Computer Aided Classification of Diabetic Macular Edema

描述(申请人提供)：目前还没有成熟的方法来识别和评估糖尿病黄斑水肿(DME)的病理生物学机制，DME是美国劳动年龄人群失明的主要原因之一。因此，针对DME的病理生理学特异性治疗药物的开发是有限的，针对个别患者的个性化治疗选择仍然是主观的。我们的长期目标是开发自动化方法，利用视网膜成像技术将DME患者分成反映特定病理生理机制的亚组。反过来，我们预计根据这些机制进行分组将有助于为每个患者选择最佳的个性化治疗。目前的研究范式分离出三种不同的病理生理机制，独立或共同作为DME的致病因素：a)毛细血管内皮细胞功能障碍，b)视网膜胶质细胞泵功能障碍，c)视网膜色素上皮细胞泵功能障碍。基于这一范式，我们提出了两个相互关联的假设：1)荧光素血管造影(FA)和SD-OCT可以使用自动化算法进行定量分析，以推断特定的疾病机制。在FA上，弥漫性与灶性渗漏面积(D/F)的比率将反映两种泵功能障碍的DME亚型相对于毛细血管渗漏亚型的相对优势。在SD-OCT上，通过层分割可以识别黄斑增厚和其他指示弥漫性和局灶性DME的形态特征。2)结合D/F比的图像分析和SD-OCT的定量分析将成为预测疗效的生物标志物。更具体地说，弥漫性DME的FA和SD-OCT标记物对药物治疗的反应更好，而灶性DME的标记物对聚焦激光的反应更好。我们将通过追求以下三个具体目标来验证这些假设：目标一：开发自动化软件来量化FA和SD-OCT上的DME亚型成像生物标记物。目的II：使用存档的DME病例来改进和验证在AIM I中开发的自动化算法。目的III：进行一项试点试验，以确定D/F比率在预测“治疗初期”和无偏见人群中抗血管内皮生长因子应答的有效性。这个项目意义重大，因为对针对疾病亚型的个性化治疗的需求尚未得到满足。该项目将提供基于FA和SD-OCT的客观的DME亚型方法，并为不同的DME机制提供推断支持。这个项目在新技术和新知识方面具有创新性和影响力。我们将利用新的数学概念和开发算法，在临床环境中以自动方式可靠地测量DME成像生物标记物。开发的软件将免费分发给公众，并有望成为临床医生用于个性化治疗选择的标准方法，或图像读取中心用于对患者进行分层以进行新二甲基甲醚药物临床试验的标准方法。最后，我们期待 我们新颖的图像处理算法及其底层数学框架将对广泛的医学图像处理研究应用产生直接影响。 公共卫生相关性：糖尿病视网膜病变影响了大约400万美国人，是劳动年龄人群失明的主要原因。然而，导致失明的具体原因和对个别糖尿病患者的最佳治疗尚不清楚。我们将开发新的计算机辅助技术，以帮助更好地了解糖尿病视网膜病变的潜在机制，这反过来有望促进针对个人特定疾病机制的个性化治疗的最佳选择。

DESCRIPTION (provided by applicant): There are currently no well-established methods to identify and evaluate the mechanisms underlying diabetic macular edema (DME) pathobiology, one of the leading causes of blindness among working-age Americans. As such, the development of pathophysiology-specific therapeutic agents for DME is limited, and the selection of therapies personalized for individual patients remains subjective. Our long-term goal is to develop automated methods that exploit retinal imaging technologies to stratify DME patients into subgroups that reflect specific pathophysiological mechanisms. In turn, we expect that subgrouping according to these mechanisms will facilitate an optimal choice of personalized therapy for each patient. The current paradigm isolates three different pathophysiologic mechanisms, independently or together, as contributing factors to DME: a) capillary endothelial cell dysfunction, b) retinal glial cellular pump dysfunction, and c) retinal pigment epithelium cel pump dysfunction. We propose two interrelated hypotheses based on this paradigm: 1) Fluorescein angiography (FA) and SD-OCT can be quantitatively analyzed using automated algorithms to infer the specific disease mechanism. On FA, the diffuse to focal leakage area (D/F) ratio will reflect the relative predominance of the two pump dysfunction DME subtypes versus the capillary leakage subtype. On SD-OCT, macular thickening and other morphological features indicative of diffuse and focal DME can be identified through layer segmentation. 2) Image analysis using both the D/F ratio and quantitative analysis of SD-OCT will serve as predictive biomarkers for therapeutic responses. More specifically, the FA and SD-OCT markers of diffuse DME will respond better to pharmacotherapy, whereas the markers of focal DME will respond better to focal laser. We will test these hypotheses by pursuit of the following three specific aims: Aim I: Develop automated software to quantify DME subtype imaging biomarkers on FA and SD-OCT. Aim II: Use archived DME cases to refine and validate the automated algorithms developed in Aim I. Aim III: Perform a pilot trial to determine the efficacy of the D/F ratio in predicting anti-VEGF responsiveness in a "treatment na¿ve" and unbiased population. This project is significant because there is an unmet need for therapies personalized to disease subtype. This project will provide objective DME subtyping methods based on FA and SD-OCT and inferential support for different DME mechanisms. This project is innovative and impactful in terms of new technology and new knowledge. We will utilize novel mathematical concepts and develop algorithms to reliably measure DME imaging biomarkers in an automated fashion in a clinical setting. Developed software will be freely distributed to the public and are expected to become the standard methodology used by clinicians to personalize the choice of therapy or by image reading centers to stratify patients for clinical trials of new DME drugs. Finally, we expect that our novel image processing algorithms and their underlying mathematical frameworks will have an immediate impact on a wide spectrum of medical image processing research applications. PUBLIC HEALTH RELEVANCE: Diabetic retinopathy affects approximately 4 million Americans and is the leading cause of blindness among working-age people. However, the specific cause for blindness and optimal treatment for individual diabetic patients is unknown. We will develop novel, computer-aided technology to help better understand the underlying mechanisms of diabetic retinopathy, which in turn is expected to facilitate the optimal choice of therapy personalized for an individual's particular disease mechanism.