Automatic liver segmentation technique for three-dimensional visualisation of CT data
Automatic liver segmentation technique for three-dimensional visualisation of CT data
复制标题
DOI：
10.1148/radiology.201.2.8888223
复制
发表时间：
1996-11-01
期刊：
影响因子：
19.7
通讯作者：
Fishman, EK
中科院分区：
文献类型：
--
作者：
Gao, LM;Heath, DG;Fishman, EK
目的：开发一种从腹部计算机断层扫描（CT）图像中自动分割肝脏以用于三维容积再现显示的系统。
材料与方法：开发了一种自动肝脏分割系统，该系统将领域知识与全局直方图分析、形态学算子以及参数可变轮廓模型相结合。通过利用相邻层面的信息对阈值化肝脏容积的边界进行逐层面修改。通过优化参数可变轮廓模型对这些边界进行细化。利用这些边界排除肝脏外的组织来创建容积再现图像。该系统在10例可能可切除的肝脏肿瘤的CT数据集上进行了测试。
结果：在10例的401个层面中，53个层面（13.2%）在分割过程中需要用户修改。放射科医生判断使用这些肝脏边界的三维再现图像的效用与手动编辑创建的三维图像相当。其中28个层面被放射科医生认为不完美，可能需要进一步修改。
结论：已经开发出一种从CT图像中自动分割肝脏的有效技术。该技术有望通过减少操作人员的干预来节省时间并简化三维肝脏图像的创建。
PURPOSE: To develop a system for automatic segmentation of the liver from computed tomographic (CT) scans of the abdomen for three-dimensional volume-rendering displays.MATERIALS AND METHODS: An automated liver segmentation system was developed, which combined domain knowledge with analysis of a global histogram, morphologic operators, and the parametrically deformable contour model. Boundaries of the thresholded liver volume were modified section-by-section by exploiting information from adjacent sections. These boundaries were refined by optimization of the parametrically deformable contour model. Volume-rendered images were created by using the boundaries to exclude tissues outside the liver. The system was tested on CT data sets from 10 cases of potentially resectable hepatic neoplasm.RESULTS: Of the 401 sections in the 10 cases, 53 sections (13.2%) required user modifications during segmentation. The utility of the three-dimensional-rendered images with use of these liver boundaries was judged by a radiologist as being comparable to that of three-dimensional images created with manual editing. Twenty-eight of the sections were deemed imperfect by the radiologist and might need further modifications. CONCLUSION: An effective technique for automatic segmentation of the liver from CT images has been developed. This technique promises to save time and simplify the creation of three-dimensional liver images by minimizing operator intervention.