Real-Time Brain-Computer Interface Control of Walking Exoskeleton with Bilateral Sensory Feedback

TL;DR

This study addresses restoring gait and bilateral leg sensation in paraplegia by implementing a bidirectional brain-computer interface (BDBCI) that uses bilateral interhemispheric leg motor/sensory cortices. The authors built an embedded BDBCI-RGE system comprising motor decoding and direct cortical sensory stimulation, interfacing with an FDA-cleared robotic exoskeleton, all operating on-device with a wireless base station for setup. In a single able-bodied epilepsy patient with implanted interhemispheric ECoG grids, they demonstrated real-time brain-controlled walking with bilateral artificial leg percepts, achieving high decoding performance (up to 0.97 on online tests; average around 0.92) and safe, artifact-free operation through interleaved decoding and stimulation. The work advances toward fully implantable, untethered BDBCI gait restoration, outlining concrete steps toward translation to SCI patients and highlighting limitations such as single-subject testing and seated rather than standing ambulation.

Abstract

Invasive brain-computer interface (BCI) technology has demonstrated the possibility of restoring brain-controlled walking in paraplegic spinal cord injury patients. However, current implementations of BCI-controlled walking still have significant drawbacks. In particular, prior systems are unidirectional and lack sensory feedback for insensate patients, have suboptimal reliance on brain signals from the bilateral arm areas of the motor cortex, and depend on external systems for signal processing. Motivated by these shortcomings, this study is the first time a bidirectional brain-computer interface (BDBCI) has demonstrated the restoration of both brain-controlled walking and leg sensory feedback while utilizing the bilateral leg motor and sensory cortices. Here, a subject undergoing subdural electrocorticogram electrode implantation for epilepsy surgery evaluation leveraged the leg representation areas of the bilateral interhemispheric primary motor and sensory cortices to operate a BDBCI with high performance. Although electrode implantation in the interhemispheric region is uncommon, electrodes can be safely implanted in this region to access rich leg motor information and deliver bilateral leg sensory feedback. Finally, we demonstrated that all BDBCI operations can be executed on a dedicated, portable embedded system. These results indicate that BDBCIs can potentially provide brain-controlled ambulation and artificial leg sensation to people with paraplegia after spinal cord injury in a manner that emulates full-implantability and is untethered from any external systems.

Figures (5)

• Figure 1: a: System diagram for BDBCI-RGE. ECoG signals are decoded to determine motor intent, which is then used to wirelessly control the RGE. RGE Interface detects right/left leg swing and commands the the BDBCI stimulator module to elicit the corresponding sensation using DCES. b: Co-registration of ECoG electrode locations (post-implantation CT image) onto a 3D render of subject's brain (post-implantation MRI). Electrodes used for motor decoding on the left/right interhemispheric grids are colored teal. Electrodes used for sensory stimulation are colored yellow. c: Diagram summarizing BDBCI-RGE task. Subject seated in ICU bed (chair configuration) wirelessly controls the RGE by following cues to idle (relax body) or move (seated stepping motion with both legs). An experimenter wearing the RGE being walked down a pathway in response to BDBCI control, triggering sensory feedback (DCES) to the subject on each step. d: Photo of BDBCI-RGE task.
• Figure 2: Left: Spatial weights of motor decoding model mapped to locations of corresponding coregistered ECoG electrodes in the left and right interhemispheric grids. Warmer colors indicate electrodes containing relevant decoding features for the given frequency band (high-$\beta$ or $\gamma$), whereas cooler colors indicate electrodes that were less relevant for decoding. Right: Time-domain power envelopes from a subset of training data in the electrodes with the strongest decoder weights (two electrodes are shown from each side). The $\gamma$ power increased during the "MOVE" state (i.e., transitioning from red shade to blue shade). Conversely, high-$\beta$ power decreased during the "MOVE" state (desynchronizations). Red and blue shades: "IDLE" and "MOVE" states, respectively. Red and blue horizontal bars: median band powers in each state. White gaps represent brief discontinuities between recording epochs, where no data were acquired.
• Figure 3: Best-case BDBCI-RGE online run. Orange: "Instructional Cue" blocks indicate periods where the "MOVE" cue was presented to the subject. Teal: "Decoded State" blocks indicate when a "MOVE" state was detected. The $P(M|f)$ line indicates the probability of "MOVE", as determined by the decoder. Each dot on the line indicates the end of one online window. $T_M$ and $T_I$ indicate the state transition thresholds to which $P(M|f)$ was compared. If the current decoded state is "IDLE" and $P(M|f) > T_M$, the decoded state changes to "MOVE". Conversely, if the current decoded state is "MOVE" and $P(M|f) < T_I$, the decoded state changes to "IDLE". Once the decoded state is "MOVE", the RGE begins to move, as indicated by the L/R Hip Angular Velocity from the motion tracking suit worn by the experimenter in the RGE. The L/R stimulation indicates the contralateral S1 stimulation corresponding to the movement of the matching RGE leg.
• Figure 4: State machine for BDBCI decoder. The state transitions between the Idle and Move state are governed by the state transition rules defined in the figure.
• Figure 5: A: High-level state diagram for online motor decoding, RGE control and stimulation operations. (1) BDBCI acquires and stores windows of ECoG data. (2) Decoded state determined using onboard training model. (3) BDBCI acts based on the decoded state. If the IDLE state is decoded, system returns to (1). If the MOVE state is decoded, (4) actuation instructions are transmitted to the RGE Interface, (5) RGE leg swing is actuated, (6) RGE leg swing triggers sensory stimulation in the corresponding leg, and upon completion of the swing, BDBCI returns to (1). B: Example of BDBCI-RGE execution in real time. The first $\sim$13 s demonstrate operations while the decoded state is IDLE. BDBCI executes operations (1) -- (3) until the decoded state changes to "MOVE," upon which decoding operations begin to be interleaved with RGE actuation and sensory stimulation (4)--(6).