Error potential detection during continuous movement of an artificial arm controlled by brain–computer interface

Abstract

Patients who benefit from Brain-Computer Interfaces (BCIs) may have difficulties to generate more than one distinct brain pattern which can be used to control applications. Other BCI issues are low performance, accuracy, and, depending on the type of BCI, a long preparation and/or training time. This study aims to show possible solutions. First, we used time-coded motor imagery (MI) with only one pattern. Second, we reduced the training time by recording only 20 trials of active MI to set up a BCI classifier. Third, we investigated a way to record error potentials (ErrPs) during continuous feedback. Ten subjects controlled an artificial arm by performing MI over target time periods between 1 and 4 s. The subsequent movement of this arm served as continuous feedback. Discrete events, which are required to elicit ErrPs, were added by mounting blinking LEDs on top of the continuously moving arm to indicate the future movements. Time epochs after these events were used to evaluate ErrPs offline. The achieved error rate for the arm movement was on average 26.9%. Obtained ErrPs looked similar to results from the previous studies dealing with error detection and the detection rate was above chance level which is a positive outcome and encourages further investigation.