Böttger Thomas, Kunert Tobias, Meinzer Hans P, Wolf Ivo
Division Medical and Biological Informatics, German Cancer Research Center, Heidelberg, Germany.
Acad Radiol. 2007 Mar;14(3):319-29. doi: 10.1016/j.acra.2006.12.001.
Medical image segmentation is still very time consuming and is therefore seldom integrated into clinical routine. Various three-dimensional (3D) segmentation approaches could facilitate the work, but they are rarely used in clinical setups because of complex initialization and parametrization of such models.
We developed a new semiautomatic 3D-segmentation tool based on deformable simplex meshes. The user can define attracting points in the original image data. The new deformation algorithm guarantees that the surface model will pass through these interactively set points. The user can directly influence the evolution of the deformable model and gets direct feedback during the segmentation process.
The segmentation tool was evaluated for cardiac image data and magnetic resonance imaging lung images. Comparison with manual segmentation showed high accuracy. Time needed for delineation of the various structures could be reduced in some cases. The model was not sensitive to noise in the input data and model initialization.
The tool is suitable for fast interactive segmentation of any kind of 3D or 3D time-resolved medical image data. It enables the clinician to influence a complex 3D-segmentation algorithm and makes this algorithm controllable. The better the quality of the data, the less interaction is required. The tool still works when the processed images have low quality.
医学图像分割仍然非常耗时，因此很少被整合到临床常规工作中。各种三维（3D）分割方法可能会简化这项工作，但由于此类模型的初始化和参数化复杂，它们在临床环境中很少被使用。
我们基于可变形单纯形网格开发了一种新的半自动3D分割工具。用户可以在原始图像数据中定义吸引点。新的变形算法保证表面模型将通过这些交互式设置的点。用户可以直接影响可变形模型的演化，并在分割过程中获得直接反馈。
该分割工具针对心脏图像数据和磁共振成像肺部图像进行了评估。与手动分割相比显示出高精度。在某些情况下，可以减少描绘各种结构所需的时间。该模型对输入数据中的噪声和模型初始化不敏感。
该工具适用于对任何类型的3D或3D时间分辨医学图像数据进行快速交互式分割。它使临床医生能够影响复杂的3D分割算法并使该算法可控。数据质量越好，所需的交互就越少。当处理的图像质量较低时，该工具仍然有效。
