Brain-computer interface (BCI) for stroke rehabilitation : effects of optimal electrode channel selection and tactile feedback in chronic stroke patients

Abstract

Brain-computer interfaces (BCIs) utilize the non-muscular channel of the brain to communicate with the outside world. Several recent studies have tried to use BCIs for stroke rehabilitation because BCIs do not rely on the residual motor function of stroke patients, unlike other conventional techniques, for example, constraint-induced movement therapy. This independence of the BCI from residual motor function allows for the training of patients with limited motor function, for example, in the subacute stage, where rehabilitation is believed to be the most effective. Major hurdles preventing the successful adoption of BCIs for stroke rehabilitation include the long electrode preparation time and lack of understanding on the most effective form of feedback modality. This study aims to suggest the optimal configurations in terms of channel selection and feedback modality for the use of BCI in stroke rehabilitation. It is a challenge to reduce the electrode preparation time while maintaining high classification accuracy. Currently, the effect of channel reduction on stroke patients is not clear, because most channel studies only focus on healthy subjects. Information regarding the optimal number of calibration sessions for channel selection is also lacking. It is unclear as to how many sessions are necessary in order to reveal the "invariant" features of subject's EEG during long-term BCI usage. There is also no single objective measure to take account of the compromise between accuracy, preparation time and number of calibration sessions for channel selection, such that different channel selection methods can be examined comparatively. This study aims to find a minimal set of electrodes for an individual stroke subject for motor imagery to control an assistive device using functional electrical stimulation for 20 sessions with an accuracy of more than 90%. The characteristics of this minimal electrode set were evaluated using two popular algorithms: Fisher's criterion and the support-vector machine recursive feature elimination (SVM-RFE). The number of calibration sessions for channel selection required for the robust control of these 20 sessions was also investigated. Five chronic stroke patients were recruited for the study. Our results suggested that the number of calibration sessions for channel selection did not have a significant effect on the classification accuracy. A performance index devised in this study showed that one training day with 12 electrodes using the SVM-RFE method achieved the best balance between the number of electrodes and accuracy in the 20-session data. Generally, 836 channels were required to maintain accuracy higher than 90% in 20 BCI training sessions for chronic stroke patients. The effects of different feedback modalities for stroke rehabilitation are currently not well-understood. Most BCIs rely on visual feedback due to its intuitive nature. However, studies have shown that successful stroke recovery relies heavily on the active participation (i.e. activation of the motor area) of patients. The Schema theory of motor learning predicts that continuous feedback can facilitate faster learning; however, until now, very few BCIs have incorporated continuous feedback into their training paradigm. Research on tactile stimulation has also shown that tactile feedback can activate the sensorimotor area of the brain, but studies devoted to investigating the effect of tactile feedback on stroke rehabilitation are very limited. This study aims to compare directly the effect of continuous tactile feedback (CTF) and visual feedback by developing a BCI with CTF capacity and then testing it on stroke subjects. CTF provides continuous feedback to the user intention by varying the speed and direction of the robotic hand opening. Eight chronic stroke subjects were recruited for this study. Each of them learned to control a BCI with visual feedback and CTF, respectively. Our results revealed that CTF produced significantly higher classification accuracy than visual feedback alone (p<0.05). Event-related desynchronization during CTF was also stronger than visual feedback at the sensorimotor area for most subjects. Our results showed that CTF provide a better feedback modality than visual feedback for stroke rehabilitation.