Differentiating amnestic from non-amnestic mild cognitive impairment subtypes using graph theoretical measures of electroencephalography

The purpose of this study was to explore different patterns of functional networks between amnestic mild cognitive impairment (aMCI) and non-aMCI (naMCI) using electroencephalography (EEG) graph theoretical analysis. The data of 197 drug-naïve individuals who complained cognitive impairment were reviewed. Resting-state EEG data was acquired. Graph analyses were performed and compared between aMCI and naMCI, as well as between early and late aMCI. Correlation analyses were conducted between the graph measures and neuropsychological test results. Machine learning algorithms were applied to determine whether the EEG graph measures could be used to distinguish aMCI from naMCI. Compared to naMCI, aMCI showed higher modularity in the beta band and lower radius in the gamma band. Modularity was negatively correlated with scores on the semantic fluency test, and the radius in the gamma band was positively correlated with visual memory, phonemic, and semantic fluency tests. The naïve Bayes algorithm classified aMCI and naMCI with 89% accuracy. Late aMCI showed inefficient and segregated network properties compared to early aMCI. Graph measures could differentiate aMCI from naMCI, suggesting that these measures might be considered as predictive markers for progression to Alzheimer’s dementia in patients with MCI.