基于粗配准和互信息的脑部MR图像配准算法

现有的医学图像配准算法一般都存在需要人工介入、配准时间过长等问题.为了寻找快速、精确、鲁棒性强的自动配准算法,在采用主轴矩方法进行脑部MR(核磁共振)图像的初始配准的基础上,提出局部搜索算法对图像求得更精确的配准.实验表明,该方法的配准精度和现有的Powell算法都可以达到亚像素级,但局部搜索方法和Powell算法相比较,平均配准时间大大缩短;即便和采用了主轴矩粗配准的Powell算法相比较,配准效率也提高了一倍左右.主轴矩粗配准算法提高了配准效率,局部搜索算法则保证了配准的精度.

An Algorithm for Brain MR Image Registration Basedon Rough Registration and Mutual Information

The existing algorithms for medical image registration always require human intervention and too much time. In order to develop a fast, accurate and anti-noise automated image registration algorithm, the Cross-Weighted Moments algorithm is applied to initialize the registration of the brain MR images for enhancing the efficiency and a local search algorithm is proposed to achieve more accurate registration. The experiments show that the local search algorithm and the Powell algorithm both can reach the sub-pixel level of accuracy but the local search algorithm can shorten much time. Compared with the Powell algorithm which is combined with the Cross-Weighted Moments initialization, the runing time is nearly half shortened.