基于改进空间模糊聚类的DTI图像分割算法  

DTI Image Segmentation Algorithm Based on the Improved Spatial Fuzzy Clustering

作　　者：刘絮雨 张相芬[1] 马燕[1] 李传江[1] 杨燕勤[1] Liu Xuyu;Zhang Xiangfen;Ma Yan;Li Chuanjiang;Yang Yanqin(College of Information,Mechanical and Electrical Engineering,Shanghai Normal University,Shanghai 200234,China)

机构地区：[1]上海师范大学信息与机电工程学院,上海200234

出　　处：《中国生物医学工程学报》2018年第4期394-403,共10页Chinese Journal of Biomedical Engineering

基　　金：国家自然科学基金(61373004);上海师范大学校级基金(A700115001005;Sk201220)

摘　　要：针对模糊C均值(FCM)聚类算法初始聚类中心选择的随机性和噪声的敏感性等问题,提出一种基于改进空间模糊聚类的图像分割算法来分割人脑DTI图像。使用局部密度核函数和中心距离函数精确选取初始聚类中心,不仅可以解决因聚类中心随机选取造成的聚类效果不稳定的问题,而且还可以使目标函数迅速收敛,提高分割效率;通过将正态分布空间信息融入模糊隶属度函数,能减小图像噪声以及人为因素对分割结果的影响。用该方法与FCM、SFCM方法对人脑DTI数据进行分割,以评价算法的聚类效果。实验对美国明尼苏达大学生物医学功能成像与神经工程实验室提供的58例DTI数据、3例FA参数图像以及6例迭加过噪声的人脑DTI图像进行分割,结果表明:该算法分割系数最高,可达到0.984 1;在同一图像中,该算法在划分系数上比FCM最高提升20.2%,并且在划分熵上比SFCM最高下降19.8%;该算法目标函数平均迭代次数为32,较FCM的52次与空间FCM的76次有明显降低。实验证明,该算法能够准确、快速地分割出重要目标,且对图像噪声不敏感。Aiming to resolve the problems of initial clustering selection randomness and noise sensitivity of fuzzy C means algorithm,this paper proposed an image segmentation algorithm based on the improved spatial fuzzy clustering to segment the DTI image of human brain. In this paper,we used the local density kernel function and the center distance function to select the initial clustering center accurately,which not only solved the problem of clustering effect instability caused by random selection of cluster center,but also made the objective function converge quickly,and improved the segmentation efficiency. Moreover,the proposed algorithm reduced the influence on the segmentation result caused by image noise and human factors by integrating normal distribution spatial information into fuzzy membership function. We segmented DTI data of human brain with the proposed method,FCM and SFCM to evaluate the clustering effect of the algorithm. In the experiments,following data were employed, including segmented 58 cases of DTI data provided by the University of Minnesota Biomedical Functional Imaging and Nerve Engineering Laboratory,3 cases of FA parameter images,and 6 cases of iterative noisy human brain DTI images. Results show that the segmentation coefficient of proposed algorithm reached 0. 9841. In the same image,the algorithm obtained the most improvement of 20. 2%than FCM on the partition coefficient,and the most decline of 19. 8% than SFCM on the partition entropy; The average number of iterations of the algorithm was 32,which is significantly lower than 52 of FCM and 76 of SFCM. Therefore,the algorithm can segment the important target accurately and quickly,and the segmentation results are insensitive to image noise.

关 键 词：FCM算法 聚类中心 局部密度 空间信息 DTI图像 

分 类 号：R318[医药卫生—生物医学工程]