基于图聚类的全脑功能分割研究

The human brain functional atlas is the basis of brain network analysis and functional connectivity studies. It is widely applied in studies of brain cognition, development, disease, etc. The brain functional atlas can be generated by functional connectivity-based parcellation. However, there is still no agreement about how the brain should be parcellated. Based on our previous work, we introduce the frontiers of graph-based clustering theories into this project and conduct whole-brain functional parcellation with these theories systematically and thoroughly. The key problem of this project is how to make the most of the properties of resting-state fMRI data in order to construct appropriate optimization problems for graph training and whole-brain parcellation. The main contents of this project include training a graph based on an existing graph, training a graph in the original feature space and in the intrinsic space simultaneously, and training a graph by low-rank representation. The project will enrich the theoretical system of graph-based clustering in the field of whole-brain parcellation, construct novel parcellation algorithms, generate more accurate brain atlases, and provide insights and tools for the theories and applications of brain science.

人脑功能图谱是进行人脑网络分析和功能连接研究的基础，在研究大脑认知、发展、疾病等方面有广泛应用。人脑功能图谱可通过对全脑静息态fMRI数据进行分割实现，然而大脑究竟该如何分割才合理一直没有定论。本项目在已有工作的基础上，引入图聚类的前沿理论和思想，系统深入地开展基于图聚类的全脑功能分割研究。项目的关键问题在于如何充分利用静息态fMRI数据的特征合理地构造优化问题以完成图的训练并实现全脑分割。项目的主要内容包括：基于已有的图训练新的图；同时在原始特征空间和本征子空间训练图；引入低秩表示训练图。本项目将丰富并完善图聚类算法在全脑分割领域中的理论体系，构建新的全脑分割算法并生成更准确的大脑图谱，为脑科学的理论发展和应用推广提供新思路和新工具。

图论是人脑网络分析和功能连接研究的理论基础。本项目基于图的思想，围绕全脑fMRI数据分析和人脑功能连接网络模式，开展了以下研究工作：基于图惩罚项的稀疏逻辑回归用于全脑fMRI数据分析；直接双侧二维主成分分析用于蕴含结构信息的数据特征提取；基于图惩罚项的稀疏平滑主成分分析用于全脑fMRI数据特征提取；应用图聚类理论探讨数学天分大脑功能连接模式特性。本项目的实施针对蕴含结构信息的数据提出了多种具有创新性的特征提取和分析方法，能更加充分地利用全脑fMRI数据的空间结构信息，并且能增强模型在神经生理学意义上的可解释性。此外，本项目应用图聚类理论成功分析出数学天分大脑的功能连接网络模式特性，为青少年大脑资源开发提供了理论依据。