Decoding Lower Limb Muscle Activity and Kinematics from Cortical Neural Spike Trains during Monkey Performing Stand and Squat Movements

Xuan Ma, Chaolin Ma, Jian Huang, Peng Zhang, Jiang Xu, Jiping HeFrontiers in Neuroscience 2017 · DOI

Abstract

Extensive literatures have shown approaches for decoding upper limb kinematics or muscle activity using multichannel cortical spike recordings toward brain machine interface (BMI) applications. However, similar topics regarding lower limb remain relatively scarce. The current study investigates whether lower limb kinematics and muscle activity characterized by electromyography (EMG) signals during monkey performing stand/squat movements can be accurately decoded from neural spike trains in primary motor cortex (M1). Using a recursive Bayesian estimation framework, specifically an unscented Kalman filter incorporated with an artificial neural network, high correlation coefficients and signal-to-noise ratios were achieved. These results demonstrate that lower limb EMG signals and kinematics can be accurately decoded from M1 neurons, providing new insights for extending BMI techniques to lower limb prostheses and exoskeletons.

Introduction

While Brain-Machine Interfaces (BMIs) for upper limbs—such as controlling robotic arms for reaching and grasping—have seen significant progress, the lower limbs remain a "final frontier." Lower limb movements are fundamentally different; they often rely on rhythmic, automatic circuits in the spinal cord rather than just direct cortical control. For patients with spinal cord injuries, restoring the ability to stand and squat is a high priority. This research seeks to bridge the gap by proving that primary motor cortex (M1) signals contain the necessary "blueprints" for complex, voluntary lower-limb motor tasks.

Experimental Design and Methods

The study utilized data from two male Rhesus monkeys trained to perform visually guided stand-and-squat tasks in a specialized primate chair. The researchers recorded multichannel spike trains from M1 (909 neurons from Monkey 1; 689 from Monkey 2) alongside EMG signals from eight specific muscles in the right leg and 3D kinematic data of the ankle.

Figure 1: Experimental apparatus and behavioral task sequence. (A) Primate chair with movable pedal. (B) Sequence of visual cues and corresponding monkey behavior during the stand-and-squat task.

The researchers employed a two-stage approach:

  1. Functional Mapping: During acute sessions, Spike Triggered Averaging (SpTA) was used to correlate individual neural spikes with muscle contractions. This created "density maps" to identify optimal locations in M1 for chronic electrode implantation.
  2. Decoding Framework: Two recursive Bayesian algorithms were compared: a standard linear Kalman Filter (KF) and an Unscented Kalman Filter (UKF) paired with an Artificial Neural Network (ANN). The ANN was incorporated to handle the non-linear relationship between neural firing and muscle activity.

Results and Key Insights

Anatomical Clustering in the Brain

The SpTA analysis revealed that neurons controlling specific muscles are not scattered randomly. Muscles sharing anatomical similarities (e.g., innervated by the same nerve) have corresponding neurons that cluster together in M1. This provides a spatial guide for future electrode implantations.

Figure 2: Spatial distribution mapping of muscle-innervating neurons in M1, showing PA (Posterior-Anterior) and DL (Dorso-Lateral) coordinates.

Decoding Performance

  • Non-linearity is Key: The UKF-ANN hybrid significantly outperformed the standard linear Kalman filter, improving the correlation coefficient (CC) by 13.6% and the Signal-to-Noise Ratio (SNR) by 17.5%. This suggests the brain's communication with muscles is inherently non-linear.
  • High Correlation: For EMG decoding, the UKF achieved a mean CC > 0.75 for most muscles. Kinematic position was predicted with high accuracy (CC ~0.90), though the system struggled with high-frequency acceleration (CC < 0.4).
  • Voluntary Control: The high decoding accuracy suggests that M1 is deeply involved in the planning and execution of these specific voluntary "posture-changing" movements, as opposed to purely rhythmic gait.

Conclusion and Future Work

This study demonstrates that lower limb muscle activity and kinematics can be accurately extracted from M1 neural activity. While the failure to accurately decode acceleration remains a hurdle for real-time robotics requiring sudden speed changes, the current results are promising for the development of brain-controlled exoskeletons. The next logical step is "closed-loop" testing, where neural signals control an actual device in real-time to assist the subject's movement.

Summary updated

Abstract

Extensive literatures have shown approaches for decoding upper limb kinematics or muscle activity using multichannel cortical spike recordings toward brain machine interface (BMI) applications. However, similar topics regarding lower limb remain relatively scarce. We previously reported a system for training monkeys to perform visually guided stand and squat tasks. The current study, as a follow-up extension, investigates whether lower limb kinematics and muscle activity characterized by electromyography (EMG) signals during monkey performing stand/squat movements can be accurately decoded from neural spike trains in primary motor cortex (M1). Two monkeys were used in this study. Subdermal intramuscular EMG electrodes were implanted to 8 right leg/thigh muscles. With ample data collected from neurons from a large brain area, we performed a spike triggered average (SpTA) analysis and got a series of density contours which revealed the spatial distributions of different muscle-innervating neurons corresponding to each given muscle. Based on the guidance of these results, we identified the locations optimal for chronic electrode implantation and subsequently carried on chronic neural data recordings. A recursive Bayesian estimation framework was proposed for decoding EMG signals together with kinematics from M1 spike trains. Two specific algorithms were implemented: a standard Kalman filter and an unscented Kalman filter. For the latter one, an artificial neural network was incorporated to deal with the nonlinearity in neural tuning. High correlation coefficient and signal to noise ratio between the predicted and the actual data were achieved for both EMG signals and kinematics on both monkeys. Higher decoding accuracy and faster convergence rate could be achieved with the unscented Kalman filter. These results demonstrate that lower limb EMG signals and kinematics during monkey stand/squat can be accurately decoded from a group of M1 neurons with the proposed algorithms. Our findings provide new insights for extending current BMI design concepts and techniques on upper limbs to lower limb circumstances. Brain controlled exoskeleton, prostheses or neuromuscular electrical stimulators for lower limbs are expected to be developed, which enables the subject to manipulate complex biomechatronic devices with mind in more harmonized manner.
BibTeX
@inproceedings{ma2017decoding,
  author = {Xuan Ma and Chaolin Ma and Jian Huang and Peng Zhang and Jiang Xu and Jiping He},
  title = {{Decoding Lower Limb Muscle Activity and Kinematics from Cortical Neural Spike Trains during Monkey Performing Stand and Squat Movements}},
  booktitle = {Frontiers in Neuroscience},
  year = {2017},
  doi = {10.3389/fnins.2017.00044},
}
Figure