基于多中心大数据构建MRI筛查新生儿脑白质损伤的预测模型研究

The incidence of neonatal white matter damage (WMI) is as high as 20%. WMI is closely related to adverse neurodevelopment outcome, such as motor, cognitive and attention damage, even the cerebral palsy. Comparing with ultrasound, MRI can accurately evaluate the degree of the WMI and provide rich prognostic information. However, MRI is not a universal screening method for WMI due to its expensive fee and longer scanning time. Previous studies demonstrated that WMI appearance may change or disappear with time, which is not conducive to the detection and accurate diagnosis of WMI. Therefore, how to screen neonates who need do MRI examination and detect and accurately evaluate WMI in time are of important clinical significance for early intervention and improvement of nervous system developmental outcome. In view of the above problems, this project will explore the clinical risk factors associated with WMI based on multi-center clinical data, and develop a model to predict WMI at birth to help screen neonates who need do MRI examination using Logistic regression analysis and machine learning algorithms. This will achieve timely detection and evaluation of WMI damage and individualized assessment. Comprehensive assessment, improved prognosis will reduce the burden on families and society.

目前新生儿脑白质损伤（WMI）发生率高达20%。且WMI与神经系统发育结局密切相关，重者可发展为脑瘫。较于超声，MRI检查可对WMI做出精确诊断、提供丰富预后信息。但由于MRI检查费用高，扫描时间长，并非普适性筛查WMI手段。且WMI病灶会随着时间改变或消失，不利于临床对WMI的诊断及评估。因此，如何通过临床因素预测发生WMI的风险，筛查须行MRI检查的新生儿，既可辅助临床医生及时做出决策，实现对WMI病灶及时检出、损伤范围精确评估；也是临床医生对WMI患儿全面评估、预测及提高神经系统发育结局的关键所在。针对上述问题，本项目将基于多中心大数据通过Logistic回归分析及多种机器学习算法构建MRI筛查新生儿脑白质损伤的普适性预测模型。实现出生时个体化评估，辅助临床医生对患儿的全面评估，提高预后；同时使医疗资源合理有效分配，减轻家庭及社会负担。

新生儿脑白质损伤（white matter injury，WMI）为目前新生儿最常见脑损伤类型，与脑瘫、 认知及行为障碍等不良结局密切相关。同时，损伤病灶离散、多变，易随时间消退或呈“假正常化”，导致临床漏诊及低估。同时，新生儿WMI病因学尚不清楚，缺乏大样本、多中心数据探索其高危因素， 制约临床对WMI 发病机理的理解和一级预防的开展。因此，WMI临床预防和早期筛查对新生儿、家庭及社会具有重要临床意。基于上述问题，本研究开展了以下几方面的工作：（1）通过分析本课题组多中心新生儿临床-MRI数据，探索了WMI患儿临床高危因素，发现了9项围产期临床高危因素，以产中-产后特征为主，为疾病的一级预防奠定了基础。（2）通过多中心数据探索了新生儿不同程度WMI临床危险因素及其差异，发现胎龄、出生体重、低血糖、先天性心脏病、电解质紊乱、新生儿惊厥、生产方式及Apgar 1分钟评分与轻–重度WMI发生有关，且不同严重程度尖，危险因素存在差异。（3）完成了基于多因素logistic回归及CHAID决策树算法的新生儿WMI危险分层模型构建，为临床提供了基于临床因素辅助筛查WMI的简便工具。（4）探讨了颅脑超声和MRI检出新生儿WMI的敏感性、特异性及诊断性能的差异，为临床选择适宜神经影像方法探查新生儿WMI提供证据支持。（5）探讨了不同层厚T1WI对新生儿WMI诊断、病灶数检出、严重程度评估的影响，为临床适宜新生儿颅脑MRI扫描方案的制定提供依据。.综上所述，通过本课题研究，基于多中心大样本数据发现了新生儿WMI临床高危因素，并基于高危因素构建了新生儿风险分层评分表，为临床决策神经影像学检查提供辅助支持工具，从而促进疾病早筛、神经影像检查技术及医保资源的合理利用。