Okada Toshiyuki, Yokota Keita, Hori Masatoshi, Nakamoto Masahiko, Nakamura Hironobu, Sato Yoshinobu
Graduate School of Information Science and Technology, Osaka University, Japan.
Med Image Comput Comput Assist Interv. 2008;11(Pt 1):502-9. doi: 10.1007/978-3-540-85988-8_60.
Hierarchical multi-organ statistical atlases are constructed with the aim of achieving fully automated segmentation of the liver and related organs from computed tomography images. Constraints on inter-relations among organs are embedded in hierarchical organization of probabilistic atlases (PAs) and statistical shape models (SSMs). Hierarchical PAs are constructed based on the hierarchical nature of inter-organ relationships. Multi-organ SSMs (MO-SSMs) are combined with previously proposed single-organ multi-level SSMs (ML-SSMs). A hierarchical segmentation procedure is then formulated using the constructed hierarchical atlases. The basic approach consists of hierarchical recursive processes of initial region extraction using PAs and subsequent refinement using ML/MO-SSMs. The experimental results show that segmentation accuracy of the liver was improved by incorporating constraints on inter-organ relationships.
构建分层多器官统计图谱的目的是从计算机断层扫描图像中实现肝脏及相关器官的全自动分割。器官间相互关系的约束被嵌入到概率图谱（PA）和统计形状模型（SSM）的分层组织中。基于器官间关系的分层性质构建分层PA。多器官SSM（MO-SSM）与先前提出的单器官多级SSM（ML-SSM）相结合。然后使用构建的分层图谱制定分层分割程序。基本方法包括使用PA进行初始区域提取以及随后使用ML/MO-SSM进行细化的分层递归过程。实验结果表明，通过纳入器官间关系的约束，肝脏的分割精度得到了提高。
