Artificial Intelligence for Assessment of Stargardt Macular Atrophy

项目摘要 Stargardt病是遗传性青少年黄斑变性最常见的形式。眼底 自体荧光（FAF）是一种广泛使用的成像技术，可以帮助诊断Stargardt 它通常用于监测疾病的进展。FAF成像提供视网膜神经元的体内测定。 这是一种间接的衡量，但只是一种间接的衡量。相比之下，谱域光学相干断层扫描（SD-OCT）， 提供视网膜微观结构的三维可视化，从而可以对其进行评估 直接和单独地在患有Stargardt病的眼睛中。在视网膜疾病终点会议与食品 2016年11月，美国食品药品监督管理局（FDA），一个可靠的测量完整性的解剖状态 的椭圆区（EZ）的视网膜，提出了一个潜在的合适的监管终点， 治疗干预临床试验。手动分割/识别EZ带，特别是在3-D中 OCT图像，已被证明是非常繁琐，耗时和昂贵的。自动物镜 分割技术，例如使用深度学习-人工智能（AI）构造的方法， 会有很大的价值此外，Stargardt病可能导致儿童和青少年严重的视力丧失， 成年人了早期预测Stargardt疾病进展可能有助于新的治疗试验。因此， 该提案开发了一种基于AI的方法，用于自动Stargardt萎缩分割和预测 FAF和OCT图像中的萎缩进展。更具体地说，我们首先记录纵向FAF和OCT 分别对图像进行处理，并将横截面FAF与OCT图像配准。然后我们开发一个2-D 使用AI方法和3-D方法从FAF图像进行Stargardt萎缩分割的方法 使用基于3-D图形的方法从OCT图像中分割EZ带。最后，基于AI的方法是 开发用于预测新的Stargardt萎缩的后续发展或现有萎缩的进展， 当前患者访视的OCT EZ带厚度和强度特征。

Project Abstract Stargardt disease is the most frequent form of inherited juvenile macular degeneration. Fundus autofluorescence (FAF) is a widely available imaging technique which may aid in the diagnosis of Stargardt disease and is commonly used to monitor its progression. FAF imaging provides an in vivo assay of the retinal layers, but is only an indirect measure. Spectral-domain optical coherence tomography (SD-OCT), in contrast, provides three-dimensional visualization of the retinal microstructure, thereby allowing it to be assessed directly and individually in eyes with Stargardt disease. At a retinal disease endpoints meeting with the Food and Drug Administration (FDA) in November of 2016, a reliable measure of the anatomic status of the integrity of the ellipsoid zone (EZ) in the retina, was proposed to be a potential suitable regulatory endpoint for therapeutic intervention clinical trials. Manual segmentation/identification of the EZ band, particularly in 3-D OCT images, has proven to be extremely tedious, time-consuming, and expensive. Automated objective segmentation techniques, such as an approach using a deep learning - artificial intelligence (AI) construct, would be of significant value. Moreover, Stargardt disease may cause severe visual loss in children and young adults. Early prediction of Stargardt disease progression may facilitate new therapeutic trials. Thus, this proposal develops an AI-based approach for automated Stargardt atrophy segmentation and the prediction of atrophy progression in FAF and OCT images. More specifically, we first register the longitudinal FAF and OCT enface images respectively, and register the cross-sectional FAF to OCT image. We then develop a 2-D approach for Stargardt atrophy segmentation from FAF images using an AI approach and a 3-D approach for EZ band segmentation from OCT images using a 3-D graph-based approach. Finally, an AI-based approach is developed to predict subsequent development of new Stargardt atrophy or progression of existing atrophy from the OCT EZ band thickness and intensity features of the current patient visit.