基于对称信息和主动轮廓模型的脑肿瘤分割系统  

Brain tumor segmentation system based on symmetry analysis and active contour models

作　　者：李丹仪[1] 李卫锋[1] 廖庆敏[1] 

机构地区：[1]清华大学深圳研究生院视觉处理信息实验室,深圳518055

出　　处：《清华大学学报（自然科学版）》2013年第7期995-1000,共6页Journal of Tsinghua University(Science and Technology)

摘　　要：该文提出了一个全自动的脑肿瘤分割系统,主要包括脑肿瘤粗分割系统和细化分割系统。粗分割系统利用脑MRI(magnetic resonance imaging)图像的对称结构信息,通过左右半脑灰度直方图的差异来确定脑肿瘤的大致灰度分布范围,从而实现脑肿瘤的粗略分割。在细化分割系统中,提出了一种新的基于局部统计直方图的主动轮廓模型(local histogram based active contour model,LHACM)方法,该方法可以有效地检测弱边缘,处理脑肿瘤图像中的灰度不均匀现象,并能实现演化曲线拓扑结构的改变。LHACM可以细化调整粗分割结果,使脑肿瘤的分割更加精确。整个系统实现了自动化运行,不需要人工的参与。通过实际脑MRI图像的实验验证,该系统具有很高的分割精度。This paper presents a system for automatically segmenting brain tumors in MRI images, including a rough segmentation subsystem and a refinement subsystem. The rough segmentation subsystem mainly utilizes symmetry information in the brain MR images. The gray range for the brain tumor is confirmed by using the difference between the gray level histogram of the left cerebral hemisphere and that of the right cerebral hemisphere to roughly segment the tumor. A local histogram based active contour model （LHACM） is used in the refinement subsystem. The model efficiently stops the contours at weak boundaries to deal with inhomogeneities in brain tumors and changes in topology. The rough segmentation result gives an initial picture for the active contour model, while the refinement makes the segmentation result more accurate. The brain tumor segmentation is then done automatically with no manual intervention. Quantitative evaluations and comparisons among several methods using real patient data demonstrate the effectiveness of the system.

关 键 词：图像分割 脑肿瘤 主动轮廓模型 对称性 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]