Effect of biased feedback on motor imagery learning in BCI-teleoperation system

Abstract

Feedback design is an important issue in motor imagery BCI systems. Regardless, to date it has not been reported how feedback presentation can optimize co-adaptation between a human brain and such systems. This paper assesses the effect of realistic visual feedback on users' BCI performance and motor imagery skills. We previously developed a tele-operation system for a pair of humanlike robotic hands and showed that BCI control of such hands along with first-person perspective visual feedback of movements can arouse a sense of embodiment in the operators. In the first stage of this study, we found that the intensity of this ownership illusion was associated with feedback presentation and subjects' performance during BCI motion control. In the second stage, we probed the effect of positive and negative feedback bias on subjects' BCI performance and motor imagery skills. Although the subject specific classifier, which was set up at the beginning of experiment, detected no significant change in the subjects' online performance, evaluation of brain activity patterns revealed that subjects' self-regulation of motor imagery features improved due to a positive bias of feedback and a possible occurrence of ownership illusion. Our findings suggest that in general training protocols for BCIs, manipulation of feedback can play an important role in the optimization of subjects' motor imagery skills.

Keywords: BCI-teleoperation; body ownership illusion; feedback effect; motor imagery learning; training.

Figure 1

Experiment setup. (A) EEG electrodes installed on a subject's sensorimotor cortex recorded brain activities during a motor imagery task. Subjects watched first-person images from a robot's perspective through a head mounted display. A lighting ball in front of each of the robot's hands gave motor imagery cues and subjects imagined grasping their corresponding hands. The classifier detected two classes of results (right or left) and motion command was sent to the robot's hand. (B) Subjects repeated the experiment under three different conditions: Still, where the robot's hand didn't move at all. Match in which the robot's hand only moved in successful trials, and Raw where the robot's hand also performed failed trials using the wrong hand. (C) Experimental procedure consisted of a non-feedback session for classifier setup, training sessions and three operational sessions.

Figure 2

Results for experiment 1 (A) Mean values and standard deviations for Q1 in each session, with Match significantly higher than Still and Raw (B) Mean values and standard deviations for Q2 in each session, with Match and Raw significantly higher than Still (C) Mean values and standard deviations of SCR values for 35 subjects, with Match associated with significantly higher responses to injection than Still (D) Subjects' performances vs. their scores of illusion in the Match condition. For those with the same performance and score, the score has been slightly modified to a non-integer neighbor value to avoid the overlap of the markers. A significantly positive correlation was found between BCI-performance and intensity of illusion.

Figure 3

Model diagram for effect of feedback design. Feedback bias can affect the interaction between BCI-performance and BOT illusion.

Figure 4

Results for experiment 2 (A) Mean value of subjects' performances in the second half of each session is demonstrated. No significant difference was found. (B) Mean value of the ratio J₂/J₁, an identifier of motor imagery quality, showed significantly higher values in the Fake-P and Match conditions compared to Raw.