基于模拟回火退火的DT-MR图像平滑和估计

Smoothing and Estimating DT-MR Images Based on Simulated Tempering Annealing

在采用高斯-马尔可夫随机场(GMRF)对扩散张量磁共振成像(DT-MRI)的原始图像进行平滑和估计时,要根据Bayes准则对图像灰度进行最大后验(MAP)估计.为了避免陷入局部最小的“陷阱”和减小计算量,MAP估计采用了模拟回火退火方法(STA).通过对未加权图像和不同梯度脉冲下的加权图像(共7幅)同时进行平滑和估计.结果表明,基于STA对图像进行平滑和估计能够大大减少噪声影响,从而在图像信噪比很低的情况下仍能保证张量场完全正定.把本方法的实验结果与传统模拟退火(SA)方法的结果进行比较,表明基于STA的方法能够更加有效地消除噪声影响,减小计算量.

To smooth and estimate the raw data of DT-MRI based on Gaussian-Markov random field （GM- RF）, maximum a posterior （MAP） estimation was resorted to estimate the parameters according to the Bayes＇ theorem. To escape the ‘cheat’of the local optimization and alleviate the burden of computing, simulated tempering annealing （STA） was adopted. An experiment was designed to smooth and estimate simultaneously one unweighted and six different gradients weighted images. The results of the experiment verify the utility of the mentioned method by the greatly decreased noise effect and the ensured positive semi-definite characteristics of the tensor field even if the signal-to-noise ratio（SNR） is very low. Compared with the traditional SA method, the method adopted can more efficiently decrease the noise effect and the computing burden.