In the scope of our master thesis, we developed and evaluated a procedure for detection and classification of event-related desynchronisation intervals of the electroencephalogram occurring during imagined motor activities of the left and right hand of a subject and also between the actual motor activities of the left and right hand. The developed procedure allows, for example, to move the cursor on the screen based on the imagined motor activities. For the development, we used electroencephalogram recordings of the EEGMMI DS database, freely available on the Physionet repository, which contains manually annotated intervals between which subjects imagined, or performed, different motor activities. Critical evaluation of the recordings showed that some of the recordings do not show the expected decrease in the signal amplitude between desynchronisation intervals, and therefore in order to obtain objective and realistic results, these recordings were treated separately. To translate the original signals into the component space, we used two methods, the common spatial patterns method and the large Laplacian mask. The classifiers chosen were the LDA and QDA classifiers. The highest achieved classification accuracy for the motor imagination intervals of activity for the common spatial patterns method and the QDA classifier was 93.30 %, while using the large Laplacian mask was 87.31 %.
|
