Abstract
Motor imagery (MI) based brain computer interfaces (BCI) detect changes in brain activity associated with imaginary limb movements, and translate them into device commands. MI based BCIs require training, during which the user gradually learns how to control his or her brain activity with the help of feedback. Additionally, machine learning techniques are frequently used to boost BCI performance and to adapt the decoding algorithm to the user's brain. Thus, both the brain and the machine need to adapt in order to improve performance. To study the utility of co-adaptive training in the BCI paradigm and the time scales involved, we investigated the performance of two groups of subjects, in a 4-day MI experiment using EEG recordings. One group (control, n = 9 subjects) performed the BCI task using a fixed classifier based on MI data from day 1. In the second group (experimental, n = 9 subjects), the classifier was regularly adapted based on brain activity patterns during the experiment days. We found that the experimental group showed a significantly larger change in performance following training compared to the control group. Specifically, although the experimental group exhibited a decrease in performance between days, it showed an increase in performance within each day, which compensated for the decrease. The control group showed decreases both within and between days. A correlation analysis in subjects who had a notable improvement in performance following training showed that performance was mainly associated with modulation of power in the α frequency band. To conclude, continuous updating of the classification algorithm improves the performance of subjects in longitudinal BCI training.
Highlights
Brain-computer interface (BCI) systems translate brain signals, e.g., electroencephalography (EEG) into control commands for a computer application or a neuroprosthesis
We found a significant difference in performance changes during training between the experimental and the control groups (Control: Mean = –2.3%, SD = 5.8%; Experimental: Mean = +11.3%, SD = 16.8%; p = 0.023)
While in the control group there was no significant change in performance (p = 0.136), the experimental group showed a significant improvement across the multi-day experiment (p = 0.04)
Summary
Brain-computer interface (BCI) systems translate brain signals, e.g., electroencephalography (EEG) into control commands for a computer application or a neuroprosthesis. A popular paradigm for BCI communication is motor imagery (MI) (Wolpaw and Wolpaw, 2012; Perdikis et al, 2016; Schultze-Kraft et al, 2017) In this paradigm, the user imagines performing a movement with a particular limb, a process which alters the rhythmic activity in locations in the sensorimotor cortex that correspond to the imagined limb. Apart from the learning of the subjects, learning takes place within the decoding component by adapting classifier parameters in a way that reduces performance errors (Vidaurre et al, 2011a,b; Perdikis et al, 2016) This classifier adaptation process can occur in parallel with human learning, thereby potentially reducing the amount of practice needed to achieve an effective use for the BCI system
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
