Classification Accuracy of an Imagined-Movements Mental Task Set for Brain-Machine Interface

We study the classification accuracy of a neural network (NN) classifier for a Brain-Machine Interface (BMI) system that uses an imagined-movements mental task set and compare the performance of the classifier when the user is subjected to a non-moving (static) reference frame (SRF) and to a moving (dynamic) reference frame (DRF) while performing the mental tasks. We use the band powers and power differences of electroencephalogram (EEG) signals recorded from 8 surface electrodes. Results show that tasks involving imagined movements are the most immune to the SRF-DRF switch but may not be the most appropriate BMI protocol for BMI classification.