Retinal Vessel Measurement and Characterization System

描述(申请人提供)：视网膜微血管异常与许多眼部和全身疾病有关，包括糖尿病视网膜病变、高血压、中风和脑微血管疾病。例如，泛发性小动脉狭窄和动静脉划痕似乎是高血压的不可逆转的长期标记物。研究结果表明，视网膜照相可能有助于评估风险分层和在适当人群中筛查视网膜疾病。高度进化的成像解决方案和计算机处理能力为量化这些异常打开了一扇门。由于计算机自动化，提出了一种高效和全面的方法，即可重复、无读者间和读者内部的可变性、精确度，并与人类分析师的技术一致。我们建议开发和验证一个系统，该系统提取视网膜血管，对其分叉和交叉进行分类，并量化视网膜血管系统的直径测量。Auto-Cal系统将使用先进的形态和高斯过滤技术来分割血管网络。一旦分割，将通过形态骨架、路径遍历和剪枝算法对交叉和分叉进行分类。然后，通过将灰度轮廓与描述典型血管的高斯参数进行拟合来计算血管直径。该系统将在50张数字图像上进行测试，这些图像将使用先前开发的计算机辅助工具手动测量其8个主要血管。与人工得出的血管直径的高度统计相关性将决定这一过程的成功。 第一阶段的具体目标是1)开发和验证全自动视网膜血管提取算法以从数字眼底照片中分割出动脉和静脉，2)开发和验证血管分叉和交叉识别算法，以及3)开发和验证血管直径测量算法。 在第二阶段，Auto-Cal工具将用于一项对比研究，检查400名增殖期前期糖尿病视网膜病变患者。这些血管已经被人工测量过了。这些数据将被用来针对大量的“地面真相”人群来验证该系统。

DESCRIPTION (provided by applicant): Retinal microvascular abnormalities are related to many ocular and systemic diseases including diabetic retinopathy, hypertension, stroke, and cerebral micro-vascular diseases. For example generalized arteriolar narrowing and arteriovenous nicking appear to be irreversible long-term markers of hypertension. Findings suggest that retinal photography may be useful for assessing risk stratification and screening for retinal disease in appropriate populations. Highly evolved imaging solutions and computer processing power have opened a door to quantify these abnormalities. A methodology is suggested that is efficient and comprehensive because of computer automation, that is repeatable, free of inter- and intra-reader variability, precise, and consistent with the human analysts' techniques. We propose to develop and validate a system that extracts retinal vessels, classifies their bifurcations and crossings, and quantifies a diameter measurement of the retinal vas-culature system. The Auto-Cal system will use advanced morphological and Gaussian filtering techniques to segment the vessel network. Once segmented the crossings and bifurcations will be classified through morphological skeletonization path traversing and pruning algorithms. The vessel diameter is then calculated by fitting the gray level profile to Gaussian parameters that describe a typical vessel. The system will be tested on 50 digital images, which will have their 8 major vessels manually measured with a previously developed computer-aided tool. High statistical correlation with manually derived vessel diameters will determine the success of the process. The specific Phase I aims are to 1) Develop and validate a fully automatic retinal vessel extraction algorithm to segment the arteries and veins from a digital fundus photograph, 2) Develop and validate a vessel bifurcation and crossing recognition algorithm, and 3) Develop and validate a vessel diameter measurement algorithm. In Phase II the Auto-Cal tool will be used in a comparative study to examine 400 pre-proliferative diabetic retinopathy patients. The vessels have been manually measured. These data will be used to validate the system against a large "ground truth" population.