Asynchronous Steady-State Visual Evoked Potential based Brain-Computer Interface: Control of a 2 DoF Artificial Upper Limb

Abstract

Event Abstract Back to Event Asynchronous Steady-State Visual Evoked Potential based Brain-Computer Interface: Control of a 2 DoF Artificial Upper Limb Aim: The realization of an asynchronous (i.e. independent of external cues) steady-state visual evoked potential (SSVEP) based brain-computer interface (BCI) according to the requirements of a 2 degrees of freedom (DOF) hand and elbow neuroprosthesis [1]. Methods: The visual stimulation was delivered via two LED bars, flickering with 8 and 13 Hz respectively. Eight healthy subjects and one tetraplegic after spinal cord injury (SCI, cervical level sub C5) participated in the online experiment. The EEG data, recorded at twenty one occipitally mounted EEG electrodes during the initial cue based calibration (320 trials, each lasting for 6 s), was used to select the six (five for the SCI subject) channels for the following online experiment. For the SSVEP frequency recognition canonical correlation analysis (CCA) was used; it is based on the premise that the measured SSVEP will contain the same frequency as the stimulus signal [2]. The performance of the SSVEP-BCI was evaluated in online control of a virtual and a robotic 2 DoF limb. A total of 8 (4 for the SCI subject) runs were conducted, whereby a run was divided into self-paced intentional control (IC, twice per run) and timed non-control (NC, 150s per run) periods. During the IC periods, the subjects performed the following predefined movement sequence, consisting of hand opening/closing (HO/HC) and elbow flexion/extension (EF/EE): HO, HC, EF, EE, HO, HC. True positive and false negative decisions were detected from the movement sequence during the IC periods, and false positive decisions were detected during the NC periods. From these numbers, the positive predictive value (PPV) and the false negative rate (FNR) were calculated. A PPV of 100% means that all the commands were intended by the subject, whereas a FNR of 0% means that all intended commands were detected. Results: We showed that eight healthy subjects and one tetraplegic patient could control a 2-DOF artificial limb with the asynchronous SSVEP-BCI. In the online experiments, the PPV varied between 69% and 83% (76 ± 4% for all nine participants), and the FNR varied between 1% and 17% (8 ± 5%). Conclusion: Our SSVEP- and a similar MI-BCI [1] approach exhibited a comparable number of average FPs per minute of NC period (1.00-2.11 for the SSVEP- and 0.20-3.10 for the MI-BCI), but the SSVEP-BCI averaged fewer FNs per movement sequence (0.06-1.25 compared to 1.38-6.50 for the MI-BCI). However, the MI-BCI does not require any external stimuli. Acknowledgments: This work was supported by EU COST Action BM0601 (Neuromath) and Wings for Life -Spinal Cord Research Foundation, project WFL-SE-016/09.