Cortical Decoding of Individual Finger Group Motions Using ReFIT Kalman Filter

Alex K. Vaskov, Zachary T. Irwin, Samuel R. Nason, Philip P. Vu, Chrono S. Nu, Autumn J. Bullard, Mackenna Hill, Naia North, Parag G. Patil, Cynthia A. ChestekFrontiers in Neuroscience 2018 · DOI

TL;DR

This paper introduces a novel manipulandum and the ReFIT Kalman filter to achieve precise, online decoding of individual finger group movements in non-human primates. This work represents a significant step towards enabling more dexterous prosthetic control by demonstrating improved performance and reduced "orbiting" (unwanted movement around a target) compared to standard decoding methods.

What Problem Does This Paper Address?

Current brain-machine interfaces (BMIs) for neuroprosthetics have made strides in restoring arm movement and basic grasping. However, achieving fine-grained control over individual fingers remains a significant challenge. Existing systems often struggle with precision as the number of controlled degrees of freedom increases, and they haven't fully quantified performance for precise finger control. This paper aims to bridge this gap by investigating the online decoding of individual finger group movements, specifically the index finger and the middle-ring-pinkie (MRP) group, to pave the way for more dexterous prosthetic hands.

Methods

The researchers developed a novel manipulandum, a device designed to isolate the movements of specific finger groups. This device uses resistive flex sensors to measure finger flexion and extension. For decoding, they employed the ReFIT (Recalibrated Feedback Intention-Trained) Kalman filter. This is a two-stage training approach that first uses an initial Kalman filter for online control and then refines the decoder by incorporating "intention estimation." This process essentially "corrects" predicted movements by assuming the user intends to move towards the target, even if the initial decoder's prediction is off. This method was applied to decode single degrees of freedom (DOF) for either the index finger or the MRP finger group.

Datasets

The study utilized data collected from two rhesus macaques (Monkey W and Monkey N) implanted with Utah arrays in their motor cortex. The experiments involved training and testing the decoding algorithms on these animals performing a target-matching task using the novel manipulandum. The data collected served as both training data for the Kalman and ReFIT filters and as evaluation data for online performance metrics.

Key Insights and Ideas

  • Finger-specific neural signals are decodable: The study demonstrates that neural activity in the motor cortex contains sufficient information to distinguish and decode movements of individual finger groups.
  • ReFIT Kalman filter significantly improves performance: The ReFIT algorithm consistently outperformed the standard Kalman filter, particularly in reducing "orbiting" (unwanted movement around a target) and improving overall decoding accuracy.
  • Novelty of precise finger control decoding: This work is highlighted as the first systematic and biomimetic separation of digits for continuous online decoding in NHPs, pushing the boundaries of BMI capabilities.
  • Potential need for non-linear approaches: The results suggest that while linear decoders are sufficient for gross movements, more complex, potentially non-linear, approaches might be necessary for achieving truly independent and precise control of individual fingers.

Key Results

  • Both monkeys were able to reliably acquire fingertip targets using either the index or MRP finger groups.
  • The ReFIT Kalman filter achieved a 31.0% and 35.2% improvement in bit rate compared to the standard Kalman filter for the index and MRP finger groups, respectively.
  • Index and MRP finger movements could be differentiated with 81.7% accuracy across both subjects.
  • The ReFIT decoder significantly reduced orbiting time, leading to an average improvement in center-out acquisition time by 33.43%.
  • The decoders remained robust even when the manipulandum was removed, suggesting reliance on motor activity rather than solely proprioceptive feedback.

Limitations and Future Work

  • Linear decoder limitations for co-contraction: The study found that a linear Kalman filter was insufficient for decoding both finger groups simultaneously due to significant unwanted movement in the stationary finger, likely caused by co-contraction. This suggests that more advanced decoding strategies might be needed for complex, interdependent finger movements.
  • Task complexity and intention estimation: The authors hypothesize that intention estimation might be more impactful for decoding multiple degrees of freedom simultaneously, as the 1-DOF task in this study might have a higher chance of initial decoder success, potentially reducing the perceived benefit of intention estimation.
  • Future directions: The researchers plan to extend this work to enable precise online control in a 2-DOF task involving both index and MRP finger groups. They also suggest exploring more accurate classification schemes to suppress unwanted movements, dimensionality reduction techniques, and incorporating models of natural muscle synergies or non-linear machine learning approaches like neural networks. Further investigation into the role of sensory context and the potential for adaptive BMI systems is also proposed.

Summary updated

Abstract

<b>Objective:</b> To date, many brain-machine interface (BMI) studies have developed decoding algorithms for neuroprostheses that provide users with precise control of upper arm reaches with some limited grasping capabilities. However, comparatively few have focused on quantifying the performance of precise finger control. Here we expand upon this work by investigating online control of individual finger groups. <b>Approach:</b> We have developed a novel training manipulandum for non-human primate (NHP) studies to isolate the movements of two specific finger groups: index and middle-ring-pinkie (MRP) fingers. We use this device in combination with the ReFIT (Recalibrated Feedback Intention-Trained) Kalman filter to decode the position of each finger group during a single degree of freedom task in two rhesus macaques with Utah arrays in motor cortex. The ReFIT Kalman filter uses a two-stage training approach that improves online control of upper arm tasks with substantial reductions in orbiting time, thus making it a logical first choice for precise finger control. <b>Results:</b> Both animals were able to reliably acquire fingertip targets with both index and MRP fingers, which they did in blocks of finger group specific trials. Decoding from motor signals online, the ReFIT Kalman filter reliably outperformed the standard Kalman filter, measured by bit rate, across all tested finger groups and movements by 31.0 and 35.2%. These decoders were robust when the manipulandum was removed during online control. While index finger movements and middle-ring-pinkie finger movements could be differentiated from each other with 81.7% accuracy across both subjects, the linear Kalman filter was not sufficient for decoding both finger groups together due to significant unwanted movement in the stationary finger, potentially due to co-contraction. <b>Significance:</b> To our knowledge, this is the first systematic and biomimetic separation of digits for continuous online decoding in a NHP as well as the first demonstration of the ReFIT Kalman filter improving the performance of precise finger decoding. These results suggest that novel nonlinear approaches, apparently not necessary for center out reaches or gross hand motions, may be necessary to achieve independent and precise control of individual fingers.
BibTeX
@inproceedings{vaskov2018cortical,
  author = {Alex K. Vaskov and Zachary T. Irwin and Samuel R. Nason and Philip P. Vu and Chrono S. Nu and Autumn J. Bullard and Mackenna Hill and Naia North and Parag G. Patil and Cynthia A. Chestek},
  title = {{Cortical Decoding of Individual Finger Group Motions Using ReFIT Kalman Filter}},
  booktitle = {Frontiers in Neuroscience},
  year = {2018},
  doi = {10.3389/fnins.2018.00751},
}

Loading related content...

Figure