Cerebral Palsy Risk Identification System

项目总结 新生儿科医生经常被要求识别出神经发育不良的婴儿。 结果，包括脑性瘫痪(CP)。脑性瘫痪是美国儿童中最常见的运动障碍 并与风险因素有关，包括胎龄低体重、早产和中风。尽管核磁共振 头颅超声为早产儿提供了有价值的结构信息，对早产儿的预后有一定的预测作用 早期CP风险识别的准确性。在过去的20年里，大量的研究证实了这种临床潜力。 一般运动评估(GMA)用于早期CP风险识别，文献中已达成共识 GMA提供最高的精确度。跨34-48个阶段的第一阶段“狭小的同步”一般动作(CSGM) “扭动”阶段和第二阶段“强迫、自愿运动”阶段的胎龄(GA) 跨度为50-59周的GA对CP的发展表现出高度的敏感性和特异性，联合范围 由训练有素的专家执行时，从80%到98%。 尽管GMA具有潜力，但它在极少数临床中心可用，因为采用和常规应用取决于 关于是否有训练有素的GMA评分员进行冗长而昂贵的床边观察或视频审查- 基于评分和手动报告创建。提出了一种脑瘫风险识别系统(CPRIS)，该系统将 成为第一个将GMA自动化用于日常应用的公司。CPRIS构成了下一代方法，将 用客观、系统、有效的运动模式取代评分者视觉格式塔，从根本上改变GMA 分类。此外，CPRIS潜在地提供了一种通知和评估新出现的STEM的效果的手段 以细胞为基础的早期发育连续性CP的干预。 成功实施第一阶段和第二阶段将完成外形小巧、可移动、高度自动化 用于脑瘫风险识别的预生产系统，可供工作人员、临床医生和健康人员轻松应用 护理人员无需任何形式的手动后处理操作或视频文件传输。一个完整的 公用事业公司将支持GMA创建和与电子健康记录(EHR)系统共享报告。一项申请- 特定的、完全集成的设备将实现最高程度的标准化，从而提高数据质量。 在两个显著的3级NICU的实地研究中，婴儿的动作将使用“RGB-D”或3D来获得 “深度”摄像机与特定应用和阶段的“深度-流”卷积神经网络相结合 (CNN)分类器方法，不需要婴儿接触(与运动学方法相比)，并捕获完整的- 身体动作。这一努力标志着首次利用这种技术使GMA自动化。结果将会被比较 对高危早产儿样本中的高级GMA评分者进行共识测定，包括阶段1和阶段2。 新方法的可行性将由ROC-AuC分析确定，成功的门槛为≥0.90 精确度。总体成果将由下列领域的公认专家组成的咨询委员会进行评估 新生儿科、儿科、脑瘫、GMA和生物统计学。

PROJECT SUMMARY Neonatologists are often required to identify infants who are likely to suffer poor neurodevelopmental outcomes, including Cerebral Palsy (CP). CP is the most common motor disability among children in the United States and is associated with risk factors including low weight for gestational age, premature birth, and stroke. Although MRI and cranial ultrasound provide valuable structural information in the preterm period, they have moderate predictive accuracy for early CP risk identification. Over the past 20 years, numerous studies have validated the clinical potential of General Movement Assessment (GMA) for early CP risk identification and there is consensus in the literature that GMA offers the highest accuracy. Stage 1 “cramped synchronized” general movements (CSGMs) spanning 34-48 weeks gestational age (GA) during the “writhing movements” period and Stage 2 “forced, voluntary movements” spanning 50-59 weeks GA have demonstrated high sensitivity and specificity for developing CP, conjointly ranging from 80%-98% when performed by extensively trained experts. Despite its potential, GMA is available in very few clinical centers, as adoption and routine application depend on the availability of highly trained GMA raters to perform lengthy and costly bedside observations or video review- based scoring and manual report creation. A Cerebral Palsy Risk Identification System (CPRIS) is proposed that will be the first to automate GMA for routine application. The CPRIS constitutes a next-generation approach that will fundamentally transform GMA by replacing rater visual gestalts with objective, systematic, validated movement pattern classification. Further, the CPRIS potentially offers a means of informing, and assessing the efficacy of emerging stem cell-based interventions for CP along the early developmental continuum. Successful implementation of Phase I&II will complete a small form factor, mobile, highly automated preproduction system for cerebral palsy risk identification that can be readily applied by staff, clinicians, and health care provider personnel without any form of manual post-processing operations or video file transfer. An integrated utility will support GMA creation and report sharing with Electronic Health Record (EHR) systems. An application- specific, fully integrated device will achieve the highest degree of standardization and thus data quality. In a field study at two prominent Level 3 NICUs, infant movements will be acquired using an “RGB-D”, or 3D “depth” camera in conjunction with an application- and stage-specific “Depth-Flow” convolutional neural network (CNN) classifier approach, that requires no infant contact (contrasting with kinematic methods) and captures whole- body movements. This effort marks the first utilization of such technology to automate GMA. Results will be compared to consensus determinations of advanced GMA raters in a sample of high risk preterm infants at both Stages 1 & 2. Viability of the new approach will be determined by ROC-AUC analyses, with a threshold for success of ≥ 0.90 accuracy. Overall results will be evaluated by an Advisory Committee of recognized experts in the fields of neonatology, pediatrics, cerebral palsy, GMA and biostatistics.