Multi-Site Clinical Data to Power MRI Biomarker of Neonatal Brain Injury

摘要 该项目旨在发布我们最近收集的新生儿缺氧缺血的现有临床数据 脑病（HIE）。 HIE 影响 1-5/1000 名足月出生的新生儿，是幼儿期的主要原因 死亡率和发病率。新生儿脑部磁共振成像 (MRI) 是常规采集用于临床诊断 照顾 HIE。新生儿脑 MRI 有望揭示不良后果的 3D 神经解剖学机制，以便 我们可以设计专门针对这些机制的新疗法。新生儿脑部 MRI 也带来希望 识别那些有可能在以后的生活中出现不良后果的新生儿，以便早期干预 该计划可以针对那些高危新生儿以获得最大利益。尽管 MRI 在 HIE 的治疗中发挥着至关重要的作用， 当前临床实践中的规范是由神经放射科医生或神经科医生目视读取 MRI。专家解读， 然而，它有很多局限性——主观的、定性的、不足以揭示机制、也不足以 预测结果。随着人工智能 (AI) 的兴起，MRI 的客观定量分析成为可能 在医学和神经影像信息学领域。然而，一个主要的限制是缺乏公开的数据 嗨。我们的项目旨在通过归档和发布我们临床获得的大规模（N = 231）和 HIE 上的多站点（2 家医院）数据。我们的数据部分由 NIH R01 (2012-2017) 资助获得， 基金会（2016-2020）。我们的数据很全面，包括临床数据元素（来自母亲和 新生儿）、新生儿脑 MRI（结构和扩散序列）、病变的专家共识注释 新生儿脑 MRI 区域、NICU 结果（死亡/生存、住院时间）和 2 岁神经认知 结果（正常/不良，发育情况是/否，听力/视觉/运动障碍是/否，以及 脑瘫是/否）。我们的数据也具有代表性，来自不同种族/民族的患者 来自不同 MRI 扫描仪（Siemens 3T 或 GE 1.5T）的具有广泛结果的患者组， 不同的成像方案，以及生命中不同日子的 MRI 扫描。我们还将从现有数据中获取新数据 数据。匿名（去识别化）数据将通过“受控”发布到 NCBI dbGaP 平台 Access”选项，需要 IRB 和数据使用协议 (DUA)。导出的数据将发布到 dbGaP “开放获取”选项，无需任何批准即可免费下载。两个发布选项都一致 dbGaP 上已发布的其他临床和 MRI 数据。我们希望这是第一个全面的数据 将促进人工智能未来的协作努力，自动识别 MRI 中的 HIE 病变，并准确地识别 结合临床和 MRI 信息预测 HIE 结果。

Abstract This project aims to release our recently gathered existing clinically-acquired data for neonatal hypoxic ischemic encephalopathy (HIE). HIE affects 1-5/1000 term-born neonates and is a major cause of early-childhood mortality and morbidity. Neonatal brain magnetic resonance imaging (MRI) is acquired routinely for the clinical care of HIE. Neonatal brain MRI is expected to reveal 3D neuroanatomic mechanisms of adverse outcomes so that we can design new treatments specifically target those mechanisms. Neonatal brain MRI also carries hope to identify those neonates who are at risk to develop adverse outcomes later in life, so that early intervention program can target those at-risk neonates for maximum benefit. Despite MRI's vital role in caring for HIE, the current norm in clinical practice is to read MRI visually by expert neuroradiologist or neurologist. Expert reads, however, has many limitations – subjective, qualitative, insufficient to reveal mechanisms, and inadequate to predict outcomes. Objective and quantitative analysis of MRI is possible with the rise of artificial intelligence (AI) in medical and neuroimaging informatics. A major limitation, however, is the lack of publicly-available data on HIE. Our project aims to fill this gap, by archiving and releasing our clinically-acquired, large-scale (N=231), and multi-site (2 hospitals) data on HIE. Our data was acquired partly funded by NIH R01 (2012-2017) and foundations (2016-2020). Our data is comprehensive, including clinical data elements (from both mothers and neonates), neonatal brain MRI (structural and diffusion sequences), expert-consensus annotation of lesion regions in neonatal brain MRI, NICU outcome (death/survival, length of stay), and 2-year-old neurocognitive outcomes (normal/adverse, yes/no for development dealy, yes/no for the hearing/visual/motor impairment, and yes/no for cerebral palsy). Our data is also representative, coming from patients with different racial/ethnicity groups, in patients with a wide range of outcomes, from different MRI scanners (Siemens 3T or GE 1.5T), with different imaging protocols, and MRI scanned on different days of life. We will also derive new data from existing data. The anonymized (de-identified) data will be released to the NCBI dbGaP platform with the “Controlled Access” option, requiring IRB and data use agreement (DUA). The derived data will be released to dbGaP with the “Open Access” option, freely downloadable without any approval. Both release options are consistent with other clinical and MRI data that have already been released on dbGaP. We hope this first comprehensive data will boost future collaborative efforts for AI to automatically identify HIE lesions in MRI, and for AI to accurately predict HIE outcome integrating clinical and MRI information.