Clinical and genetic analysis of retinopathy of prematurity

项目摘要 本项目的长期目标是建立早产儿视网膜病变（ROP）的定量框架 基于临床、影像学、遗传学和信息学原理的护理。在上一个资助期内， 开发了用于ROP诊断的人工智能方法，但由于缺乏 前瞻性验证，以及将这些系统视为无法解释其原理的“黑匣子” 进行诊断。此外，虽然生物医学研究数据正在以巨大的速度产生， 在整合从基因组学到基因组学的各种不同的科学发现方面， 影像学应用于临床医学。这一更新将解决目前在这些领域的知识差距。我们的整体 假设是使用人工智能和分析来开发ROP护理的定量框架 将改善临床疾病管理，建立“可解释的”人工智能系统将提高 临床接受和教育机会，以及临床，成像， 环境和遗传的发现，将提高对疾病发病机制和风险的理解。 这些假设将使用三个具体目标进行测试：（1）评估人工 智能化ROP诊断和筛查系统。这将包括：（a）招募一个目标， 2000次眼睛检查，包括来自5个中心的375名受试者的广角视网膜图像，（B）优化图像 质量检测算法，我们最近开发的，和（c）分析系统的准确性，为ROP诊断 和筛选（使用新的定量血管严重程度量表）。(2)提高我们的可解释性 用于ROP诊断的现有人工智能方法。这将包括：（a）增加“可解释性”， 通过将深度学习与传统的特征提取方法相结合，（B）开发神经网络， 识别序列图像之间的变化，以及（c）通过系统反馈评估这些方法， 专家(3)开发ROP发病机制和风险的综合模型。这将包括：（a）建立和改进 基于临床、图像和人口统计学特征的ROP风险预测模型，以及（B）整合遗传、 通过机器学习进行遗传风险预测的成像、临床和环境变量， 调查与遗传变异和遗传风险评分的因果关系，并通过将SNP 与基因表达测量相关，以鉴定ROP的功能基因。最终，这些 研究将大大减少临床医生采用人工智能等技术的障碍， 并将展示一个健康信息管理的原型， 表型数据该项目将由一个多学科的调查小组进行， 成功地在一起近10年，谁拥有眼科，生物医学信息学， 计算机科学、计算生物学、眼科遗传学、遗传分析和统计遗传学。

Project Summary The long-term goal of this project is to establish a quantitative framework for retinopathy of prematurity (ROP) care based on clinical, imaging, genetic, and informatics principles. In the previous grant period, we have developed artificial intelligence methods for ROP diagnosis, but real-world adoption has been limited by lack of prospective validation and by perception of these systems as “black boxes” that do not explain their rationale for diagnosis. Furthermore, although biomedical research data are being generated at an enormous pace, much less work has been done to integrate disparate scientific findings across the spectrum from genomics to imaging to clinical medicine. This renewal will address current gaps in knowledge in these areas. Our overall hypotheses are that developing a quantitative framework for ROP care using artificial intelligence and analytics will improve clinical disease management, that building “explainable” artificial intelligence systems will enhance clinical acceptance and educational opportunities, and that analysis of relationships among clinical, imaging, environmental, and genetic findings, in ROP will improve understanding of disease pathogenesis and risk. These hypotheses will be tested using three Specific Aims: (1) Evaluation performance of an artificial intelligence system for ROP diagnosis and screening prospectively. This will include: (a) recruit a target of over 2000 eye exams including wide-angle retinal images from 375 subjects at 5 centers, (b) optimize an image quality detection algorithm we have recently developed, and (c) analyze system accuracy for ROP diagnosis and screening (using a novel quantitative vascular severity scale). (2) Improve the interpretability of our existing artificial intelligence methods for ROP diagnosis. This will include: (a) increase “explainability” of systems by combining deep learning with traditional feature extraction methods, (b) develop neural networks to identify changes between serial images, and (c) evaluate these methods through systematic feedback by experts. (3) Develop integrated models for ROP pathogenesis and risk. This will include: (a) build and improve ROP risk prediction models based on clinical, image, and demographic features, and (b) integrate genetic, imaging, clinical, and environmental variables through genetic risk prediction by machine learning, by investigating casual relationships with genetic variants and genetic risk scores, and by incorporating SNP associations with gene expression measurements to identify functional genes of ROP. Ultimately, these studies will significantly reduce barriers to adoption of technologies such as artificial intelligence for clinicians, and will demonstrate a prototype for health information management which combines genotypic and phenotypic data. This project will be performed by a multi-disciplinary team of investigators who have worked successfully together for nearly 10 years, and who have expertise in ophthalmology, biomedical informatics, computer science, computational biology, ophthalmic genetics, genetic analysis, and statistical genetics.