Extracting Spatial Spectral Patterns from EEG Signals for Diagnosis of Mild Cognitive Impairment

Mild cognitive impairment )MCI( is an early stage of Alzheimer's disease)AD). Early diagnosis of this disease increases the likelihood of the treatment and prevents its conversion to dementia. The purpose of this study is the discrimination and the classification of two groups of healthy and MCI subjects by an advanced method of extracting spatial spectral features from electroencephalographic (EEG) signals. Filter bank common spatial pattern (FBCSP) has been recently used for classifying motor imagery EEG data in brain-computer interface researches. Here, we propose using FBCSP for classifying EEG data from healthy and MCI subjects. The proposed method was tested and compared to the popular method of frequency band-power and to the classic common spatial pattern (CSP) using a dataset of 9 MCI patients and 12 healthy subjects. A leave-one-out cross validation, using FBCSP resulted in a classification accuracy of 100% and outperformed both the frequency band-power and classic CSP methods. These results reveal the important role of using the learned spatial combinations of EEG signals in different frequency bands as effective features for discrimination of MCI and normal subjects.