An empirical mode decomposition (EMD)-based scheme for alcoholism identification

Abstract

Alcohol addiction can severely affect the brain structure and function. The brain state is usually determined with the help of the ElectroEncephaloGram (EEG) signals, which consist of different neuronal activities that are recorded from the different regions of the brain. These signals have varying characteristics and low magnitudes (µV). All these factors can hinder the human interpretation of these signals and make it time-consuming. Furthermore, these different EEG signals are also prone to several inter/intra variability errors. On the other hand, they exhibit some non-stationary and non-linear characteristics. As a result, a lot of effort is needed for deciphering the signals with the use of linear frequency and time-domain processes. The nonlinear parameters, in addition to the time-frequency scale domain processes, help in detecting the signal variations. Here, the researchers have proposed a novel technique for identifying the effect of alcoholism with the help of an Empirical Mode Decomposition (EMD) method that extracts the relevant features from the EEG signals. These features were derived from the EEG band spectra, which was directly affected by alcohol. An Independent Component Analysis (ICA) was applied for the feature reduction, which was followed by the weighted k-Nearest-Neighbour (wk-NN) technique for classifying the normal and the alcoholic EEG signals. For ensuring the reliability of the classification system, the researchers adopted a 10-fold cross-validation approach. This system generated results with a 98.91% mean accuracy, 99.02% mean sensitivity and a 99.24% specificity, indicating the robustness and the dependability of the system, which makes it effective under different environmental conditions.