一种精确提取三维脑图像对称面的方法

A robust and accurate symmetry plane extraction method from 3D brain images

在脑图像的结构和功能研究中,提取三维脑图像中间矢状位对称面(MSP)起着很重要的作用。本文提出一种自动分步精确优化的计算方法,通过结合子空间粗略定位及原始空间小范围精确优化计算,提高计算精度,可有效简化初始对称面的选择,并使用Powell优化算法高效计算出对称面。采用本算法对多种条件的模拟数据和真实数据进行性能测试,对于不同模态、噪声水平和不均匀场等数据均获得了较其他代表性方法更好的提取结果。

The automatic extraction of the mid-sagittal symmetry plane（MSP） in 3D brain images is significant for many anatomical and functional studies.This article proposed an automatic multistage optimization method to detect the final symmetry plane.The rough computation in the sub-scale image and the shrinking of searching scope in the original-scale image dramatically and robustly were used to improve the accuracy.The way of selecting a valid initial plane was simplified,and the local-based Powell method was used to compute the symmetry measure efficiently.The algorithm was validly tested both on synthetic and real human brain images,and gave the better results than the representative methods on different modalities,noise levels and non-uniformities.