复小波域混合概率图模型的超声医学图像分割

针对存在大量不规则斑点噪声、目标边缘弱化的超声医学图像分割中较难识别目标的问题, 提出了一种复小波域中混合概率图模型的超声医学图像分割算法.采用具有近似平移不变性和良好方向选择性的双树复小波变换(Dual tree-complex wavelet transform, DT-CWT)提取超声医学图像6个方向的高频特征信息; 其次, 为关联目标的弱特征信息并抑制统计独立的高频噪声, 构建了复小波域混合概率图模型; 尺度间"父—子"节点间标记采用贝叶斯网络进行建模, 尺度内邻域间标记采用马尔科夫随机场(Markov random field, MRF)无向图建模, 对复小波域中同尺度的特征系数采用高斯混合模型建模, 尺度内同标记的观测特征采用高斯模型建模; 最后, 用迭代条件模式(Iterated conditional mode, ICM)实现MRF中误分割率最小的能量函数最优解, 获取标记场, 实现超声医学图像分割.实验结果从视觉效果和定量分析两方面验证表明, 本文算法能有效地提取超声图像的弱目标信息, 较好地定位目标区域, 具有较高的分割精度和鲁棒性.

Ultrasound Medical Image Segmentation Based on Hybrid Probabilistic Graphical Model in Complex-wavelet Domain

For ultrasound medical image with a lot of irregular speckle noise and the weakened object edge,it is difficult to identify the object in image segmentation.An ultrasound medical image segmentation algorithm based on hybrid probabilistic graphical model in complex-wavelet domain is proposed.First,dual tree-complex wavelet transform(DTCWT)has approximate shift invariance and nice directional selectivity,so using DT-CWT to extract the high frequency characteristic information.Then,hybrid probabilistic graphical model in complex-wavelet domain is constructed,for associating the weak characteristic information of object and suppressing statistical independent of high-frequency noise.The labels of the father-child nodes in the inter-scale is modeled by Bayesian network.The mark within the intra-scale of the neighborhood is modeled by Markov random field(MRF)undirected graph model.The characteristic coefficients of the same scale in the complex-wavelet domain is modeled by Gaussian hybrid model.Using Gaussian model describes the observational characteristics with a same mark in the intra-scale.Finally,the optimal solution of the energy function with the smallest error segmentation rate in the MRF is obtained by using iterated conditional mode(ICM)to get the tag field,and complete the ultrasound medical image segmentation.From two aspects of visual effect and quantitative analysis,the proposed algorithm can extract the weak object information of ultrasound image effectively,which can better locate the target region,and has higher segmentation accuracy and robustness.