基于模糊C均值自动随机游走算法在脑肿瘤分割中的应用

Fuzzy c-means random walks algorithm in brain tumor segmentation

目的:经典随机游走算法是使用最为广泛的交互式图像分割方法之一,分割精度较高,但是该算法执行过程需要人工参与,由医生手动勾选种子点,这在临床使用中是比较麻烦和耗时的,在一定程度上降低了该算法的效率。针对该问题,本文期望提出一种自动的随机游走分割方法。方法:首先使用模糊C均值算法对待分割图像进行聚类,根据聚类的隶属度进行阈值分割得到初步分割结果;再对分割结果进行形态学开放处理,仅保留初步分割结果的主要区域,并将处理后的区域中的所有像素点作为随机游走算法的种子点,利用随机游走算法对图像实施分割,得到最终分割结果。本文结合模糊C均值算法和形态学处理,实现了自动的随机游走。使用改进后的算法对MR图像中的脑肿瘤和水肿区域实施自动分割,验证本文算法的有效性和准确性。结果:本文方法在实现自动化分割的同时,其分割精度明显优于模糊C均值算法的分割结果。结论:本文提出的改进方法能准确地自动分割出目标区域,并且较模糊C均值方法的分割结果有显著性提高。

Objective The typical random walks algorithm is one of most popular interactive segmentation methods, with a high precision. However, the random walks algorithm needs to select the initial seed points manually, which is usually tedious and time-consuming in clinical, lowering the efficiency. An automatic random walks algorithm is proposed in this paper to solve these problems, Methods The images to be segmented were firstly clustered by fuzzy c-means （FCM） algorithm. The primary segmentation results were obtained by threshold segmentation based on the membership of the cluster. And then the segmentation results were processed by morphological processes to preserve only the main segmentation regions of primary segmentation results. All the pixels in the processed segmentation regions were taken as the seed points for the random walks algorithm to get the final results. This algorithm, combined with FCM algorithm and morphological process, achieved the automatic random walks. Finally, the proposed algorithm segmented the brain tumor and edema in magnetic resonance images to verify its affectivity and accuracy. Results The proposed method could achieve the automatic segmentation, and the segmentation results of proposed method were more precise than those of FCM algorithm. Conclusion The results show that the proposed method can segment the target region automatically and precisely, significantly improving the segmentation accuracy.