Data driven public health approaches for diabetic retinopathy and age-related macular degeneration

该研究金将包括基于昆士兰大学公共卫生中心研究优势的数据驱动项目，重点关注电子健康记录（EHR）和视网膜成像数据的使用，以改善人口的眼睛健康。开发慢性眼病的新结果测量：在英国，年龄相关性黄斑变性（AMD）和青光眼是最常见的威胁视力的眼病，如果疾病进展加速，需要定期监测和快速治疗。眼部结构或视觉功能的变化都可能表明病情正在恶化。以光学相干断层扫描（OCT）技术为中心的3D视网膜成像技术取得了快速进展，该技术可以以前所未有的细节解析眼部结构。然而，缺乏充分利用这些信息的分析方法，特别是当试图将视网膜的结构和功能变化联系起来时。该研究员将开发新的统计方法，将AMD患者的大型视网膜成像数据集与从电子病历中提取的视觉功能测量数据整合起来。这项工作将与昆士兰大学眼科医生和OCT专家（由Tunde Peto教授和Ruth Hogg博士领导的小组）以及伦敦城市大学的统计学家一起进行。目的是为新疗法的临床试验开发有意义的AMD进展结果测量。优化糖尿病视网膜病变筛查：糖尿病视网膜病变（DR）是导致工作年龄人群视力丧失的最常见原因之一，当高血糖损害视网膜血管时就会发生。对那些有DR风险的人定期拍摄视网膜照片进行筛查。图像是手动分级的具体变化的存在，以确定那些需要治疗。该项目的目的是探索将自动图像分析整合到北爱尔兰DR筛查计划中的潜力，以更有效地针对治疗并降低成本。将发展新的分析方法，以充分利用方案大量筛选库中所载的资料。一个关键的挑战是预测哪些患者会在短期内发展为威胁视力的DR。治疗可以针对这一群体，而不是针对在多次筛查中处于疾病中期稳定的许多患者。对于病情稳定的患者，准确的进展预测也可以为基于风险的筛查提供信息，延长筛查间隔，减少人群的筛查总次数和相关费用。目前的方法很难预测DR的进展。可能有一些细微的视网膜变化模式可以预测DR的进展，只有使用可以同时分析数千名患者数据的自动化方法才能检测到。最新一代的机器学习技术（深度学习算法）几乎可以与人类评分员在视网膜图像中检测DR的能力相匹配。下一步将是确定这些技术是否可以应用于预测DR的进展，利用图像序列中的全套信息。这些数据将来自北爱尔兰DR筛查项目数据库（临床负责人，Peto教授），这是一个独特的视网膜图像存储库，大约有87,000名患者，最近被集中起来，并连接到2002年。该研究员将与Peto教授和伦敦国王学院的数学家合作，开发最新的机器学习技术，并将其应用于大量筛选图像，以检测预测DR进展的新特征。北爱尔兰是这项研究的理想地点，因为老年人很少迁移，所以比英国其他地区更容易监测患者的结果。将评估自动化方法的性能以及改进筛选程序的潜力。

This fellowship will consist of data-driven projects building on the research strengths of the Centre for Public Health, QUB, focusing on use of electronic health records (EHR) and retinal imaging data to improve population eye health. Developing novel outcome measures for chronic eye disease:Management of the most common sight-threatening eye diseases in the UK, age-related macular degeneration (AMD) and glaucoma, requires regular monitoring and rapid treatment if disease progression accelerates. Changes in either ocular structures or visual function can signal that progression is occurring. There have been rapid advances in 3D retinal imaging technology, centred on a technique known as optical coherence tomography (OCT), that can resolve ocular structures in unprecedented detail. However, analytical methods to make full use of this information are lacking, especially when attempting to link structural and functional changes in the retina. The fellow will develop novel statistical methods to integrate a large retinal imaging dataset of AMD patients with measurements of visual function drawn from EHRs. This work will be conducted with QUB ophthalmologists and OCT experts (groups led by Prof Tunde Peto and Dr Ruth Hogg) and statisticians at City, University of London. The aim is to develop meaningful outcome measures of AMD progression for use in clinical trials of new treatments. Optimising diabetic retinopathy screening:Diabetic retinopathy (DR), one of the most common causes of sight loss among working-age people, occurs when high blood sugar damages blood vessels in the retina. Those at risk of DR are screened with retinal photographs taken at regular intervals. Images are manually graded for the presence of specific changes to identify those in need of treatment. The aim of this project is to explore the potential of integrating automated image analysis into the Northern Ireland DR screening programme to target treatment more effectively and reduce costs. New analytical approaches will be developed to fully exploit information contained within the programme's substantial screening libraries. A key challenge is predicting which patients will progress to sight-threatening DR in the short term. Treatment could be targeted towards this group, rather than towards the many patients that remain stable in the intermediate stages of the disease across multiple screenings. Accurate prediction of progression could also inform risk-based screening with longer screening intervals for stable patients, reducing the overall number of screening visits for the population and the associated costs.DR progression is difficult to predict using current methods. There may be subtle patterns of retinal changes predicting DR progression detectable only using automated approaches that can analyse data from thousands of patients simultaneously. The latest generation of machine learning techniques (deep learning algorithms) can almost match the ability of human graders to detect DR in retinal images. The next step will be to determine whether these techniques can be applied to predict progression of DR, leveraging the full set of information within image sequences. These will be drawn from the Northern Ireland DR screening programme databank (clinical lead, Prof Peto), a unique repository of retinal images for approximately 87,000 patients that has recently been centralised and linked backed as far as 2002. Working with Prof Peto and mathematicians at King's College London, the fellow will develop and apply the latest machine learning techniques to a large set of screening images to detect novel features predictive of DR progression. Northern Ireland is an ideal for this study as there is little migration among older people so patient outcomes can be monitored more easily than in other parts of the UK. Performance of the automated methods will be assessed along with the potential for improvements to the screening programme.