基于聚类算法的脑部MR图像分割

摘要

ObjectiveTo explore an improved clustering segmentation algorithm and apply it on the automatic segmentation of brain MR image.Methods First, a novel colorization method is proposed to transform a gray brain MR into a color one and increase the image contrast of anatomical structure. Second, a probability density curve is drawn with gray histogram to split region at intersection points. Finally, the segmented image would be achieved by using the selected centroids in clustering method. Results Different segmentation algorithms are selected to conduct simulation experiment of the brain MR image. Qualitative evaluation results show that the proposed method can enhance contrast of gray matter, white matter and cerebrospinal fluid, and improve image quality. Quantitative evaluation results indicate the improved cluster algorithm generates a higherJaccardcoefficient than others, and has less computation time than other methods.ConclusionTheprobabilistic line which is based on gray histogram can effectively avoid the blindness of the initial centroids selection, and make the segmented results more rapid and accurate. The proposed algorithm has higher clinical application value for the analysis of region of interest.%目的：探讨改进的聚类分割算法，并将其应用于脑部MR图像的自动分割。方法采用彩色编码将灰度图像转换到彩色空间，提高图像各解剖结构对比度；利用灰度直方图绘制概率密度曲线获得各类区域峰值点；将此峰值点作为聚类分割算法的初始聚类中心，达到图像自动分割的效果。结果选用不同分割算法对脑部MR图像进行仿真实验。定性分析表明基于本文分割算法的图像中灰质、白质和脑脊液部分容易辨别，且清晰度更高；定量评估结果显示基于本文分割算法能获得最优的Jaccard系数和最少的平均分割时间。结论基于灰度直方图绘制的概率曲线有效地避免初始聚类中心选取的盲目性，使得分割结果更快速、更准确，在目标分析中具有较高的临床应用价值。