摘要： 脑肿瘤分割是计算机辅助脑病诊断的基础。为提高脑肿瘤分割精度，提出一种融合高斯混合模型的测地线脑肿瘤分割方法。根据相邻2个像素点间互相到达时间构造离散且带有权重的网格图，通过高斯混合模型估计每个像素点属于目标物体的罚度，并融合高斯混合模型的概率密度差异表示区域属性与边缘属性构成能量函数，利用快速最短路径算法求解前景与背景间的测地线距离，并根据该距离最小化能量函数，得到脑肿瘤的分割结果。利用10组脑部核磁共振图像数据对算法进行评估，结果表明，该算法分割结果与金标准的重叠率在0.60~0.85之间，可有效避免局部最优解的情况，对非匀质区域具有较好的分割效果。

关键词: 测地线距离, 高斯混合模型, 期望值最大化算法, 最短路径快速算法, 脑肿瘤图像, 图像分割

Abstract: Brain tumor segmentation plays an import role in computer-aided diagnosis. In order to improve the precision of brain tumor segmentation, this paper proposes a brain tumor segmentation method using geodesic combined with Gaussian Mixed Model(GMM). A discrete weighted graph which the edge-weights present the arrival time between the neighbors is constructed, the penalization that the pixel belongs to the target object is computed. The cost energy function consists of the region and edge terms. The geodesic distance is computed by using the shortest path fast algorithm. The brain tumor is detected by the minimizing the energy function according to the geodesic distance. The algorithm is tested on ten sets of MR image datasets and the overlap values between the segmentation result of the proposed algorithm and the ground truth is 0.60~0.85. Experimental results illustrate that the algorithm can reduce the local minimization and it has high efficiency in heterogeneous regions.

Key words: geodesic distance, Gaussian Mixed Model(GMM), Expectation Maximization(EM) algorithm, the shortest path fast algorithm, brain tumor image, image segmentation

中图分类号: 

• TP911.73