Predicting the risk of glaucoma from structural, functional, and genetic factors using artificial intelligence

利用人工智能从结构、功能和遗传因素预测青光眼风险

• 批准号: 10597998
• 负责人: Siamak Yousefi
• 金额: $ 44.39万
• 依托单位: UNIVERSITY OF TENNESSEE HEALTH SCI CTR
• 依托单位国家: 美国
• 财政年份: 2022
• 资助国家: 美国
• 起止时间: 2022-04-01 至 2027-03-31
• 项目状态: 未结题
• 来源: https://reporter.nih.gov/project-details/10597998
• 关键词: Address; Affect; Anatomy; Artificial Intelligence; Atlases; Axon; Blindness; Caring; Clinical; Clinical Trials; Complex; Country; Cox Proportional Hazards Models; Data; Developing Countries; Development; Diagnosis; Disease; Elderly; Evaluation; Eye; Functional disorder; Fundus; Future; Genetic; Genetic Markers; Glaucoma; Growth; Hereditary Disease; High Prevalence; Image; Incidence; Individual; Machine Learning; Maps; Modality; Modeling; Nerve Degeneration; Ocular Hypertension; Older Population; Ophthalmologist; Optic Disk; Outcome; Participant; Patients; Pattern; Persons; Phenotype; Population; Populations at Risk; Prevalence; Primary Open Angle Glaucoma; Resolution; Resources; Retina; Risk; Risk Factors; Screening procedure; Side; Single Nucleotide Polymorphism; Specific qualifier value; Surrogate Endpoint; Testing; Validation; Visual Fields; Work; aging population; archetypal analysis; care burden; clinical care; clinical practice; clinically relevant; deep learning; deep learning model; demographics; endophenotype; experience; field study; genome wide association study; high risk; hypertension treatment; improved; machine learning model; multimodality; novel; novel marker; optical disc; population based; predictive modeling; prevent; retinal ganglion cell degeneration; retinal nerve fiber layer; risk prediction; screening; sight restoration; statistical and machine learning; statistical learning; tool; usability

Glaucoma is a complex neurodegenerative blinding disease that causes the degeneration of retinal ganglion cells and their axons. The prevalence of glaucoma is projected to increase by almost 50% over the next two decades as older people making up the fastest growing part of the global population. The burden of glaucoma care will therefore continue to grow, without a competing increase in the number of ophthalmologists or available resources. As a result, the required demand for glaucoma care will likely exceed capacity and resources leading to prioritizing care for those patients at highest risk of vision loss. There is no concrete evidence in support of an individual test, or group of tests, that show superiority for identifying people at-risk of developing glaucoma or those at higher risk of glaucoma progression. Glaucoma risk factors are too insensitive in identifying individuals who will likely develop glaucoma. Fundus photographs lack detailed and high-resolution information of the optic disc and surrounding retinal nerve fiber layer for glaucoma assessment and visual field tests provide surprisingly inconsistent and variable results, especially in subclinical glaucoma and in patients with more severe visual field loss (both sides of glaucoma spectrum). Although glaucoma is a highly inheritable disease, genetic factors yet explain only slight segment of all glaucoma. Reliable and accurate models for detecting individuals at higher risk of visual loss is an unmet need. We propose to use artificial intelligence (AI) constructs to discover visual field and imaging signatures of glaucoma and synthesize these signatures with classic risk factors and genetic data to identify individuals at-risk of developing glaucoma and future vision loss. The central hypothesis of this proposal is that AI applied to fundus photographs, visual fields and genetic factors may recognize and quantify the glaucoma-induced signs, yielding better signatures for glaucoma development and vision loss compared to current subjectively specified or conventionally identified features. As such, we will develop AI models to predict glaucoma from fundus photographs and visual fields then extract fundus and visual field endophenotypes (signatures) of glaucoma. We will then develop genome-wide association study (GWAS) and machine learning models to address underpower GWAS limitation and develop AI models to predict glaucoma from identified genetic markers. We finally develop an AI construct to synthesizes the discovered fundus and visual field signatures with classic glaucoma risk factors and genetic data to predict glaucoma. This AI construct can work with any or all of these modalities as well thus providing a potential tool for screening purposes as well. To achieve these objectives, we have assembled a team of interdisciplinary experts with access to large clinically annotated multi-modal glaucoma data. Our proposed studies will potentially uncover novel genetic factors of glaucoma as well as visual field and imaging endophenotypes of glaucoma that may serve as surrogate endpoints to improve glaucoma clinical trials and offer improvements in identifying individuals at-risk of developing glaucoma and future vision loss.

青光眼是一种复杂的神经退行性致盲性疾病，它导致视网膜神经节变性 细胞和它们的轴突青光眼的患病率预计将在未来两年增加近50%， 几十年来，老年人是全球人口中增长最快的部分。青光眼的危害 因此，护理将继续增长，而不会增加眼科医生的数量， 现有资源。因此，对青光眼护理的需求可能会超过能力， 资源导致优先照顾那些患者在视力丧失的最高风险。没有具体 支持单个测试或一组测试的证据，这些测试显示出识别有风险的人的优势。 发展中的青光眼或青光眼进展风险较高的患者。青光眼的危险因素也 在识别可能患青光眼的个体方面不敏感。眼底照片缺乏细节， 高分辨率视盘和周围视网膜神经纤维层信息用于青光眼评估 和视野测试提供了令人惊讶的不一致和可变的结果，特别是在亚临床青光眼中 并且在具有更严重的视野丧失的患者中（青光眼光谱的两侧）。虽然青光眼是一种 高度遗传性疾病，遗传因素仅能解释所有青光眼中的一小部分。可靠和 用于检测具有较高视力丧失风险的个体的精确模型是一种未满足的需求。我们建议使用 人工智能（AI）构建，以发现青光眼的视野和成像特征， 这些签名与经典的风险因素和遗传数据，以确定个人在发展青光眼的风险 和未来视力丧失。这项提议的中心假设是，人工智能应用于眼底照片，视觉 磁场和遗传因素可以识别和量化脑卒中引起的症状， 与当前主观规定或常规规定相比， 识别特征。因此，我们将开发人工智能模型，从眼底照片预测青光眼， 然后，视野提取青光眼的眼底和视野内表型（特征）。然后我们将 开发全基因组关联研究（GWAS）和机器学习模型，以解决GWAS不足的问题 限制和开发AI模型来预测青光眼从确定的遗传标记。我们终于开发出一种人工智能 将发现的眼底和视野特征与经典青光眼风险因素相结合 和遗传数据来预测青光眼。这个人工智能结构也可以与任何或所有这些模态一起工作 因此也提供了用于筛选目的的潜在工具。为了实现这些目标，我们聚集了 一个跨学科的专家团队，可以访问大量临床注释的多模式青光眼数据。 我们提出的研究将潜在地揭示青光眼以及视野的新遗传因素， 青光眼的内表型成像可作为替代终点，以改善青光眼临床 试验，并提供改进，以确定个人的风险发展青光眼和未来的视力丧失。