Detection of Alzheimer’s Disease Using Extreme Band Powers in EEG Data

Alzheimer’s disease (AD) is one of the leading category among the neurodegenerative diseases and characterized by cognitive deficits, deterioration in memory, behavioral changes etc. In this paper, we present a methodology for improvement in classification of AD and normal subjects using scalp Electroencephalogram (EEG) signals, captured during resting with eyes closed. The extreme values of the distribution of window-wise EEG band powers are used as novel features. Using standard correlation based methods, the lower extreme of Alpha band power, upper extremes of Delta band, Theta band and Hjorth Complexity are selected as the final features. Among 19 EEG electrodes / channels, finally C3, Cz, Pz, P4 are considered by removing the dependent channels for selected features. The proposed approach along with Random Forest based classification gives an F1-score of 81.69%. This gives an improvement of approximately 8%-12% as against the baseline methods with standard relative band power features and classifiers like lightGBM (76.12%), SVM (73.74%), KNN (72.81%), MLP (74.82%) and Random Forest (75.31%).