Biomedical Image Engineering of Macular Images

描述(由申请人提供)：这项建议将汇集一个在眼科疾病、成像和生物医学工程方面拥有世界级专业知识的团队，从照片和扫描激光眼底镜(SLO)图像中合成信息，以提高我们对老年性黄斑变性(AMD)的了解，AMD是发达国家的主要致盲原因。随着我们开发分析这些图像的技术，我们将与国际公认的AMD研究中心分享和测试我们的方法。 我们建议开发准确、经济、自动的数字图像分析工具，这些工具比目前的手动方法更有效。在日益紧张的卫生保健环境中，成本考虑尤为重要。例如，在NEI的一项重要试验--年龄相关眼病研究中，仅人工图像分析一项的成本就约为550万美元。 自动化方法还有助于简化各机构之间的共享，以进一步实现规模经济和成本效益。 AMD的诊断和治疗是基于对黄斑的摄影，因此这些图像中的信息至关重要。例如，被称为玻璃体的视网膜下沉积物就是这种疾病的标志。二十年来，研究人员一直在尝试使用数字技术来量化玻璃疱疹，但结果有限。识别玻璃疣的一个主要困难是，它们嵌入的背景本身是不均匀的。我们已经开发了黄斑的自动数学模型原型，具有足够的力量来克服这一障碍，并在不到一分钟的时间内准确地识别遇到的大多数图像中的斑疹。随着复杂的神经网络、水平集可变形图像和我们在生物医学工程领域的同事们证明对MRI图像和乳房X光图像的分析有效的其他图像工程技术，我们建议将我们的能力扩展到所有黄斑图像。到目前为止，有一个概念是我们努力的关键：通过一个用于统一对象识别的数学模型来平整图像背景。这一突破可能具有更广泛的意义：从部分背景图像数据计算这样一个模型，并使用它来消除图像中的背景变异性，可能有助于识别其他类型的医学图像中的病理结构。 扫描激光眼底镜(SLO)的最新成像技术可以提供AMD的生化基础信息，并在红外模式下成像视网膜下结构。SLO测量自体荧光(AF)，这揭示了一种名为脂褐素的潜在有毒物质在老化的眼睛中积累。脂褐素的积聚可能至少部分是由基因决定的。这种被称为ABCR的基因肯定会导致青少年黄斑变性患者的积聚，并已被认为与AMD有关。我们的姊妹研究哥伦比亚黄斑遗传学研究(CMGS)的目标是将该基因或其他候选基因与成年MD联系起来。在最初的三年中，我们已经能够生成一个包含2200多个研究对象的临床数据库，并预计在两年内达到3600个对象的目标，包括每个对象的黄斑照片、SLO图像和DNA样本。 我们建议通过将我们对单个图像的自动分析与三种图像类型的图像配准相结合，从CMG的照片和SLO图像(自动对焦和红外)中合成丰富的信息。例如，我们已经分析了与玻璃疣照片配准的AF图像，以提供证据表明玻璃疣和过度荧光的共同定位从阶段3(玻璃疣)到阶段4(玻璃疣和地理萎缩)AMD发生了戏剧性的转变(从75%到20%)，这可能为自然历史提供新的见解。在与伦敦国王学院医院的合作下，我们将寻求对这些发现的验证和理解，这现在是这项申请的主要目标。这类工作可以将特定的DNA突变与特定的AMD图像特征联系起来。这些知识可以为早期诊断处于危险中的个人奠定基础，然后他们可以根据特定的分子缺陷接受特定的治疗。这些进步将为我们老龄化的人口带来深远的健康和社会利益。

DESCRIPTION (provided by applicant): This proposal will bring together a team with world-class expertise in ophthalmic disease, imaging and biomedical engineering to synthesize information from photographs and scanning laser ophthalmoscope (SLO) images to improve our understanding of age-related macular degeneration (AMD), the leading cause of blindness in the developed world. As we develop the technology to analyze these images, we will reach out to share and test our methods with internationally recognized centers for AMD research. We propose to develop accurate, cost-effective, automatic digital image analysis tools that are more efficient than the present manual methods. Cost considerations are particularly important in the increasingly strained health-care environment. For example, the cost for manual image analysis alone in an important NEI trial, the Age Related Eye Disease Study, was about $5.5M. Automated methods also lend themselves to streamlined sharing between institutions for further economies of scale and cost. The diagnosis and treatment of AMD are based on photography of the macula, hence the paramount importance of the information in these images. For example, the subretinal deposits known as drusen are the hallmark of this disease. Researchers have been trying for two decades, with limited results, to use digital techniques to quantify drusen. A major difficulty in identifying drusen is that the background in which they are embedded is inherently non-uniform. We have developed prototype automated mathematical models of the macula with sufficient power to overcome this obstacle and accurately identify drusen in the majority of images encountered in less than a minute. With sophisticated neural networks, level set deformable images, and other image engineering techniques from our colleagues in biomedical engineering that have proved effective for analysis of MRI images and mammograms, we propose to extend our capability to all macular images. One concept has been key to our efforts so far: leveling image background by a mathematical model for uniform object identification. This breakthrough may have wider significance: the ideas of computing such a model from partial background image data, and using it to remove background variability from the image, could be useful for identifying pathological structures in other types of medical images. The newer imaging technique of scanning laser ophthalmoscopy (SLO) can provide information about the biochemical basis of AMD, and in infrared mode, image subretinal structures. The SLO measures autofluorescence (AF), which reveals the accumulation of a potentially toxic substance called lipofuscin in the aging eye. The build-up of lipofuscin may be at least in part genetically determined. The gene known as ABCR definitely causes build-up in patients with juvenile macular degeneration and has been implicated in AMD. Linking this gene, or other candidate genes, with adult MD is the goal of our sister study, the Columbia Macular Genetics Study (CMGS). During the first three years we have been able to generate a clinical database of over 2,200 study subjects, and expect to reach our target of 3600 subjects in two more years, with macular photographs, SLO images and DNA samples from each subject. We propose to synthesize a wealth of information from the photographs and SLO images (AF and infrared) of the CMGS by combining our automated analyses of individual images with image registration of the three image types. For example, we have already analyzed AF images in registration with drusen photographs to provide evidence for a dramatic shift (from 75 percent to 20 percent) in the co-localization of drusen and hyperfluorescence from stage 3 (drusen) to stage 4 (drusen and geographic atrophy) AMD that may provide new insight on the natural history. In collaboration with King's College Hospital, London, we will pursue validation and understanding of these findings, which is now a major goal of this application. This is the type of work that could provide the linkage of specific DNA mutations to specific AMD image characteristics. Such knowledge could form the basis for early diagnosis of individuals at risk, who could then receive specific therapies based on specific molecular defects. These advances would extend profound health and social benefits to our aging population.