摘要
目的本文提出一种基于聚类的无监督脑部MR图像分割新算法,有别于传统的基于灰度阈值和一维空间MR图像分割算法。方法首先,估算输入图像的质子密度和弛豫时间;然后,描述输入图像的概率分布;最后,采用基于空间关联决策准则识别最佳分类区域,达到图像分割的效果。结果选用不同分割算法对人工合成图像和临床实例MR图像进行仿真实验。定性分析结果是本文算法的分割图像边缘和细节部分保存的完整清晰;定量评估结果显示基于本文分割算法能获得探测率最大和误报率最小,且在15~30 dB信噪比范围内的戴斯相似性系数和杰卡德相似性系数均最大。结论基于质子密度和弛豫时间的统计算法是一种可行的脑部MR分割算法,在噪声环境、图像灰度不均和临床实例等情况下均表现出强健性,具有较高的临床应用价值。
Objective This paper proposed a brain joint segmentation and classification algorithm based on proton density（ρ） and relaxation time（T_1） and（T_2）, instead of the acquired gray level image. MethodsEstimation of proton density and relaxation time was made, then the approach exploited the statistical distribution of the involved signals in the complex domain; at last a novel method for identifying the optimal decision regions was proposed, which could achieve the ideal segmentation results. ResultsBoth simulated and real datasets were evaluated by using different methods. Qualitative analysis showed that edges were well retrieved and small structures were preserved and completely clear. Quantitative evaluation results showed that the proposed segmentation algorithm in this paper could provide the best detection probability and false alarm probability. And it could acquire the maximal Dice coefficient and Jaccard similarity indexes in case of different SNR（15~30 dB）. Conclusion The proposed method based on ρ, T_1 and T_2 maps was a feasible segmentation algorithm. And it could provide better robustness in the noise environment, intensity inhomogeneity and clinical applications, which was of great value in clinical popularization.
出处
《中国医疗设备》
2016年第10期25-28,共4页
China Medical Devices
关键词
质子密度
弛豫时间
概率分布
空间关联准则
MR图像分割
proton density
relaxation time
statistical distribution
spatial correlation
MR image segmentation
