Abstract: Diffusion Weighted Image(DWI) does not contain signal bias caused by thermal noise but also outliers introduced by physiological noise. The Least Square(LS) estimator, the classic diffusion tensor estimator in the estimator of Diffusion Tensor Imaging(DTI), produces the optimal result in diffusion tensor on the condition that the distribution of the error is Gaussian, while it cannot lead to a robust result if outliers exist. According to the problem in LS method, this paper presents robust MM estimator to estimate the diffusion tensor, which has the advantages of both high breakdown point and high efficiency under normal errors simultaneously. MM estimator uses simple initial estimates as input and two step of M estimator is performed subsequently. Through applying MM robust estimator in the synthetic and real data, a more robust and more effective result can be produced in diffusion tensor estimator.

Key words: Diffusion Weighted Image(DWI), Diffusion Tensor Imaging(DTI), robust MM estimator, Least Square(LS) method

摘要： 在扩散加权图像中存在由热噪声产生的高斯分布偏差和生理噪声产生的异常点，最小二乘(LS)法对于高斯分布偏差具有较好的估算效果，但是对异常点不稳健。为此，采用稳健MM估计方法对扩散张量成像(DTI)数据进行张量估算，将高失效点算法的估计结果作为初始估计值，进行两步M估计。模拟数据与真实数据的实验结果表明，该估计方法具有较好的稳健性，并能有效估算扩散张量。

关键词: 扩散加权图像, 扩散张量成像, 稳健MM估计, 最小二乘法

CLC Number: 

• TP306