Brain network analysis of working memory in schizophrenia based on multi graph attention network

Ping Lin,Geng Zhu,Xinyi Xu,Zhen Wang,Xiaoou Li,Bin Li
DOI: https://doi.org/10.1016/j.brainres.2024.148816
IF: 3.61
2024-02-22
Brain Research
Abstract:The cognitive impairment in schizophrenia (SZ) is characterized by significant deficits in working memory task. In order to explore the brain changes of SZ during a working memory task, we performed time-domain and time–frequency analysis of event related potentials (ERP) of SZ during a 0-back task. The P3 wave amplitude was found to be significantly lower in SZ patients than in healthy controls (HC) (p < 0.05). The power in the θ and α bands was significantly enhanced in the SZ group 200 ms after stimulation, while the θ band was significantly enhanced and the β band was weakened in the HC group. Furthermore, phase lag index (PLI) based brain functional connectivity maps showed differences in the connections between parietal and frontotemporal lobes between SZ and HC (p < 0.05). Due to the natural similarity between brain networks and graph data, and the fact that graph attention network can aggregate the features of adjacent nodes, it has more advantages in learning the features of brain regions. We propose a multi graph attention network model combined with adaptive initial residual (AIR) for SZ classification, which achieves an accuracy of 90.90 % and 78.57 % on an open dataset (Zenodo) and our 0-back dataset, respectively. Overall, the proposed methodology offers promising potential for understanding the brain functional connections of schizophrenia.
neurosciences
What problem does this paper attempt to address?