基于共轭梯度的大尺度医学图像非刚性配准算法

Non-rigid Registration Based on Conjugate Gradient Method for Large Deformation of Medical Image

人体胸腹腔中大量非刚性软组织器官由于呼吸、脏器蠕动和体位变动等会产生大尺度非线性形变,粘性流体配准将图像的形变模拟为流体的流动过程而适宜于大的形变.针对粘性流体配准算法应用超松驰迭代求解离散化偏微分方程组耗时量大,且其最优松驰参数难以获得的问题,使用基于不需要预先估计参数的共轭梯度迭代求解形变参数,实现粘性流体配准.该方法所需存储量小,且具有步收敛性、稳定性高的优点,实验证明该方法在不损失配准精度的前提下提高了配准速度.

Due to respiration, organ motility, postural changes of non-rigid soft tissues in human thoracic and abdominal cavity, the large-scale nonlinear deformation is produced. Registration based on viscous fluid simulates image deformation as the process of fluid flowing and is adaptive for large deformation. Viscous fluid registration algorithm employs Successive Over Relaxation（SOR） iterative methods to solve time-con- suming discrete partial differential equations capacity and the method is difficult to obtain the optimal re- laxation parameters. In the paper, conjugate gradient iterations with no need for anticipate parameters is used to achieve viscous fluid registration with a small amount of memory required, step convergence and ad- vanced stability. Experiments show that the method improves registration rate without losing registration accuracy.