Predicting ECMO NeuroLogICal Injuries using mAchiNe Learning (PELICAN)

项目摘要 体外膜氧合（ECMO）是一种用于体外循环的形式， 重症儿童和成人，以支持心脏和肺时，传统的治疗失败。 迄今为止，全球已有超过79，762名儿童得到了支持，全球使用这一工具的人数 扩张ECMO和重症监护的进步改善了其他致命疾病的生存率 疾病，从而揭示神经损伤，这本身减少了50-60%的生存率，并导致 严重的长期神经系统疾病ECMO相关脑损伤机制的研究进展 我们对此知之甚少。现有的研究集中在评估离散元素，如 基础疾病、凝血异常、抗凝治疗或终末- 器官灌注是脑损伤的相关因素。以前的研究没有考虑到 临床事件的时间和动态因素，可能在脑的发生中发挥重要作用 很少有人探索哪些变量可以预测显著的神经系统损伤， 预先选择感兴趣的变量的偏差。机器学习是人工智能的一种 它采用算法直接从输入数据中以迭代方式发现模式： 在ECMO的背景下，它可以识别动态模式和变量之间的关系， 神经损伤 这项研究的长期目标是确定神经系统疾病的可修改的床边预测因子。 从而推动早期干预措施的发展，以改善神经功能 接受ECMO的儿童的结果。为了实现这一目标，我们建立了一个多- 具有临床和计算专业知识的学科团队。我们的核心假设是， ECMO患者SNI的风险预测模型可以基于生理和 在现实世界的临床环境中定期收集的实验室数据并且可以使用该模型 以确定SNI的参数进行潜在干预。该提案将利用先进的机器 学习算法以在大型多中心队列（0-18岁，n=750）中构建该预测模型。 在目标1中，我们将使用新的概率机器学习算法来训练和开发模型 通过神经影像学预测SNI。在目标2中，我们将使用 神经成像评分，并探索个性化的随时查询算法，预测时间 和SNI的类型。临床科学家和计算科学家之间的这一合作提案将奠定 通过确定可改变的风险因素，为神经保护性干预研究奠定基础， 改善ECMO幸存者的神经系统发病率和死亡率。

Project Summary Extracorporeal Membrane Oxygenation (ECMO) is a form of cardiopulmonary bypass used in critically ill children and adults to support the heart and lungs when conventional therapies fail. More than 79,762 children worldwide have been supported to date, and global use of this tool is expanding. Advances in ECMO and critical care have improved survival of otherwise fatal illnesses, thereby unmasking neurologic injury which itself reduces survival by 50-60% and leads to significant long-term neurologic morbidity. The mechanisms of ECMO-related cerebral injuries are poorly understood. Existing research focuses on evaluating discrete elements, such as underlying illness, coagulation abnormalities, anticoagulation management, or markers of end- organ perfusion as factors associated with brain injury. Prior studies have not considered the temporal and dynamic element of clinical events that may play a large role in the genesis of brain injury, and few have explored which variables could predict significant neurologic injury without the bias of pre-selecting variables of interest. Machine learning is a form of artificial intelligence that employs algorithms to discover patterns in an iterative manner directly from input data: in the context of ECMO, it can identify dynamic patterns and relationships between variables prior to neurologic injury. The long-term goal of this research is to identify modifiable bedside predictors of neurologic impairment and thereby drive the development of early interventions to improve neurologic outcomes of children undergoing ECMO. Towards this goal, we have assembled a multi- disciplinary team with clinical and computational expertise. Our central hypothesis is that a robust risk predictive model for SNI in ECMO patients can be developed based on the physiological and laboratory data routinely collected in real-world clinical settings and that this model can be used to identify parameters of SNI for potential intervention. This proposal will utilize advanced machine learning algorithms to build this prediction model in a large multicenter cohort (0-18 years, n=750). In Aim 1, we will use novel probabilistic machine learning algorithms to train and develop a model to predict SNI by neuroimaging. In Aim 2, we will validate and refine the model from Aim 1 using neuroimaging scores and explore a personalized anytime query algorithm that predicts the timing and type of SNI. This collaborative proposal between clinical and computational scientists will lay the groundwork for a neuroprotective interventional study by identifying modifiable risk factors to improve the tragically high neurologic morbidity and mortality in ECMO survivors.