An Automated Early Motor Development Risk Screener from Observational Video Recordings of Infants and Toddlers

运动发育在婴儿的全面发展中起着举足轻重的作用，并对社会、 认知、记忆和语言能力。这些功能可能会在运动开发被破坏时受到影响 运动障碍与全身性发育迟缓有关。早期筛查和 识别延迟可以进行早期干预，目标不仅是改善运动功能，而且是整体 其他领域的发展。独立或电动爬行的经验表明，可以改善空间 婴儿的记忆力。积极的运动训练，甚至早在3个月前，就被证明在目标上取得了进步 在培训课程结束后很长一段时间内持续的探索和社会参与，表明及早 干预。 接受发育筛查的儿童更有可能被确认为发育迟缓并及早接受筛查 干预服务与那些只接受适合年龄的里程碑检查作为儿科部分的服务相比较 好的-孩子们的来访。然而，全面的运动筛查，使用标准化的仪器，用于发育 监督是劳动密集型的，需要在管理和评估方面进行专门培训。短缺 经常执行黄金标准的全面运动评估的资源可能会导致遗漏 为早期干预提供机会，并对有行为风险的婴儿的终生结局产生负面影响 和发育障碍。在美国，15%-20%的儿童有发育或行为障碍，但 只有不到三分之一的人在入学前被确诊，这阻碍了早期干预的机会。 在大脑发育的关键时期。 在第一阶段，我们构建了一个原型人类动作识别引擎(HARE)来自动提取婴儿运动 从视频中移动；然后利用该引擎来演示自动化开发的可行性 通过利用机器学习的最新进展来实现最先进的风险筛选器 对婴儿运动发育的评估。在第二阶段，我们的目标是建立风险筛选程序SCOREIT，作为 临床上可部署的、经济高效的发育风险评估工具，以确定哪些儿童需要进一步 临床随访。SCOREIT有可能通过带来力量来改变早期发育筛查 全面的运动筛查，通常只对使用专门资源的高危儿童进行， 服务不足和经济脆弱的社区。不到三分之一的美国儿童接受发育 放映。SCOREIT可以为目前70%以上没有接受风险筛查的儿童带来风险筛查， 增加行为和发育障碍的早期干预机会。此外， Hare系统能够从视频中自动提取和量化运动动作，这有助于 探索行为和发育障碍的早期标志，极大地简化了手动视频 编码负担和加快发现的步伐。

Motor development plays a pivotal role in an infant’s overall development and has a cascading effect on social, cognitive, memory, and language ability. These functions can be compromised when motor development is disrupted; motor disruption is associated with generalized developmental delays. Early screening and identification of delays can allow early interventions with the goal of not just improving motor function but overall development in other domains. Experience with independent or motorized crawling is shown to improve spatial memory in infants. Active motor training, even as early as 3 months, is shown to achieve gains in object exploration and social engagement that persist long after the training sessions, indicating the benefits of early intervention. Children who get developmental screening are more likely to be identified with delays and receive early intervention services compared to those that only receive age-appropriate milestone checks as part of pediatric well-child visits. However, comprehensive motor screening, using standardized instruments, for developmental surveillance is labor intensive and requires specialized training in administration and evaluation. Shortage of resources to perform frequent, gold-standard, comprehensive motor assessments can lead to missed opportunities for early intervention and contribute negatively to lifelong outcomes for infants at risk for behavioral and developmental disorders. In the US, 15-20% of children have a developmental or behavioral disability, but less than a third of them get diagnosed before entering school, preventing opportunities for early intervention during a critical period of brain development. In Phase I, we built a prototype Human Action Recognition Engine (HARE) to automate extraction of infant motor movement from video; then utilized this engine to demonstrate the feasibility of an automated developmental risk screener by leveraging recent advances in machine learning to achieve state of the art accuracies in assessment of infant motor development. In Phase II, we aim to build out the risk screener, SCOREIT, as a clinically deployable, cost-effective, developmental risk assessment tool to identify children who need further clinical follow-up. SCOREIT has the potential to transform early developmental screening by bringing the power of comprehensive motor screening, usually administered only to at-risk children utilizing specialized resources, to underserved and economically-fragile communities. Less than a third of US children receive developmental screening. SCOREIT can bring risk screenings to the over 70% of children that currently do not receive them, increasing opportunities for early intervention for behavioral and developmental disorders. Additionally, the HARE system’s ability to automatically extract and quantify motor movements from video can be an enabler for explorations of early markers for behavioral and developmental disorders, significantly easing manual video coding burdens and accelerating the pace of discoveries.