Motor imagery decoded from electroencephalography signal by means of self-organizing maps

Abstract

Motor imagery is a cognitive process which involves a planning of a movement without performing its execution. This planning can be decoded from the electroencephalography signal and be used for therapeutic purposes. Self-organized maps are artificial neural networks that have a potential to be used as motor imagery classifiers. In this work 11 channels were recorded from 2 healthy subjects and from 2 stroke patients. The data from the channels was pre-processed with a Laplacian type spatial filter applied to channels C3, Cz and C4, then power features where computed with a wavelet transform in the alpha and beta bands. Self-organized maps were trained and tested for each participant with the power features obtained from 3 different conditions. Correct classification accuracies above the chance level were achieved and even reached 80%. The proposed classifiers are promising for motor imagery decoding in the EEG signal, the performance of the proposed classifiers can be increased if a feedback system could be used to train the system’s users.