Statistical Atlases and Computational Models of the Heart. Imaging and Modelling Challenges
Statistical Atlases and Computational Models of the Heart. Imaging and Modelling Challenges
复制标题
DOI：
10.1007/978-3-319-28712-6
复制
发表时间：
2012
影响因子：
--
通讯作者：
O. Camara;Tommaso Mansi;M. Pop;K. Rhode;Maxime Sermesant;A. Young
中科院分区：
文献类型：
--
作者：
O. Camara;Tommaso Mansi;M. Pop;K. Rhode;Maxime Sermesant;A. Young
我们提出了一个完全自动化的系统分割左心室（LV）在心脏MR图像的基础上统计和变形模型。3D LV形状的投影输出逆组成主动外观模型产生使用统一的统计/确定性可变形模型细化的分割。一种新的多尺度检测器，基于直方图的方向性（HoG），产生初始估计LV的位置和规模的MR体积。在30幅MICCAI Grand Challenge测试图像上评估HoG检测器和基于可变形模型的分割组件的性能。检测器边界框重叠的平均F测量值为0.89。轮廓重叠的平均F测量值为0.80（endo）、0.82（epi）和0.46（myocardial）。
We present a fully automated system for segmenting the Left Ventricle (LV) in cardiac MR images based on statistical and deformable models. A Project-Out Inverse Compositional Active Appearance Model of 3D LV shape produces segmentations that are refined using a unified statistical/deterministic deformable model. A new multi-scale detector, based on the Histogram of Oriented Gradients (HoG), produces initial estimates of LV position and scale in the MR volume. The performance of the HoG detector and the deformable-model-based segmentation components are evaluated on the 30 MICCAI Grand Challenge test images. The average F-measure for detector bounding box overlap is 0.89. The average F-measures for contour overlap are 0.80 (endo), 0.82 (epi), and 0.46 (myocardium).