[Paper Review] Decoding hand kinematics from population responses in sensorimotor cortex during grasping

Elizaveta V. Okorokova, James M. Goodman+2 more|arXiv (Cornell University)|Apr 6, 2019

EEG and Brain-Computer Interfaces54 references4 citations

TL;DR

This study decodes hand kinematics from population neural activity in primary motor (M1) and somatosensory (S1) cortices during grasping in three monkeys. Using a small number of neurons, the researchers achieved high-accuracy decoding of joint angles, outperforming velocity decoding—demonstrating the feasibility of using cortical signals for dexterous brain-machine interfaces and highlighting the potential of somatosensory cortex for closed-loop neuromodulation.

ABSTRACT

The hand, a complex effector comprising dozens of degrees of freedom of movement, endows us with the ability to flexibly, precisely, and effortlessly interact with objects. The neural signals associated with dexterous hand movements in primary motor cortex (M1) and somatosensory cortex (SC) have received comparatively less attention than have those that are associated with proximal limb control. To fill this gap, we trained three monkeys to grasp objects varying in size, shape and orientation while tracking their hand postures and recording single-unit activity from M1 and SC. We then decoded their hand kinematics across 30 joints from population activity in these areas. We found that we could accurately decode kinematics with a small number of neural signals and that performance was higher for decoding joint angles than joint angular velocities, in contrast to what has been found with proximal limb decoders. We conclude that cortical signals can be used for dexterous hand control in brain machine interface applications and that postural representations in SC may be exploited via intracortical stimulation to close the sensorimotor loop.

Motivation & Objective

• To investigate how population activity in primary motor (M1) and somatosensory (S1) cortices encodes dexterous hand movements during grasping.
• To determine whether neural signals from M1 and S1 can be used to decode fine-grained hand kinematics across 30 joints.
• To compare decoding performance between joint angles and angular velocities in the context of hand movement.
• To evaluate the potential of somatosensory cortex representations for use in intracortical stimulation-based brain-machine interfaces.

Proposed method

• Monkeys performed grasping tasks with objects of varying size, shape, and orientation while hand posture was tracked in 3D.
• Single-unit neural activity was recorded simultaneously from M1 and S1 during task performance.
• A linear decoder was trained to predict hand joint angles and angular velocities from population neural activity.
• Decoding performance was evaluated using cross-validated correlation coefficients between predicted and actual kinematics.
• The study compared decoding accuracy between M1 and S1, and between joint angle and velocity representations.
• A population-based decoding framework was applied to assess the minimal number of neurons required for accurate kinematic reconstruction.

Experimental results

Research questions

• RQ1 Can population activity in M1 and S1 accurately decode 30-joint hand kinematics during natural grasping movements?
• RQ2 How does decoding performance differ between joint angles and angular velocities in the sensorimotor cortex?
• RQ3 Which cortical area—M1 or S1—provides more informative signals for dexterous hand kinematics?
• RQ4 What is the minimum number of neurons required to achieve high-accuracy decoding of hand posture?
• RQ5 Can somatosensory cortex representations be leveraged for closed-loop brain-machine interfaces via intracortical stimulation?

Key findings

• Decoding of joint angles achieved significantly higher accuracy than decoding of angular velocities, contradicting findings from proximal limb decoding.
• High-accuracy reconstruction of hand kinematics was achieved using only a small population of neurons from M1 and S1.
• Somatosensory cortex (S1) provided robust representations of hand posture, suggesting its potential for use in closed-loop brain-machine interfaces.
• The study demonstrated that cortical signals from both M1 and S1 can be effectively decoded to reconstruct complex hand movements.
• The results support the feasibility of using intracortical neural signals for controlling dexterous prosthetic hands in brain-machine interface systems.