IEEE Trans Med Imaging. 2014 May;33(5):1109-18. doi: 10.1109/TMI.2014.2304959.
Digital image-based elasto-tomography (DIET) is a prototype system for breast cancer screening. A breast is imaged while being vibrated, and the observed surface motion is used to infer the internal stiffness of the breast, hence identifying tumors. This paper describes a computer vision system for accurately measuring 3-D surface motion. A model-based segmentation is used to identify the profile of the breast in each image, and the 3-D surface is reconstructed by fitting a model to the profiles. The surface motion is measured using a modern optical flow implementation customized to the application, then trajectories of points on the 3-D surface are given by fusing the optical flow with the reconstructed surfaces. On data from human trials, the system is shown to exceed the performance of an earlier marker-based system at tracking skin surface motion. We demonstrate that the system can detect a 10 mm tumor in a silicone phantom breast.
基于数字图像的弹性层析成像（DIET）是一种用于乳腺癌筛查的原型系统。在对乳房进行成像时，乳房会被振动，而观察到的表面运动被用于推断乳房内部的硬度，从而识别肿瘤。本文描述了一种用于精确测量 3-D 表面运动的计算机视觉系统。基于模型的分割用于识别每个图像中的乳房轮廓，然后通过拟合模型来重建 3-D 表面。使用针对应用定制的现代光流实现来测量表面运动，然后通过将光流与重建的表面融合来给出 3-D 表面上点的轨迹。在人体试验数据上的结果表明，该系统在跟踪皮肤表面运动方面的性能超过了早期基于标记的系统。我们证明该系统能够检测到硅树脂模型乳房中的 10 毫米肿瘤。
