Automated Screening of Diabetic Retinopathy by Content

描述(申请人提供)：我们的项目“糖尿病视网膜病变的内容自动筛查”(R01 EY017065)的目标是研究使用基于内容的图像检索来检测并准确描述和索引人类视网膜疾病，特别是糖尿病视网膜病变的可行性，这些疾病是从低成本、非扩张的视网膜照片远程收集的。基于内容的图像检索(CBIR)是从非常大的数据库集合中根据图片内容检索相关图像的过程。我们的概念性假设预测，通过从数字图像(内容信息)中提取特征，并将图像和相关元数据(上下文信息)与从大型编辑的视网膜图像库检索到的相似的、经验证的图像进行比较，基于计算机的(即自动的)诊断能力将出现。 利用CBIR，我们成功地开发了一种基于网络的方法，通过远程访问基于计算机的诊断图像分析方法，允许在初级保健保健环境中实时远程诊断DR。我们在这次竞争性更新中提出的研究旨在解决我们当前算法在自动机器分割性能方面的关键方法。我们的目标是将性能提高到允许作为具有专家能力的全自动患者护理范例实施的水平，从而产生尽可能高的疾病检测灵敏度和特异性。我们还将编制一个足够大的库，以验证我们的假设，即临床元数据(上下文数据)可以有助于CBIR方法的性能(敏感度和特异度)，为远程检测和诊断Dr.公共卫生相关性提供可靠的诊断方法：到2030年，全球每天将有必要对100万名糖尿病眼病患者进行检查。糖尿病视网膜病变的治疗是可用的；我们的挑战在于找到一种在大量人群中检测和管理糖尿病眼病的经济有效的方法。利用新颖的图像分析和临床元数据算法，将计算机成像应用于视网膜疾病的诊断，有望实现低成本、自动化的诊断方法，通过为迅速扩大的高危人群中服务不足的患者提供基于图像的“专家”诊断来改善社区眼睛健康。

DESCRIPTION (provided by applicant): The goal of our project "Automated Screening for Diabetic Retinopathy by Content" (R01 EY017065) is to investigate the feasibility of using content-based image retrieval to detect and accurately describe and index human retinal disease, specifically diabetic retinopathy, collected remotely from low-cost, non-dilated retinal photographs. Content-based image retrieval (CBIR) is the process of retrieving related images from very large database collections, based upon their pictorial content. Our conceptual hypothesis predicts that by extracting features from digital images (content information) and comparing the image and associated metadata (contextual information) to similar, validated images retrieved from a large compiled retinal image library, computer-based, (i.e. automated) diagnostic capabilities would emerge. Using CBIR, we have successfully developed a web-based method that permits remote diagnosis of DR in the primary care health setting, in real time, through remote access to a computer-based, diagnostic, image analysis method. The studies we propose in this competitive renewal are designed to address key methods in the performance of automated machine segmentation by our current algorithms. Our goal is to improve the performance to a level which will permit implementation as a fully automated patient care paradigm with expert capabilities that yield the highest possible sensitivity and specificity of disease detection. We will also compile a library large enough to validate our hypothesis that clinical metadata (contextual data) can contribute to the performance (sensitivity and specificity) of the CBIR method to provide a robust diagnostic method for remote detection and diagnosis of DR. PUBLIC HEALTH RELEVANCE: By 2030 it will be necessary to examine 1 million patients for diabetic eye disease every day worldwide. Treatment for DR is available; our challenge lies in finding a cost-effective approach to detecting and managing diabetic eye disease in large populations. The application of computer-based imaging to the diagnosis of retinal disease, using novel image analysis and clinical metadata algorithms hold the promise of achieving low-cost, automated, diagnostic methods to improve community eye health through access to image-based "expert" diagnosis for underserved patients in rapidly expanding at-risk populations.