Alzheimer’s disease detection with Optimal EEG channel selection using Wavelet Transform

Abstract

Alzheimer’s disease (AD) is a neuro degenerative disorder having higher fatality in the elderly due to delayed treatment caused by a lower detection rate. State-of-the-art multi- channel Electroencephalogram (EEG) techniques have been re- ported to assist clinical practitioners in early AD detection. However, a large number of EEG channels comprises several redundant channels resulting in higher computational complexity. Paper presents a channel ranking using sub-band-based energy to entropy ratio for automatic detection of AD. Wavelet packet analysis is used to calculate the ratio of energy to entropy for each wavelet sub-band of the EEG signal. The signal channels are ranked based on calculated sub-band ratio values such that the most important channel has the highest ratio. The feature vector for a channel comprises the mean, standard deviation, kurtosis, minimum value, maximum value, and energy of each wavelet packet sub-band. The optimal number of channels is selected using proposed rank-based sequential backward feature elimination. Six different classifiers, namely, support vector machine (SVM), multi-layer perceptron neural network, k-nearest neighbors, random forest, Naive Bayes, and AdaBoost are used in a 10-fold cross-validation framework. Evaluation is performed using 16 channel Alzheimer’s Patients’ Relatives Association of Valladolid dataset. The experimental results showed the highest accuracy of 97.50%, with 97.08% sensitivity and 97.45% specificity for six channels (T4, P3, P4, O1, O2, and Cz) and SVM classifier.