A unifying model of concurrent spatial and temporal modularity in muscle activity

Ioannis Delis, Stefano Panzeri, Thierry Pozzo, Bastien BerretJournal of Neurophysiology 2013 · DOI

TL;DR

This paper introduces a novel "space-by-time" decomposition model and a corresponding algorithm (sNM3F) to unify existing approaches for understanding muscle activation patterns. By simultaneously extracting spatial and temporal modules, the model provides a more flexible and comprehensive representation of motor control, offering insights into how the brain generates complex movements from simpler building blocks.

What Problem Does This Paper Address?

The control of movement, particularly generating a wide range of actions, is thought to rely on a "modular" organization in the central nervous system. This means the brain uses a limited set of fundamental "modules" that are combined to produce diverse movements. Researchers have identified evidence for this modularity by looking at muscle activation patterns (electromyography, or EMG). However, previous methods for identifying these muscle modules differed in their assumptions about the nature and dimensionality of these modules, leading to a fragmented understanding. This paper aims to bridge this gap by proposing a unified framework that can encompass and generalize existing models, offering a more complete picture of how spatial and temporal aspects of muscle activity contribute to motor control.

Methods

The core contribution is the space-by-time decomposition model. This model represents muscle activity as a combination of spatial modules (which describe how muscles are grouped and co-activate) and temporal modules (which describe the timing and activation patterns of these groups). This is a departure from previous models that focused on either temporal or spatial aspects exclusively, or treated them as inseparable spatiotemporal units.

To extract these modules, the authors developed a new algorithm called sample-based non-negative matrix tri-factorization (sNM3F). This algorithm is an extension of non-negative matrix factorization (NMF) and is designed to simultaneously identify both spatial and temporal modules from EMG data. The sNM3F algorithm aims to find a low-dimensional representation of the data by minimizing reconstruction error, while ensuring that the extracted modules are "sample-independent" (meaning they are consistent across different trials of a task). The paper also explores variants of this algorithm, including one that incorporates time-shifts to account for variations in movement timing.

Datasets

The primary dataset used in this study was collected from six healthy right-handed participants performing arm pointing movements in the horizontal plane. The experiment involved 16 distinct motor tasks, varying in direction and speed, with each task consisting of 40 trials. EMG activity was recorded from 9 upper body and arm muscles. The paper also mentions using simulated data to test the algorithm's effectiveness and robustness.

Key Insights and Ideas

  • Unifying Framework: The space-by-time decomposition successfully integrates previous models (temporal, spatial, and spatiotemporal synergies) into a single, more comprehensive framework.
  • Primal Building Blocks: The identified spatial and temporal modules appear to be fundamental building blocks that can be combined to generate more complex spatiotemporal muscle patterns, including time-varying synergies.
  • Efficient Trade-off: The model achieves a good balance between dimensionality reduction (simplifying motor control) and task discriminability (ability to distinguish between different movements).
  • Functional Interpretation: The extracted modules have clear anatomical and functional interpretations, with spatial modules representing muscle groupings (e.g., elbow flexors) and temporal modules reflecting distinct phases of muscle activation during movement.
  • Robustness: The identified modules are robust across subjects and consistent with findings from other studies, suggesting a generalizable principle of motor control.

Key Results

  • The space-by-time decomposition, using 4 spatial and 3 temporal modules, achieved 68% Variance Accounted For (VAF) and 80% decoding performance on the arm pointing task dataset.
  • The temporal modules identified corresponded to three independently controlled activation bursts, consistent with known phases of goal-directed movement.
  • The spatial modules revealed anatomically and functionally interpretable muscle groupings, such as elbow flexors, elbow extensors, shoulder flexors, and shoulder extensors, which act as agonist-antagonist pairs.
  • The model demonstrated task-dependence, showing that activation parameters varied with movement direction and speed, with speed modulation being particularly apparent.
  • The sNM3F algorithm showed significantly better performance and robustness compared to alternative methods like successive NMFs, especially when dealing with simulated data.
  • When comparing to spatiotemporal models, the space-by-time decomposition offered higher task discrimination power (80% vs. 70% with equal complexity), although spatiotemporal models achieved slightly higher VAF.

Limitations and Future Work

  • Time-Shifts: While incorporating time-shifts improved VAF, it did not significantly enhance decoding performance and could even reduce it. The authors suggest that time-shifts might primarily account for task-irrelevant variability or refine temporal resolution rather than encoding task goals directly.
  • Dataset Complexity: The current study focused on arm pointing movements. The authors plan to apply the space-by-time decomposition to more complex motor behaviors involving a wider range of tasks and a larger number of muscles at different time scales.
  • Neural Data Integration: Future work could involve applying the extraction algorithm to simultaneously recorded neural and muscle activities to directly link neural patterns to spatial and temporal muscle modules, providing deeper insights into the neural basis of modularity.
  • Algorithmic Considerations: The paper notes that updating time-shifts can be computationally challenging and may lead to suboptimal solutions. Further research could explore alternative update rules or convergence properties.

Summary updated

Abstract

Modularity in the central nervous system (CNS), i.e., the brain capability to generate a wide repertoire of movements by combining a small number of building blocks ("modules"), is thought to underlie the control of movement. Numerous studies reported evidence for such a modular organization by identifying invariant muscle activation patterns across various tasks. However, previous studies relied on decompositions differing in both the nature and dimensionality of the identified modules. Here, we derive a single framework that encompasses all influential models of muscle activation modularity. We introduce a new model (named space-by-time decomposition) that factorizes muscle activations into concurrent spatial and temporal modules. To infer these modules, we develop an algorithm, referred to as sample-based nonnegative matrix trifactorization (sNM3F). We test the space-by-time decomposition on a comprehensive electromyographic dataset recorded during execution of arm pointing movements and show that it provides a low-dimensional yet accurate, highly flexible and task-relevant representation of muscle patterns. The extracted modules have a well characterized functional meaning and implement an efficient trade-off between replication of the original muscle patterns and task discriminability. Furthermore, they are compatible with the modules extracted from existing models, such as synchronous synergies and temporal primitives, and generalize time-varying synergies. Our results indicate the effectiveness of a simultaneous but separate condensation of spatial and temporal dimensions of muscle patterns. The space-by-time decomposition accommodates a unified view of the hierarchical mapping from task parameters to coordinated muscle activations, which could be employed as a reference framework for studying compositional motor control.
BibTeX
@article{delis2013unifying,
  author = {Ioannis Delis and Stefano Panzeri and Thierry Pozzo and Bastien Berret},
  title = {{A unifying model of concurrent spatial and temporal modularity in muscle activity}},
  journal = {Journal of Neurophysiology},
  year = {2013},
  doi = {10.1152/jn.00245.2013},
}
Figure