Deep Learning Approaches for Personalized Modeling and Forecasting of Glaucomatous Changes

用于青光眼变化个性化建模和预测的深度学习方法

• 批准号: 10533641
• 负责人: HIROSHI ISHIKAWA
• 金额: $ 25.14万
• 依托单位: OREGON HEALTH & SCIENCE UNIVERSITY
• 依托单位国家: 美国
• 财政年份: 2020
• 资助国家: 美国
• 起止时间: 2020-02-01 至 2025-01-31
• 项目状态: 未结题
• 来源: https://reporter.nih.gov/project-details/10533641
• 关键词: Deep; Learning; Approaches; Personalized; Modeling

Project Summary Glaucoma is a leading cause of vision morbidity and blindness worldwide. Early disease detection and sensitive monitoring of progression are crucial to allow timely treatment for preservation of vision. The introduction of ocular imaging technologies significantly improves these capabilities, but in clinical practice there are still substantial challenges at managing the optimal care for individual cases due to difficulties of accurately assessing the potential progression and its speed and magnitude. These difficulties are due to a variety of causes that change over the course of the disease, including large inter-subject variability, inherent measurement variability, image quality, varying dynamic ranges of measurements, minimal measurable level of tissues, etc. In this proposal, we propose novel agnostic data-driven deep learning approaches to detect glaucoma and accurately forecast its progression that are optimized to each individual case. We will use state- of-the-art automated computerized machine learning methods, namely the deep learning approach, to identify structural features embedded within OCT images that are associated with glaucoma and its progression without any a priori assumptions. This will provide novel insight into structural information, and has shown very encouraging preliminary results. Instead of relying on the conventional knowledge-based approaches (e.g. quantifying tissues known to be significantly associated with glaucoma such as retinal nerve fiber layer), the proposed cutting-edge agnostic deep learning approaches determine the features responsible for future structural and functional changes out of thousands of features autonomously by learning from the provided large longitudinal dataset. This program will advance the use of structural and functional information obtained in the clinics with a substantial impact on the clinical management of subjects with glaucoma. Furthermore, the developed methods have potentials to be applied to various clinical applications beyond glaucoma and ophthalmology.

项目摘要 青光眼是世界范围内视力发病和失明的主要原因。早期疾病检测和 敏感地监测进展对于及时治疗以保护视力是至关重要的。的 眼成像技术的引入显著提高了这些能力，但在临床实践中， 在管理个别病例的最佳护理方面仍然存在重大挑战， 准确评估潜在的进展及其速度和幅度。这些困难是由于 在疾病过程中发生变化的各种原因，包括受试者间的大变异性，固有的 测量可变性、图像质量、测量的变化动态范围、 在这项提案中，我们提出了新的不可知数据驱动的深度学习方法来检测 青光眼和准确地预测其进展，优化每个个案。我们将使用国家- 最先进的自动化计算机化机器学习方法，即深度学习方法， OCT图像中与青光眼及其进展相关的结构特征 没有任何先验假设。这将为结构信息提供新的见解，并已显示出非常 初步成果令人鼓舞。而不是依赖于传统的基于知识的方法（例如， 定量已知与青光眼显著相关的组织，例如视网膜神经纤维层）， 提出的尖端不可知的深度学习方法确定了未来负责的功能， 通过学习提供的功能，自主地从数千个功能中进行结构和功能更改， 大型纵向数据集。该计划将促进所获得的结构和功能信息的使用 在临床上对青光眼患者的临床管理具有实质性影响。而且 所开发的方法具有应用于青光眼以外的各种临床应用的潜力， 眼科