Neural Substrate of Outcomes after Neonatal Hypoxic Ischemic Encephalopathy

摘要 新生儿缺氧缺血性脑病（HIE）造成的脑损伤影响1-5/1000足月出生的婴儿。HIE 通常会导致不良后果，定义为死亡或认知贝利婴儿发育量表（cBSID）<85 2年。目前的治疗试验正在测试是否晚期，更深，和/或更长版本的标准 治疗性低温可进一步减少2年不良结局。然而，治疗创新是 缓慢且不确定，因为确切的神经解剖学基底（即，结果背后的解剖结构） 目前尚不清楚，缺乏可靠的工具来预测HIE后2年的不良结局。了解 神经基质，从而预测不良的2年结果可以：（a）在新生儿治疗前， 阶段，确定那些高风险的患者，以发展不良的2年结果，所以我们的治疗试验可以集中在 对那些高风险患者进行治疗，并避免对低风险患者进行不必要的治疗;以及（B）在治疗后， 早期（新生儿阶段）评估治疗效果，避免等到2年才观察结果 从而加速治疗创新。脑磁共振成像（MRI）正在探索， NIH新生儿研究网络（NRN）作为一种潜在的生物标志物，以量化神经基质和预测2- 年不良后果。然而，专家解释和评分系统，目前的规范，只考虑一个 选定的一组脑区域（例如，丘脑、基底神经节、内囊和其他通常沿着 皮质脊髓束），并且仅考虑MRI指标，而不是最佳地将MRI与临床变量组合。的 最近兴起的新的病变-症状映射（LSM）技术使我们能够进行第一项研究，以量化 体素，区域和整个大脑的纤维神经基质，在一个更全面， 多变量方式比NRN评分系统。我们使用第一代LSM的初步结果 （voxel-wise LSM，或V-LSM）已经导致了新的发现，新的区域和左半球的优势，在大脑中。 神经基质，以前没有考虑在NRN评分系统。在R21中，我们的Aim 1将 进一步使用第二代LSM（即，多变量LSM，或M-LSM）和第三代LSM（即， 连接组LSM或C-LSM），以更全面地探索2年不良HIE的神经基质 结果。我们的目标2将提出一种新的机器学习和深度学习算法，称为预测LSM 或P-LSM，比较和联合收割机的三代LSM的结果，并应用这种新的算法 为了预测HIE患者的个体结果，不添加和添加临床变量（目的2a和2b， 分别）。我们的总体假设是，定量和彻底表征的神经基板， 我们即将开发的机器学习和深度学习框架可以预测2年的不良HIE结果 比目前的NRN分数更准确。

Abstract Neonatal brain injury caused by hypoxic ischemic encephalopathy (HIE) affects 1-5/1000 term-born infants. HIE often causes adverse outcomes, defined as death or cognitive Bayley Scales of Infant Development (cBSID)<85 by 2 years. Current therapeutic trials are testing whether late, deeper, and/or longer versions of the standard therapeutic hypothermia could further reduce the adverse 2-year outcome. However, therapeutic innovation is slow and inconclusive, because the exact neuroanatomic substrate (i.e., the anatomy that underlies outcomes) is unknown and there is a lack of a reliable tool to predict 2-year adverse outcomes after HIE. Understanding the neural substrate and hence prediction of the adverse 2-year outcome can: (a) before therapy in the neonatal stage, identify those patients at high risk to develop adverse 2-year outcomes, so our therapeutic trials can focus on those high-risk patients and avoid unnecessary therapies to low-risk patients; and (b) right after therapy, to evaluate therapeutic effects early (in the neonatal stage), avoiding the wait until 2 years to observe the outcome and therefore expediting therapeutic innovations. Brain magnetic resonance imaging (MRI) is being explored by NIH Neonatal Research Network (NRN) as a potential biomarker to quantify neural substrate and predict the 2- year adverse outcome. However, expert interpretation and scoring systems, the current norm, only consider a selected set of brain regions (e.g., thalamus, basal ganglia, internal capsules, and others usually along the corticospinal tract), and only consider MRI metrics, not optimally combining MRI with clinical variables. The recent rise of novel lesion-symptom mapping (LSM) technologies allows us to conduct the first study to quantify voxel-, region-, and fiber-wise neural substrate throughout the brain, in a much more comprehensive and multivariate manner than the NRN scoring systems. Our preliminary results using the first-generation of LSM (voxel-wise LSM, or V-LSM) have led to novel findings of new regions and left hemisphere dominance in the neural substrate that were previously not considered in the NRN scoring system. In this R21, our Aim 1 will further use the second-generation LSM (i.e., multivariate LSM, or M-LSM) and the third-generation LSM (i.e., connectome LSM, or C-LSM) to more comprehensively explore neural substrate underlying 2-year adverse HIE outcome. Our Aim 2 will propose a novel machine learning and deep learning algorithm, termed predictive LSM or P-LSM, to compare and combine the findings from three generations of LSM, and apply this novel algorithm to predict individual outcomes in HIE patients, without and with adding clinical variables (Aims 2a and 2b, respectively). Our overall hypothesis is that quantitative and thorough characterization of the neural substrate in our to-be-developed machine learning and deep learning framework can predict 2-year adverse HIE outcomes more accurately than the current NRN scores.