Automated Image-based Biomarker Computation Tools for Diabetic Retinopathy

描述(申请人提供)：在这个STTR项目中，我们介绍了一套先进的图像分析工具EyeMark，用于使用视网膜眼底图像自动计算糖尿病视网膜病变(DR)的生物标记物。具体地说，我们将开发计算微动脉瘤(MA)出现和消失率(统称为周转率)的工具，作为量化DR进展风险的生物标记物。可靠的基于图像的生物标记物的可用性将对DR护理的各个方面产生高度积极的影响，包括筛查、监测进展、药物发现和临床研究。测量MA周转涉及两个劳动密集型步骤：仔细对当前和基线图像进行对齐，以及对单个MA进行标记。如果完全由评分员来完成，这个过程非常耗时，而且容易出错。该项目的主要目标是通过自动化MA营业额测量中涉及的两个步骤来克服这些限制：准确的图像配准和MA检测。在第一阶段，我们设计和开发了MA周转率计算原型工具，该工具能够健壮地配准纵向图像(即使有多个病变变化)，并有效地检测MA(病变级别AUROC=0.92)。该工具通过报告MA周转作为一个范围来提供对混杂图像因素的优雅降级，从而捕获MA检测的内在置信度。到第二阶段结束时，我们将开发一个经过临床验证的端到端桌面软件，用于稳健、自动化地计算MA周转生物标记物，该软件可以在云上工作，在几乎恒定的时间内(对于大数据集)产生结果，并为临床医生提供直观的可视化工具，以更有效地监控DR进展。

DESCRIPTION (provided by applicant): In this STTR project, we present EyeMark, a set of advanced image analysis tools for automated computation of biomarkers for diabetic retinopathy (DR) using retinal fundus images. Specifically, we will develop tools for computation of microaneurysm (MA) appearance and disappearance rates (jointly known as turnover rates) for use as a biomarker in quantifying DR progression risk. The availability of a reliable image-based biomarker will have high positive influence on various aspects of DR care, including screening, monitoring progression, drug discovery and clinical research. Measuring MA turnover involves two labor intensive steps: careful alignment of current and baseline images, and marking of individual MAs. This process is very time consuming and prone to error, if done entirely by human graders. The primary goal of this project is to overcome these limitations by automating both the steps involved in MA turnover measurement: accurate image registration, and MA detection. In Phase I we have designed and developed a MA turnover computation prototype tool that robustly registers longitudinal images (even with multiple lesion changes) and effectively detects MAs (lesion level AUROC=0.92). The tool provides graceful degradation to confounding image factors by reporting MA turnover as a range, thereby capturing the inherent confidence in MA detection. By the end of Phase II we will develop a clinically validated end-to-end desktop software for robust, automated computation of MA turnover biomarker, that can work on the cloud to produces results in near constant time (for large datasets), and also provide intuitive visualization tools for clinicians to more effectively monitor DR progression.