结合空间域模糊聚类与DRLSE模型的颅脑出血CT图像分割  

Segmentation of Intracranial Hemorrhage CT Image Based on Spatial Fuzzy Clustering and DRLSE Model

作　　者：王玉梁[1] 金福江[1] 高宏志[2] 黄彩虹[1] 

机构地区：[1]华侨大学信息科学与工程学院,福建厦门361000 [2]福建医科大学第二附属医院,福建泉州362000

出　　处：《控制工程》2016年第3期353-360,共8页Control Engineering of China

基　　金：国家自然科学基金(61273069);福建省自然科学基金(2015J01383)

摘　　要：针对颅脑出血CT图像中存在出血病灶不明显、边界不规则、不连续及含有高噪声现象,提出一种结合空间域模糊聚类与DRLSE模型算法,用于脑CT图像出血病灶区分割。首先,采用基于空间域信息的模糊C-均值聚类算法对出血CT图像初始聚类分割,然后,利用模糊聚类结果对距离规则化水平集演化(Distance Regularized Level Set Evolution,DRLSE)模型初始化。新算法引入了图像空间域信息,不需要手工初始化,DRLSE模型使曲线精确、稳定的演化。实验结果表明,与传统的FCM算法、阈值分割算法相比,该算法具有更好的分割效果、更快的分割速度、更强的鲁棒性和抗噪性。In view of the problem that intracranial hemorrhage usually appears inconspicuous, irregular, discontinuous and noisy on CT image, an algorithm integrated with spatial fuzzy clustering and DRLSE model is proposed to facilitate brain CT image segmentation. Firstly, a fuzzy C-means clustering algorithm based on spatial information is employed to segment intracranial hemorrhage CT image. Secondly, level set function is initialized based on the results of the fuzzy clustering. The new algorithm introduces spatial information of CT image and does not need re-initialization, DRLSE model makes curve evolve steadily and accurately. The experiment results show that the new algorithm has salient advantages over classical FCM segmentation algorithm and threshold segmentation algorithm with respect of segmented effect, computational efficiency, robustness and noise immunity.

关 键 词：颅脑出血CT 病灶不明显 不规则 不连续 高噪声 DRLSE模型 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]