Real-Time Algorithms for Automatic Vasculature Map Generation and Feature-Based Location Determination From Intra-Ocular Image Sequences

激光眼科手术是治疗脉络膜新生血管（CNV）的已知最有效的方法，这些脉络膜新生血管出现在各种情况下，包括年龄相关性黄斑变性、组织浆性脉络膜炎等。然而，目前这种方法在一次治疗后根除CNV的成功率不到50%，复发率和/或持续率约为50%。这项研究将使大大降低故障率的仪器成为可能。该仪器由一个计算机化的多光谱三维眼底相机和一个相关的平视显示器和一个实时激光跟踪系统组成。跟踪系统使用一套自动图像分析程序，可以精确绘制病理，并进行计算机化激光治疗计划。正在解决的核心图像处理问题是：(i)从图像序列构建视网膜血管系统的广域图；以及（ii）以至少每秒30帧的帧率以2像素或更好的精度确定当前帧相对于在步骤(i)中构造的广域地图的位置。上述问题正在通过开发一种快速算法来解决，该算法用于在存在少量噪声点的情况下匹配具有未知对应的点集。该算法通过将搜索限制在最可靠的图像特征点的最小足够子集中来运行。此外，所提出的算法正在线性阵列处理器上实现，进一步实现实时操作。

Laser eye surgery is the most effective known procedure for treating choroidal neovascularization (CNV) found in a variety of conditions including age-related macular degeneration, histoplasmic choroiditis, etc. However, the current rate of success of this procedure is less than 50% for eradication of the CNV following one treatment session, with a recurrence and/or persistence rate of about 50%. This research will make possible an instrument that will drastically reduce this failure rate. The instrument consists of a computerized multi-spectral 3-D fundus camera with an associated head-up display and a real-time laser tracking system. The tracking system uses a set of automatic image analysis routines that allow accurate mapping of the pathologies, and computerized laser treatment planning. The core image-processing problems that are being addressed are: (i) constructing a wide-area map of the retinal vasculature from an image sequence; and (ii) determining the location of the current frame relative to the wide-area map constructed in step (i) to an accuracy of 2 pixels, or better, at a frame rate of at least 30 per second. The above problems are being solved by the development of a fast algorithm for matching point sets with unknown correspondences, in the presence of a small number of noise points. This algorithm operates by limiting the search to a minimally- sufficient subset of the most reliable image feature points. In addition, the proposed algorithm is being implemented on a linear array of processors, further enabling real-time operation.