Comparative analysis of spectral approaches to feature extraction for EEG-based motor imagery classification

The quantification of the spectral content of electroencephalogram (EEG) recordings has a substantial role in clinical and scientific applications. It is of particular relevance in the analysis of event-related brain oscillatory responses. This work is focused on the identification and quantification of relevant frequency patterns in motor imagery (MI) related EEGs utilized for brain-computer interface (BCI) purposes. The main objective of the paper is to perform comparative analysis of different approaches to spectral signal representation such as power spectral density (PSD) techniques, atomic decompositions, time-frequency (t-f) energy distributions, continuous and discrete wavelet approaches, from which band power features can be extracted and used in the framework of MI classification. The emphasis is on identifying discriminative properties of the feature sets representing EEG trials recorded during imagination of either left- or right-hand movement. Feature separability is quantified in the offline study using the classification accuracy (CA) rate obtained with linear and nonlinear classifiers. PSD approaches demonstrate the most consistent robustness and effectiveness in extracting the distinctive spectral patterns for accurately discriminating between left and right MI induced EEGs. This observation is based on an analysis of data recorded from eleven subjects over two sessions of BCI experiments. In addition, generalization capabilities of the classifiers reflected in their intersession performance are discussed in the paper.