Brain-machine interface cursor position only weakly affects monkey and human motor cortical activity in the absence of arm movements

Sergey D Stavisky,Leigh R Hochberg,Christine Blabe,Stephen I Ryu,Jaimie M Henderson,Chethan Pandarinath,Krishna V Shenoy,Jonathan C Kao,Paul Nuyujukian

doi:10.1038/s41598-018-34711-1

Abstract

Brain-machine interfaces (BMIs) that decode movement intentions should ignore neural modulation sources distinct from the intended command. However, neurophysiology and control theory suggest that motor cortex reflects the motor effector’s position, which could be a nuisance variable. We investigated motor cortical correlates of BMI cursor position with or without concurrent arm movement. We show in two monkeys that subtracting away estimated neural correlates of position improves online BMI performance only if the animals were allowed to move their arm. To understand why, we compared the neural variance attributable to cursor position when the same task was performed using arm reaching, versus arms-restrained BMI use. Firing rates correlated with both BMI cursor and hand positions, but hand positional effects were greater. To examine whether BMI position influences decoding in people with paralysis, we analyzed data from two intracortical BMI clinical trial participants and performed an online decoder comparison in one participant. We found only small motor cortical correlates, which did not affect performance. These results suggest that arm movement and proprioception are the major contributors to position-related motor cortical correlates. Cursor position visual feedback is therefore unlikely to affect the performance of BMI-driven prosthetic systems being developed for people with paralysis.

Highlights

Brain-machine interfaces (BMIs) that decode movement intentions should ignore neural modulation sources distinct from the intended command
We have presented multiple complementary lines of evidence that support the same conclusion: motor cortical correlates of the position of a two-dimensional BMI-controlled cursor, without accompanying arm movements, are weak and unlikely to be a meaningful nuisance variable during decoding
We first showed that subtracting the expected cursor position’s neural contribution as described in[19] did improve BMI performance if the monkeys were allowed to move their arm during BMI use, but not when the BMI was used in the absence of arm movements (Fig. 1)

Summary

Introduction

Brain-machine interfaces (BMIs) that decode movement intentions should ignore neural modulation sources distinct from the intended command. BMIs are an emerging medical technology that can be used to bypass motor disabilities due to injury and disease These systems read out the movement intentions of people with paralysis to restore function[1], for example by controlling computer cursors for communication[2], commanding arm and hand movements of robotic limbs[3], or electrically stimulating the person’s own paralyzed muscles[4]. Motor cortical modulation due to effector state estimation and visual feedback is germane to BMI decoding, since these factors would presumably still affect the observed neural activity even when BMI output commands should be the same. Nuyujukian and colleagues showed that this operation improves performance in monkeys, but this test was performed during BMI use accompanied by arm reaching[19] During this behavioral context, motor cortex is expected to receive strong proprioceptive inputs and contribute efferent commands to the arm. When ReFIT was successfully translated to clinical study human BMI users[2,24], the Cursor Position Subtraction operation led to mixed results[24]

Methods

Results

Conclusion