动态脑功能连接的多模态成像研究

This project aims at investigating chronnectomics of the resting-state human brain and its applications based on fusion of multi-modality neuroimaging techniques including electroencephalograph (EEG), functional magnetic resonance imaging (fMRI) and diffusion tensor imaging (DTI). Firstly, we will develop novel models and estimation methods of the brain's functional connectivity (FC) states, by constructing feasible fusion models for data with different modality. Secondly, we would apply data-driven strategies based on machine learning to explore and characterize low-dimensional temporal and spatial structures embedded in observed data of the FC state-space. Further, we will also perform dynamic parcellation of brain regions and networks, to investigate the principles of dynamic interaction and functional organization across different functional units in the brain at the time-scale of several seconds, which are validated by the task-based experimental data and clinical datasets of some typical brain disorders. Finally, the obtained results are applied to the prediction of individual behavioral ability and assistant clinical classification of patients with brain disorders. In brief, the ultimate objective of this project is to make significant advancement in FC state-space theory and analysis methodology, neurophysiological meanings of FC states, and specific biomarkers derived from dynamic functional connectivity for brain disorders, which consequently further our understanding to principles of dynamic interaction across functional sub-systems in the brain’s cortices, as well as their links to individual cognitive and behavioral abilities.

项目旨在结合脑电(EEG)、功能磁共振(fMRI)、弥散张量成像(DTI)等多模态脑成像技术，开展静息条件下的脑时变连接组学及其应用研究。首先，通过构建合理的多模态融合计算模型，发展高维脑连接状态的建模与估计方法；其次，利用机器学习的数据驱动方法，探索脑状态观测空间的低维时空结构和特性，通过状态依赖的脑区/脑网络动态功能剖分，理解大脑在秒级时间尺度上，各功能单元的动态交互及功能组织规律，并采用任务实验及典型脑疾病的临床数据验证结果的合理性。最后，开展动态功能连接在个体行为能力预测及脑疾病的辅助临床分类的应用研究。项目预期在基于EEG-fMRI融合分析的脑连接状态空间描述与分析方法、脑状态的神经生理意义以及基于动态功能连接的脑疾病特异性指标等方面取得显著进展，促进对脑皮层各功能子系统的动态交互规律，以及它们与个体认知和行为能力之间关系的理解。

项目旨在结合脑电(EEG)和功能磁共振成像(fMRI)等多模态无创脑成像技术，开展静息条件下的脑动态功能连接及其应用研究。完成的研究内容主要包括：首先，我们基于EEG-fMRI多模态数据揭示了全局信号时间波动性的神经生理基础，发现全局信号回归将对脑功能连接状态的估计产生显著影响；其次，我们提出基于动态功能连接特征的脑功能网络及脑区子区剖分方法，得到了海马体更为精细的功能子区剖分结果，并且发现采用该方法获得的全脑功能脑区划分具有更好的脑区内动态一致性；进一步，我们提出面向脑(动态)功能连接模式分类的深度学习模型，设计域适应迁移学习策略解决大数据影像分析多中心导致的数据分布不一致问题，提升了小样本条件下脑功能连接模式分类的准确性；最后，我们将动态脑连接的模型算法等成果应用于临床数据，依次开展针对睡眠剥夺、抑郁症、精神分裂症等脑状态或脑疾病的临床验证，证实脑静息态功能连接的动态特性可以作为一种有效的识别和辅助诊断脑认知损伤的影像生物学标记。项目结果发表于IEEE Trans. on PAMI, NeuroImage, Cerebral Cortex等重要期刊，获得的模型、方法及实验结果，对于进一步研究精神疾病的客观影像学辅助诊断，理解脑认识损伤的神经机制具有科学价值。