| The execution of specific cognitive tasks requires the mutual coordination and interaction of different brain regions.However,the complex differences and dynamic variability of multiple brain regions’ interactions(brain networks)in different cognitive tasks make it difficult for existing cognitive science research methods to intuitively characterize them,which also increases the difficulty of human understanding of brain cognitive mechanisms.The continuous development of deep learning technology and brain network omics provides feasible means for personalized and intelligent excavation of dynamic evolution laws of brain networks in cognitive processes.This article constructs a graph deep learning network model based on EEG brain networks from the perspective of identifying multiple cognitive tasks and analyzing cognitive processes.It achieves accurate identification of different cognitive tasks and proposes a model interpretation and understanding method1.It explores the relationship between EEG-related temporal features,brain network structure,and cognitive tasks,providing a feasible basis for the in-depth understanding of brain cognitive mechanisms and obtaining potential biological markers associated with cognition.It also helps early diagnosis and intervention of cognitive impairment diseasesThis study includes three specific research contents:(1)Aiming at the problems of the meager signal-to-noise ratio of EEG signals,lack of whole-brain perspective under traditional analysis,and complexity of brain cognitive state,this paper proposed a graph deep learning network based on induction-expression thinking design: EEG-brain function network cognition recognition network(EBNCRN).Firstly,the EEG brain function network was constructed using the amplitude-frequency consistency analysis method,and the sample size was expanded by the data augmentation strategy.Then EBNCRN was used to realize classification recognition of brain networks under different cognitive tasks.A series of model experiments verified the validity,reliability,and crosssubject ability of model classification.This study improves the accuracy and robustness of cognitive task recognition and provides effective models and data for subsequent interpretability analysis.(2)In order to explore the key features and structures that graph deep learning models rely on in EEG brain network classification and their representation methods,this study proposes an interpretation method based on temporal features and topological structure features of the EEG brain network and the corresponding model interpretation understanding method.Interpretation methods include gradient-type class activation mapping method,integral gradient interpretation method,and parameterized model interpreter,which optimize the application of the above methods in EEG brain network.Model interpretation understanding methods include understanding methods for signal temporal features and key graph structure features.This study provides effective research tools and analysis guidance for subsequent experimental results analysis.(3)In order to explore the relationship between key brain features based on EBNCRN in the cognitive task process and subjects’ cognitive tasks,this study proposes a scientific research analysis framework based on the deep learning model of the EEG brain network.By applying the EBNCRN model and model interpretation understanding method,signal features and key graph structures that affect model output under different cognitive tasks and different subjects are obtained.The results obtained in this paper are compared with existing EEG cognitive research conclusions to reveal functional connection relationship between different brain regions under different cognitive tasks and their temporal characteristics.This study verifies that the proposed method has strong effectiveness and reliability in EEG cognitive research,provides a new research tool and research process for cognitive neuroscience from a data-driven perspective,explores dynamic changes in brain functional networks and mechanisms of cognitive processes from a data-driven perspective,promotes cross-disciplinary integration between cognitive neuroscience and artificial intelligence. |
