为了提高大脑磁共振图像的分割精度, 提出了一种新的图像分割算法。首先, 分析了常用于大脑磁共振图像分割的高斯混合模型和主动轮廓模型的优缺点, 联合高斯混合模型的概率密度函数和主动轮廓模型的能量函数构造了一个新的能量函数。然后, 利用遗传算法和最大期望算法获取概率密度函数的参数值。最后, 利用水平集方法和梯度下降流法, 对获得的能量函数进行最小化, 从而得到最终的分割结果。与传统方法相比, 本文算法对脑组织中的白质和灰质的分割精度分别提高了6.73%和14.07%。该算法利用像素点的区域信息和概率值驱动主动轮廓曲线的演化, 能有效区分具有相近灰度值的不同区域, 从而提高了大脑磁共振图像的分割精度。

To improve the segmentation precision of brain Magnetic Resonance(MR) imaging, a novel brain tissue automated segmentation method was proposed. Firstly, the merits and demerits of Gaussian mixture model and active contour model used for MR image segmentation were analyzed, and a new energy function was constructed through combining the probability density function of the Gaussian mixture model with the energy function of the active contour model. Then, the genetic algorithm and expectation maximization algorithm were used to get the parameter values of the probability density function. Finally, segmentation results were achieved through minimizing the novel energy function by using the level set method and the gradient descent algorithm. The experiment results clearly indicate that the segmentation accuracies of white matter and gray matter in brain tissue by the proposed method are increased by 6.73% and 14.07%, respectively as compared with that of the traditional methods. By using the area information and probability values of pixel points to drive the active contour curve, the proposed method automatically segments the brain MR image with high enough accuracy and improves the segmentation accuracy of brain MR images.