology of China Hefei, Medical Image Center, the First Affiliated Hospital Anhui Traditional Medicine College Hefei, Anhui

会议类型: 国际会议

会议名称: The 4th International Conference on Bioinformatics and Biomedical Engineering(第四届IEEE生物信息与生物医学工程国际会议 iCBBE 2010)

会议地点: 成都

会议语种:英文

页码: 1-4

在线出版日期: 2010-06-18（万方平台首次上网日期，不代表论文的发表时间）