基于多模态MRI与机器学习的终末期肾病脑铁过载性认知损害神经网络机制与早期预测模型的研究

End-stage renal disease patients with hemodialysis suffered brain iron overload-induced cognitive impairment due to the regular use of intravenous iron agent, which affected the understanding and decision-making ability, medication compliance of patients. To date, the mechanism of cognitive impairment was not clear, which caused the lack of clinical diagnostic thresholds, can not make the diagnosis early and delay the disease, increased the incidence of disability and mortality. The previous studies found that the iron overload of brain nuclei was correlated with cognitive impairment; brain iron overload, cognitive impairment were both correlated with brain atrophy, brain blood flow and oxygen metabolism, etc. So we speculated that excess iron deliberately deposits in the cognitive-related regions, which influences its functional connective regions and neural pathways between brain regions, causes the changes of neural pathways and brain structures, metabolism and functions, finally causes the cognitive impairment. We aim to use multi-modality MRI to confirm the cerebral iron overload regions, which was correlated with cognitive impairment; to find the functional cerebral regions connected with the cerebral iron overload regions using functional connective, whole brain network analysis; finally to analyze the changes in neural pathways (network, nodes, and connection paths) between regions and the changes in cerebral structures, metabolism, and functional biomarkers in cerebral iron overload regions and functional connective regions. This study can help us to clarify the neural network mechanism of cerebral iron overload-induced cognitive impairment; to find the special imaging biomarker for early preventation of cognitive impairment and clinical diagnostic thresholds; and to build early predictive model based on machine learning technique. It is of great significance in guiding the clinical treatment of intravenous iron supplementation, preventing and delaying cognitive impairment caused by iron overload.

终末期肾病血透患者常规静脉补铁引起脑铁过载性认知损害，影响患者理解与决策力、药物治疗依从性。目前其发生机制不明确，导致缺乏临床诊断阈值，不能早期诊断而延误病情，增加致残率与死亡率。课题组前期发现血透患者脑核团铁过载与认知损害有关；且脑铁过载、认知损害与脑体积、血流及氧代谢有关。因此提出过量铁特异性沉积于认知相关脑区，影响其功能连接脑区与脑区间神经通路，引起神经通路及各脑区结构、代谢与功能改变，导致认知损害的机制假说。本课题采用多模态MRI明确认知损害相关铁过载脑区；采用功能连接、全脑网络分析获得铁过载脑区的功能连接脑区；分析脑区间神经通路（网络、节点与连接路径）及各脑区结构、代谢与功能影像指标的变化，阐明铁过载性认知损害的神经网络机制，探讨早期预防认知损害的特异性影像指标与临床诊断阈值，并基于机器学习技术构建早期预测模型。研究对指导临床静脉补铁治疗、预防和延缓铁过载性认知损害具有重要意义。

1.项目背景.终末期肾病血透患者常规静脉补铁引起脑铁过载性认知损害，影响患者理解与决策力、药物治疗依从性。目前其发生机制不明确，导致缺乏临床诊断阈值，不能早期诊断而延误病情，增加致残率与死亡率。..2.研究目的.明确脑铁过载对脑结构、血流等变化的影响，阐明铁过载性认知损害的神经网络机制；探讨早期预防认知损害的特异性影像指标与临床诊断阈值；基于机器学习技术实现脑内核团的自动化分割及铁含量、体积的评估。..3.主要研究内容.（1）通过横向获得ESRD血透患者认知损害的特异性影像学指标，评价脑铁过载性认知损害的诊断阈值；（2）基于机器学习技术实现脑内核团的自动化分割与铁含量定量评估，深入挖掘多模态MRI数据中的深层信息，提取、筛选关键影像学特征，结合临床特征构建认知损害的早期预测模型；（3）探讨脑内核团铁过载与相关区域其他影像指标（脑血流、体积等）的关系，探索脑铁过载导致认知损害的可能神经网络络机制。...4.重要结果及关键数据.（1）横向与纵向随访观察ESRD血透患者部分脑区脑血流增加，部分增加的脑血流与认知功能损害有关。贫血是脑血流增加的主要危险因素，调整患者的贫血状态后，相关性依然存在；（2）横向与纵向随访观察ESRD血透患者部分脑区体积萎缩，部分萎缩的脑体积也与认知功能损害有关；（3）中介分析揭示贫血引起ESRD血透患者脑血流增加，增加的血流导致相关脑区的铁过载，进一步引起认知功能损害。增加的铁沉积也导致相关脑区脑萎缩，从而导致认知损害；（4）基于人工智能技术实现颅内核团的自动化分割与铁含量、体积的定量评估。..5.科学意义.通过深度学习的方法对影像学数据进行自动分割，相关研究成果可进行推广，以期为临床医生提供铁过载的可靠自动化评估方法。总体的研究对于临床医生进一步防止终末期肾病患者铁过载引起的神经毒性作用，改善患者智能和工作生活能力、降低ESRD透析病人因脑铁沉积引发认知功能损害有非常重要的意义。