Automated Assessment of Optic Nerve Edema with Low-Cost Imaging

 描述（由申请人提供）： 视神经乳头水肿--一种与颅内压升高相关的视神经水肿--发生在退伍军人中，原因包括颅内肿块或出血、阻塞性脑积水、创伤性脑损伤和特发性原因。目前，视乳头水肿的诊断及其严重程度是通过直接眼底镜观察或数字眼底照片使用六阶段分级量表进行评估。该量表是定性的，分类的，受评分者间变异性的影响，需要临床专业知识，并且不能区分视神经乳头水肿与其他视神经水肿原因。光学相干断层扫描（OCT）图像的自动化图像分析的最新发展已经导致水肿严重程度的连续定量测量（如总视网膜体积）的可用性。然而，光学相干断层扫描仍然过于昂贵，无法在专业眼科诊所之外轻易获得。使用低成本的便携式彩色眼底照相机是远程医疗和紧急情况下更经济可行的成像选择，但目前还没有客观的方法来分析和分类视神经水肿的原因和严重程度。该项目的总体目标是开发基于彩色眼底荧光的自动图像分析策略，以快速准确地确定视神经水肿的严重程度，并确定区分其原因的特征。核心假设是，通过使用 将基于光学相干断层扫描的严重性测量作为基于计算机的彩色眼底照片特征学习的参考标准，将提高在远程医疗设置（其中可能只有彩色眼底照片可用）中自动估计视神经水肿严重性的能力。拟议研究的基本原理是，通过眼底照片确定视神经水肿的严重程度和原因的自动化方法将能够在退伍军人的远程医疗和紧急情况下早期诊断视力和危及生命的疾病。具体目标如下：1.发展并评估彩色眼底照片计算肿胀视神经头相关特征的自动化方法。这将通过改进和增强自动眼底图像分析方法来完成，用于计算视神经水肿患者的全局/区域血管、纹理和视盘边界特征。这些特征的相关性将首先通过与专家的联系进行评估， 定义的严重度等级。 2.从彩色眼底照片中识别与基于OCT的视神经水肿患者严重程度测量最佳相关的特征，并开发基于眼底照片的连续严重程度量表。这将通过使用机器学习方法来实现，以将基于全球/区域眼底的特征与最近开发的基于OCT的新型自动化严重程度测量（例如，总视网膜体积）。预计使用开发的自动化方法进行的基于眼底的严重程度测量将更好地与基于OCT的严重程度测量相关（比专家定义的Frisén量表等级-当前基于眼底的临床顺序严重程度量表）。 3.从彩色眼底照片中识别区分视乳头水肿和其他视盘肿胀原因的特征，并开发预测分类器。工作的假设是，视盘周围的区域差异将在自动分化过程中贡献最大。该方法是创新的，因为使用自动图像分析算法来评估低成本图像中视神经水肿的程度和原因，反映了依赖定性信息的现状的重要部门。拟议的研究意义重大，因为它将有助于建立急需的定量和客观方法，通过这些方法来评估远程医疗和紧急情况下退伍军人视盘肿胀的严重程度和原因。

 DESCRIPTION (provided by applicant): Papilledema - a specific type of optic nerve edema associated with elevated intracranial pressure - occurs in veterans due to intracranial masses or hemorrhage, obstructive hydrocephalus, following traumatic brain injury, and from idiopathic causes. Currently, the diagnosis of papilledema and its severity are assessed by direct funduscopic observation or digital fundus photographs using a six-stage grading scale. The scale is qualitative, categorical, subject to inter-rater variability, requires clinical expertise, and does not distinguish papilledea from other causes of optic nerve edema. Recent developments in the automated image-analysis of optical coherence tomography (OCT) images have led to the availability of continuous quantitative measures (such as the total retinal volume) of edema severity. However, optical coherence tomography is still too expensive to become readily available outside specialized ophthalmology clinics. Use of low-cost portable color fundus cameras is a more economically viable imaging option in telemedical and emergency settings, but there currently exists no objective method of analyzing and classifying the cause and severity of optic nerve edema. The overall objective of this project is to develop automated color-fundus-photograph-based image-analysis strategies to rapidly and accurately determine the severity of optic nerve edema and to ascertain features that differentiate its cause. The central hypothesis is that, through the use of optical-coherence-tomgraphy-based severity measures as a reference standard for computer-based learning of color-fundus-photograph features, the ability to automatically estimate the severity of optic nerve edema in telemedical settings (where only color fundus photographs are likely to be available) will be improved. The rationale for the proposed research is that having automated methods for determination of the severity and cause of optic nerve edema from fundus photographs will enable earlier diagnosis of vision- and life-threatening conditions in telemedical and emergency settings in veterans. The following specific aims will be pursued: 1. Develop and evaluate the automated methodology for computing swollen-optic-nerve-head- relevant features from color fundus photographs. This will be completed by refining and augmenting automated fundus-image-analysis approaches for computation of global/regional vascular, textural, and optic-disc boundary features in patients with optic nerve edema. The relevance of these features will first be evaluated in this aim via their associations with expert defined severity rankings. 2. Identify features from color fundus photographs that optimally correlate with OCT-based measurements of severity in patients with optic nerve edema and develop a fundus-photograph- based continuous severity scale. This will be accomplished by using machine-learning approaches to relate global/regional fundus-based features to recently developed novel automated OCT-based measures of severity (e.g., the total retinal volume). It is expected that fundus-based severity measures using the developed automated approach will better correlate (than expert-defined Frisén scale grades - the current clinical fundus-based ordinal severity scale) with OCT-based measures of severity. 3. Identify features from color fundus photographs that differentiate papilledema from other causes of optic disc swelling and develop a predictive classifier. The working hypothesis is that regional differences around the optic disc will contribute the most in the automatic differentiation process. The approach is innovative because the use of automated image-analysis algorithms for assessing the extent and cause of optic nerve edema in low-cost images reflects a significant department from the status quo of relying on qualitative information. The proposed research is significant because it will help to establish much- needed quantitative and objective methods by which to assess the severity and cause of optic disc swelling of veterans in telemedical and emergency settings.