Towards a Real-Time Motor Imagery-based BCI with FES

There are many patients in the US and around the world who have been diagnosed with a stroke, or other ailments caused by a neurological problem that precludes them from conscious motor control. For these patients, rehabilitation can be difficult to regain any voluntary motor function. Functional Electrical Stimulation (FES), which involves using electricity to artificially stimulate nerves in the body, can be used to stimulate the muscle neurons in the patients’ affected limbs to improve their rehabilitation outcomes. To allow the patients to control the FES stimulation voluntarily, Motor Imagery (MI) based Brain-Computer Interfaces (BCIs) can be used. BCIs are typically used as a direct communication pathway between the brain and a computer for individuals who cannot otherwise control devices through normal methods, either due to illness or injury to the central nervous system. Motor Imagery entails simply thinking about moving a limb, and these thoughts can then be used as inputs by a BCI for controlling an external device, which in this study is an FES unit. Previous studies have shown that this paradigm increases rehabilitation performance in stroke patients. However, previous research has only been conducted with BCI control of one limb in this manner. The goal of this study is to validate if this research can be extended to control two limbs, which could increase the rehabilitation outcomes of stroke patients.

Subjects were seated in a comfortable armchair with two sets of FES electrodes attached to their forearms. The FES electrodes were attached to an FES stimulator that was modified to connect to an external circuit board. This allows for stimulation parameters to be controlled by a computer by simulating button presses on the FES unit to configure it as desired. The subjects wore an EEG cap with 8 active EEG electrodes connected to a g.USBamp biosignal amplifier for recording their brain activity. The electrical activity of the brain was sampled at 512 Hz and bandpass filtered from 0.5 Hz to 30 Hz to remove unwanted noise, and notch filtered at 60 Hz to remove mains hum.

Subjects were presented with a series of trials that involved giving the subject a cue, left or right, as to which hand to imagine grasping. As the subject imagined grasping their hand, the MI classifier would take the recorded EEG data in 500ms blocks, calculate the Event-Related Desynchronization (ERD), and then determine which hand the subject imagined grasping. The amplitude of the FES stimulation applied to the attended limb was then changed in accordance with the MI classification output. Classification accuracies generated from the motor imagery tasks showed above 70% for all subjects, which is a commonly accepted accuracy threshold accepted by the BCI community (Kubler et al. 2004).