A new method for classification of subjects with major cognitive disorder, Alzheimer type, based on electroencephalographic biomarkers

Abstract

1 Abstract: The diagnosis of Alzheimer's disease (AD), Mild Cognitive Impairment (MCI), and healthy subjects (Healthy) in clinical practice currently lacks an automated tool. Discriminating between Healthy and MCI subjects requires experienced neuropsychologists and have a sensitivity of 80%. In this paper, a new classification method to separate the three groups using EEG resting-state is evaluated. The method integrates features as wavelet entropy's Pearson correlation coefficient, theta relative power and P300 latency and amplitude. The use of Pearson correlation to analyze wavelet entropy is a new approach since entropy analysis is traditionally based on the average of entropy values of all epochs. The biomarkers integration is also novel since, to our knowledge, there are no proposed methods combining connectivity, spectral characteristics, complexity and P300 for this task. The results obtained show that it is possible to automatically classify a subject with an overall accuracy of 94.44%, close to the best result found in the literature, 97.90%. The classification method here proposed could be the base for implementing a quantitative diagnosis-support tool. The selected features also may be used as endogenous indicators of progression from MCI to AD.

Keywords

• Cognitive disorder
• EEG complexity
• Entropy
• Pearson correlation coefficient
• P300
• Theta relative power