Classification of imaginary movements using the magnitude-squared coherence feature extractor

This study investigates the use of the Magnitude - Squared Coherence (MSC) to extract features from three events: spontaneous electroencephalogram (EEG), EEG-based motor task, and EEG-based motor imagination. We extracted such characteristics by using the Delta band (0.1-2 Hz), Alpha band (8-13 Hz) and Beta band (14-30 Hz). Tasks were classified by using Hidden Markov Models (HMM) and Multilayer Perceptron (MLP). From three healthy subjects, we recorded EEG with electrodes placed according to the international 10-20 and 10-10 systems. HMM observations and networks´ inputs were obtained by the MSC calculated with 12 trials and the frequency range with higher MSC was adopted as a feature for classifiers. The hit rate in classification with HMM was 80%, 73% and 73% for subjects #1, #2 and #3, respectively. When we used MLP, the rates were 75%, 75% and 79%. These findings have shown we can extract features from brain activities related to different events by using coherence and that HMM and MLP are useful in the classification of imaginary movements.