一种基于遗传算法的脑MR图像去偏移场模型

A Brain MR Images De-bias Model Based on Genetics Algorithm

由于磁共振图像(magnetic resonance images,MRI)常含有偏移场而影响后继图像分割,针对这种图像的分割,采用Legendre多项式基函数来拟合偏移场,可以去除偏移场对图像分割的影响。当使得恢复图像的信息熵达到最小时,则求得的偏移场最优。在求偏移场的过程中,需要求解基函数的参数,由于传统的梯度下降法易陷入局部最优,为解决此问题,提出将遗传算法引入到参数求解过程中,然而传统的遗传算法不仅时间复杂度高,且易陷入局部最优,为此需对遗传算法进行改进,使得不仅更容易得到全局最优解,且时间复杂度较低。实验证明,该改进算法可以得到精确的偏移场,并可得到准确的分割结果。

Intrascan intensity inhomogeneities are a common source of difficulty for MRI segmentation. We estimate the bias field by Legendre polynomials. The bias field could be the best when we get minimum entropy. It needs to work out parameters of the base function in the process of finding bias field, but conventional methods such as gradient-descent method often find local best. To find global best, we present genetics algorithm to find best parameters to estimate the bias field, however the result was not satisfying. Then we make some modification of genetics algorithm to make it easier to find global best. Experiments on the segmentation of brain magnetic resonance images show our modification can achieve optimal bias field and accurate segmentation results.