(in English) |
Thanks to development of machine learning methods based on large-scale data, it has made huge impacts on not only image and natural language processing, but also processing of neuroimaging data. For example, Human Connectome Project in U.S.A has been collecting more than thousands of brain data with demographic data including scores of cognitive functions and sharing them as open-public data. Therefore, researchers of methodology or statistics have abundantly reported advanced analysis methods. In particular, functional connectivity calculated from brain activity at rest represents not only age and sex, but also cognitive functions and psychiatric disorders. Recently, interactions of several networks (central executive network, default mode network and saliency network) represent individual creative cognition. In general, data of functional connectivity is high dimension, therefore, it often occurs over-fitting problem. To avoid this problem, data of functional connectivity should be assumed sparseness. In this study, I examined connectome-based prediction model which combined feature extractions machine learning with data-driven manner to illustrate brain network which represented the cognitive functions associated with creativity. I conducted several machine learning methods which assumed sparseness for large-scale brain data. In addition, I prepared several parcellation methods to optimize feature extraction methods. I would like to discuss feasibility of prediction model for the cognitive functions based on results of the connectome-based prediction models. |