Classification of executed upper limb movements by means of EEG

This work presents the performance of a Linear Discriminant Analysis classifier that used EEG data from 3 different subsets of the signal, which was gathered during the execution of 4 upper limb movements. The mean Power of the signal, segmented in 8 EEG frequency bands, was used as the features for the classifier and the effect of spatial feature selection was also investigated. A non-conventional potential difference based on an 8-electrode clinical transversal setup was used in the acquisition of EEG signal during arm and hand movements, which were segmented in Movement Planning, Movement Execution and Steady Position. The results showed that the Movement Planning subset achieved the best classification accuracy, suggesting that the speed for a BCI can be improved by using pre-movement information. Spatial feature selection showed that non-motor areas should be considered as an information source. Best classification accuracy of right and left limbs was 67.95%, hands versus arms achieved 82.69%, and 49.36% of classification was the best result for the 4-class set up. Results are promising, however further experiments are required to obtain better classification accuracy and to generalize these conclusions.