Kurashige Hiroki, Kaneko Jun, Yamashita Yuichi, Osu Rieko, Otaka Yohei, Hanakawa Takashi, Honda Manabu, Kawabata Hideaki
Institute of Innovative Science and Technology, Tokai University, Tokyo, Japan.
National Institute of Neuroscience, National Center of Neurology and Psychiatry, Tokyo, Japan.
Front Hum Neurosci. 2020 Jan 10;13:457. doi: 10.3389/fnhum.2019.00457. eCollection 2019.
To characterize each cognitive function and to understand the brain as an aggregate of those functions, it is vital to relate dozens of these functions to each other. Knowledge about the relationships among cognitive functions is informative not only for basic neuroscientific research but also for clinical applications and developments of brain-inspired artificial intelligence. In the present study, we propose an exhaustive data mining approach to reveal relationships among cognitive functions based on functional brain mapping and network analysis. We began our analysis with 109 pseudo-activation maps (cognitive function maps; CFM) that were reconstructed from a functional magnetic resonance imaging meta-analysis database, each of which corresponds to one of 109 cognitive functions such as 'emotion,' 'attention,' 'episodic memory,' etc. Based on the resting-state functional connectivity between the CFMs, we mapped the cognitive functions onto a two-dimensional space where the relevant functions were located close to each other, which provided a rough picture of the brain as an aggregate of cognitive functions. Then, we conducted so-called conceptual analysis of cognitive functions using clustering of voxels in each CFM connected to the other 108 CFMs with various strengths. As a result, a CFM for each cognitive function was subdivided into several parts, each of which is strongly associated with some CFMs for a subset of the other cognitive functions, which brought in sub-concepts (i.e., sub-functions) of the cognitive function. Moreover, we conducted network analysis for the network whose nodes were parcels derived from whole-brain parcellation based on the whole-brain voxel-to-CFM resting-state functional connectivities. Since each parcel is characterized by associations with the 109 cognitive functions, network analyses using them are expected to inform about relationships between cognitive and network characteristics. Indeed, we found that informational diversities of interaction between parcels and densities of local connectivity were dependent on the kinds of associated functions. In addition, we identified the homogeneous and inhomogeneous network communities about the associated functions. Altogether, we suggested the effectiveness of our approach in which we fused the large-scale meta-analysis of functional brain mapping with the methods of network neuroscience to investigate the relationships among cognitive functions.
为了描述每种认知功能，并将大脑理解为这些功能的集合，将几十种这样的功能相互联系起来至关重要。关于认知功能之间关系的知识不仅对基础神经科学研究有参考价值，对临床应用以及受脑启发的人工智能的发展也有参考价值。在本研究中，我们提出一种详尽的数据挖掘方法，基于功能性脑图谱和网络分析来揭示认知功能之间的关系。我们从109个伪激活图谱（认知功能图谱；CFM）开始分析，这些图谱是从一个功能磁共振成像元分析数据库重建而来的，每个图谱对应109种认知功能中的一种，如“情绪”“注意力”“情景记忆”等。基于CFM之间的静息态功能连接性，我们将认知功能映射到一个二维空间，其中相关功能彼此靠近，这提供了一幅大脑作为认知功能集合的大致图景。然后，我们使用每个CFM中与其他108个CFM具有不同强度连接的体素聚类，对认知功能进行所谓的概念分析。结果，每个认知功能的CFM被细分为几个部分，每个部分与其他认知功能子集中的一些CFM密切相关，这引出了该认知功能的子概念（即子功能）。此外，我们对一个网络进行了网络分析，该网络的节点是基于全脑体素到CFM的静息态功能连接性从全脑分割得到的脑区。由于每个脑区的特征是与109种认知功能相关联，使用它们进行网络分析有望揭示认知特征与网络特征之间的关系。事实上，我们发现脑区之间相互作用的信息多样性和局部连接密度取决于相关功能的种类。此外，我们确定了关于相关功能的同质性和异质性网络群落。总之，我们提出了我们方法的有效性，即我们将功能性脑图谱的大规模元分析与网络神经科学方法相结合，以研究认知功能之间的关系。
