A Single-Channel EEG-Based Discrimination of Major Depressive Disorder with Wavelets Features and Machine Learning

Electroencephalography is largely used as a technique for neuroscience studies and applications. However, using all of its electrode set is not required for every EEG-related problem, especially when reducing the number of channels will lead to an optimized experience by minimizing both time and computational resources. In this study, we tested the possibility of using single-channel EEG signals for prescreening major depressive disorder within a group of 34 depressive patients and 30 age-matched healthy controls. Discrete wavelet transform (DWT) with 4 levels of decomposition was applied on 364 EEG segments in order to extract relevant brain rhythms (Delta, Theta, Alpha, Beta, Gamma). A 364 by 35 matrix of wavelet features was generated and then ranked using the ReliefF algorithm. Features from Delta, theta, and Alpha bands proved to be the most relevant and an accuracy of 97.28 % had been reached by the Decision Tree classifier in conjunction with 10-fold Cross Validation (10-fold CV).