先验形状约束的超声前列腺图像分割方法

摘要

A method for prostate segmentation in ultrasound image is presented using shape prior constraint and local image feature in this paper. A shape prior constraint is added into conventional active contour model to improve segmentation results which always interrupted by speckle noise,artifacts and intensity in homogeneity in ultrasound image. The method has two stages: shape model learning and shape prior segmentation. In shape model learning,principle component analysis is applied to learn the prostate shape,and then Gaussian distribution is utilized to estimate shape deformation parameters; in shape prior segmentation,local Gaussian fitting energy and shape prior constraint are combined together for prostate segmentation. Experimental results demonstrate the efficiency of presented method,which provides quantitative analysis for clinical diagnosis and therapy.%提出一种结合超声前列腺图像的局部特征和前列腺的先验形状知识的分割方法.该方法在传统图像分割方法中引入了前列腺的先验形状约束,使得分割能够一定程度地避免由于超声图像中噪声、伪影、灰度分布不均匀等因素对前列腺分割所造成的影响.算法分为两个部分:先验形状模型的学习和先验形状约束的分割.在先验形状模型学习阶段,采用主成分分析方法对形状作特征提取,以高斯分布作为形变参数的估计；在先验形状约束分割阶段,将基于局部高斯拟合特征的活动轮廓模型与形状模型相结合对前列腺图像分割.实验表明,所提出的方法在超声前列腺图像中取得了良好的分割效果,为临床诊断和治疗提供了定量分析的工具.