Brinkley J F
Department of Biological Structure, University of Washington, Seattle 98195.
Proc Annu Symp Comput Appl Med Care. 1991:460-4.
A representation is described for two-dimensional anatomic shapes which can be described by single-valued distortions of a circle. The representation, called a radial contour model, is both generic, in that it captures the expected shape as well as the range of variation for an anatomic shape class, and flexible, in that the model can deform to fit an individual instance of the shape class. The model is implemented in a program called SCANNER (version 0.61) for 2-D interactive image segmentation and matching. An initial evaluation was performed using 7 shape models learned from a training set of 93 contours, and a control model containing no shape knowledge. Evaluation using 60 additional contours showed that in general the shape knowledge should reduce interactive segmentation time by a factor of two over the control, and that for specific shapes such as the eye, the improvement is much greater. A matching function was also devised which showed that the radial contour model should allow diagnosis of subtle shape changes. These results suggest that the use of spatial anatomic knowledge, when combined with good interactive tools, can help to alleviate the segmentation bottleneck in medical imaging. The models, when extended to more complex shapes, will form the spatial component of a knowledge base of anatomy that could have many uses in addition to image segmentation.
描述了一种用于二维解剖形状的表示方法，这些形状可以通过圆的单值变形来描述。这种表示方法称为径向轮廓模型，它具有通用性，因为它能捕捉解剖形状类别的预期形状以及变化范围；同时具有灵活性，因为该模型可以变形以适应形状类别的单个实例。该模型在一个名为SCANNER（版本0.61）的程序中实现，用于二维交互式图像分割和匹配。使用从93个轮廓的训练集中学习到的7个形状模型以及一个不包含形状知识的控制模型进行了初步评估。使用另外60个轮廓进行的评估表明，一般来说，形状知识应使交互式分割时间比控制模型减少一半，对于诸如眼睛等特定形状，改进幅度更大。还设计了一个匹配函数，结果表明径向轮廓模型应能诊断出细微的形状变化。这些结果表明，当空间解剖知识与良好的交互式工具相结合时，有助于缓解医学成像中的分割瓶颈。这些模型在扩展到更复杂的形状时，将构成解剖学知识库的空间组件，除了图像分割外还可能有许多用途。
