摘要
针对CT影像存在伪影、分割困难的问题,提出了复Contourlet域树结构马尔可夫随机场(tree-structured Markov random filed, TS-MRF)的医学CT图像分割算法。采用复Contourlet分析提取CT图像各尺度中的特征信息,特征信息的相关性以其对应标记的相关性表征;其次,相邻尺度间标记的相关性通过构建一阶Markov模型建立联系;尺度内通过构建TS-MRF模型,采用父节点标记对子节点标记的约束以及子节点邻域间标记的相关性描述尺度内节点标记的局部相关性;CT图像特征场通过在每一尺度内构建同标记的高斯模型表征;最后,图像分割的结果通过极大化特征场与标记场联合分布来实现。实验结果表明,相对于空域TS-MRF、小波域TS-MRF、空域马尔科夫随机场(Markov random filed, MRF)、复小波域MRF等4种算法,复域(ontourle, TS-MRF)算法反映分割区域一致性的概率Rand指数(probabilistic rand index, PRI)提高0.091 3以上;同区域分割误差指标全局一致性误差指数(global consistency error, GCE)降低了0.002 8以上;分割边缘连续性指标边界偏移误差指数(boundary displacement error, BDE)降低0.617 9以上;分割后信息丢失指标信息变化指数(variation of information, VoI)降低了0.889 6以上。因此,算法对医学CT图像分割具有较好的区域一致性、分割精度和鲁棒性。
Medical CT image segmentation algorithm based on tree-structured Markov random filed(TS-MRF) in complex contourlet domain is proposed. Using complex contourlet analysis to extract feature information in each scale of CT images, the correlation of feature information and the correlation characterization of corresponding marks. Second, the correlation of the marks between adjacent scales is established by constructing a first-order Markov model;By constructing the TS-MRF model within the scale, we use the constraints of the parent node label on the child node label and the correlation between the label of the child node neighborhood to describe the local correlation of the node label within the scale. The CT image feature field is characterized by constructing the same-labeled Gaussian model in each scale. Finally, the results of image segmentation are achieved by maximizing the joint distribution of feature fields and label fields. The experimental results show that, compared with the other four algorithms, the Probabilistic rand index(PRI) of the algorithm in this paper has increased by more than 0.091 3. The global consistency error(GCE) of the same region segmentation error index is reduced by more than 0.002 8. The boundary displacement error(BDE) of the segmentation edge continuity index is reduced by more than 0.617 9. after the segmentation, the information loss index called variation of information(VOI) decreased by more than 0.889 6. This algorithm has good regional consistency, segmentation accuracy and robustness for medical CT image segmentation.
作者
夏平
彭程
施宇
雷帮军
Xia Ping;Peng Cheng;ShiYul;Lei Bangjun(Hubei Key Laboratory of Intelligent Vision based Monitoring for Hydroelectric Engineering,Three Gorges University,Yichang 443002,China;College of Computer and Information Technology,Three Gorges University,Yichang 443002,China)
出处
《国外电子测量技术》
北大核心
2022年第10期155-163,共9页
Foreign Electronic Measurement Technology
基金
国家自然科学基金(联合基金)(U1401252)
湖北省重点实验室开放基金(2018SDSJ07)项目资助。
