Alzheimer's disease detection from optimal electroencephalogram channels and tunable Q-wavelet transform

Abstract

Alzheimer's disease (AD) is a non-curable neuro-degenerative disorder that has no cure to date. However, it can be delayed through daily activity assessment using a robust electroencephalogram (EEG) based system at an early stage. A selection technique using a Shannon entropy to signal energy ratio is proposed to select optimal EEG channels for AD detection. A threshold for channel selection is calculated using the best detection accuracy during backward elimination. The selected EEG channels are decomposed using tunable Q-wavelet transform (TQWT) into nine different subbands (SBs). Four features: Katz's fractal dimension, Tsallis entropy, Relyi's entropy, and kurtosis are extracted for each SB. These features are used to train and test support vector machine, k-nearest neighbor, ensemble bagged tree (EBT), decision tree, and neural network for detecting AD patients from normal subjects. 16-channel EEG signals from 12 AD and 11 normal subjects recorded using the 10-20 electrode placement method are used for evaluation. Ten optimized channels are selected, resulting in 32.5% compression. The experimental results of the proposed method showed promising classification accuracy of 96.20% with the seventh SB features and EBT classifier. The significance of these features was inspected by using the Kruskal-Wallis test.