Built Environment, Pedestrian Injuries and Deep Learning (BEPIDL) Study

项目总结 道路交通伤害是全球疾病负担的主要贡献者，有近130万人死亡 全球每年有多达5000万人受伤，低收入和中等收入的行人和骑自行车的人 受影响最大的国家(LMIC)。建筑环境的道路基础设施(例如人行道)， 社区设计(如街道连通性)和城市发展(如城市蔓延)是关键 行人受伤风险的决定因素。在LMIC，糟糕的道路基础设施和社区设计 被认为是道路交通伤亡人数上升的重要因素，但有 很少有研究系统地识别和量化建筑环境的具体特征 在这些环境中导致机动车相撞。在LMIC城市内部，往往存在很大的差距 基础设施得到改善，反映社会经济特征，导致道路卫生不平等 交通事故致伤。LMIC中关于建成环境的地理参考数据的匮乏使得道路成为研究的焦点 交通伤害更加困难，尽管计算机视觉和图像分析的最新进展结合了Big 公开可用的、地理参考的全球道路图像数据(例如，Google Street View，GSV)可以提供帮助 克服数据匮乏以及在建筑物上收集和分析数据的成本和时间限制 LMIC中的环境。自动图像分析在很大程度上是通过深度学习实现的，深度学习是一个子领域 并依靠训练神经网络来检测和标记特定的 图像中的对象。这些方法可以大幅降低全市建筑环境的障碍，并 在LMIC城市进行交通安全研究，从而大大提高了研究能力和推广能力。我的 职业目标是成为全球城市健康的独立调查者，专注于道路安全和 LMIC中的建筑环境。我建议对应用的深度学习方法进行研究和培训 对于哥伦比亚波哥大的公共卫生：1)开发神经网络以创建BE数据库 从图像数据中获取道路基础设施的特征，并从这些特征中创建邻里类型； 2)评估邻域水平的BE特征和类型与行人碰撞之间的关联 3)社区社会环境关联性评价 行人碰撞和死亡的特征、感知和BE特征和类型。我是 寻求额外的培训：1)培养应用于公共卫生的深度学习方法的能力；2) 创建健康和建成环境的邻里指标和类型；3)应用贝叶斯 时空模型，以了解邻居特征和类型如何影响健康；4) 培养多国合作、赠款撰写和监督LMICs研究项目的技能。

PROJECT SUMMARY Road traffic injuries are a major contributor to the burden of disease globally with nearly 1.3 million deaths globally and as many as 50 million injured annually with pedestrians and cyclists in low and middle-income countries (LMICs) among the most affected. Road infrastructure of the built environment (e.g., sidewalks), neighborhood design (e.g., street connectivity) and urban development (e.g., urban sprawl) are key determinants of the risk of pedestrian injuries. In LMICs, poor road infrastructure and neighborhood design are acknowledged as being important contributors to rising numbers of road traffic injuries and deaths, but there are few studies systematically identifying and quantifying what specific features of the built environment are contributing to motor vehicle collisions in these settings. Within LMIC cities, there are often large disparities where infrastructure is improved that reflect socioeconomic characteristics, leading to health inequities in road traffic injury. The paucity of georeferenced data on the built environment in LMICs has made research on road traffic injuries more difficult, though recent advances in computer vision and image analysis combined with Big Data of publicly available, georeferenced, images of roads worldwide (e.g., Google Street View, GSV) can help overcome the paucity of data and the cost and time limitations of collecting and analyzing data on the built environment in LMICs. Automated image analysis has largely been made possible via deep learning, a subfield of artificial intelligence and machine learning and relies on training neural networks to detect and label specific objects within images. These methods can drastically reduce the barriers to citywide built environment and traffic safety research in LMIC cities, thus substantially increasing research capacity and generalizability. My career goal is to become an independent investigator in global urban health with a focus on road safety and the built environment in LMICs. I propose undertaking research and training in deep learning methods applied to public health in the setting of Bogota, Colombia: 1) Develop neural networks to create a database of BE features of the road infrastructure from image data and to create neighborhood typologies from those features; 2) Assess the association between neighborhood-level BE features and typologies and pedestrian collisions and fatalities and road safety perceptions; 3) Assess the association of neighborhood social environment characteristics with pedestrian collision and fatalities, perceptions, and BE features and typologies. I am seeking additional training in 1) developing competency in deep learning methods applied to public health; 2) creating neighborhood indictors and typologies of health and the built environment; 3) applying Bayesian spatiotemporal models to understand how neighborhood characteristics and typologies influence health; 4) develop skills in multi-country collaboration, grant writing and overseeing research projects in LMICs.