结合先验稀疏字典和空洞填充的CT图像肝脏分割  

Liver segmentation in CT image based on priori sparse dictionary and hole filling

作　　者：王雪虎[1] 杨健[1] 艾丹妮 王涌天[1] 

机构地区：[1]北京理工大学光电学院北京市混合现实与新型显示工程技术研究中心,北京100081

出　　处：《光学精密工程》2015年第9期2687-2697,共11页Optics and Precision Engineering

基　　金：国家973重点基础研究发展计划资助项目(No.2013CB328806);国家863高技术研究发展计划资助项目(No.2013AA013703);国家十二五科技支撑计划资助项目(No.2013BAI01B01)

摘　　要：针对复杂多变的肝脏图像,提出了一种基于先验稀疏字典和空洞填充的三维肝脏图像分割方法。对腹部CT图像进行Gabor特征提取,并分别在Gabor图像和灰度图像的肝脏金标准边界上选择大小相同的图像块作为两组训练集,利用训练集得到两种查询字典及稀疏编码。将金标准图像与待分割图像配准,并将配准后的肝脏边界作为待分割图像的肝脏初始边界;在初始边界点上的十邻域内选择大小相同的两组图像块作为测试样本,利用测试样本与查询字典计算稀疏编码及重构误差,并选择重构误差最小的图像块的中心作为待分割肝脏的边界点;最后,设计一种空洞填充方法对肝脏边界进行补全和平滑处理,得到最终分割结果。利用医学图像计算和计算机辅助介入国际会议中提供的肝脏数据进行了实验验证。结果表明,该方法对肝脏分割图像具有较好的适用性和鲁棒性,并获得了较高的分割精度。其中,平均体积重叠率误差为(5.21±0.45)%,平均相对体积误差为(0.72±0.12)%,平均对称表面距离误差为(0.93±0.14)mm。For complicated liver images, this segmentation method based on sparse dictionary paper presents a three-dimensional automatic liver and hole filling technologies. The Gabor feature of an abdominal CT image was extracted. The image blocks with the same size on the border of liver gold standard in Gabor images and CT images were selected as two groups of train sets. Then, the training sets were used to get the dictionaries and sparse coding. The golden standard image was registered with the image to be segmented, and registered liver boundary was taken as the initial liver boundary of the image to be segmented. Furthermore, the training sets in ten neighborhoods on two sets of images with the same size were selected as the initial boundary. The sparse coding and image reconstruction error were computed by using the testing sets and the block-sparse dictionary, and the final liver boundary with the smallest image reconstruction error was ohtained. Finally, a hole filling method was designed for liver boundary completion and smoothing to obtain the final segmentation results. The proposed method for the liver segmentation was evaluated by using the data sets of MICCAI 2007. The results show that this method has better segmentation applicability and robustness for the liver. It shows a higher segmentation accuracy, the volume overlap error rate is reduced to 5.21±0. 004 5, the relative volume error is 0.72±0. 001 2, and the average symmetric surface distance error is reduced to （0.93!0.14） mm.

关 键 词：计算机层析(CT)图像 肝脏分割 稀疏编码 字典学习 空洞填充 

分 类 号：TP391[自动化与计算机技术—计算机应用技术] R814.42[自动化与计算机技术—计算机科学与技术]