Unlocking the Full Potential of High-Density Surface EMG: Novel Non-Invasive High-Yield Motor Unit Decomposition

Agnese Grison, Irene Mendez Guerra, Alexander Kenneth Clarke, Silvia Muceli, Jaime Ibanez Pereda, Dario Farina

TL;DR

The paper adapts Swarm-Contrastive Decomposition (SCD) to high-density surface electromyography (HD-sEMG), allowing researchers to non-invasively decode significantly more motor units with higher precision than current state-of-the-art methods. By dynamically tuning the algorithm's separation function and sequentially "peeling off" identified signals, SCD successfully extracts low-amplitude and deeply buried motor unit activities even in highly complex, high-interference muscle contractions.

What Problem Does This Paper Address?

To understand how the nervous system controls movement, scientists look at the firing patterns of motor units (a motor neuron and the muscle fibers it activates). Historically, getting a clear picture required invasive techniques like inserting needle electrodes directly into the muscle (intramuscular EMG). While accurate, this only captures a few motor units at a time and is uncomfortable for participants.

High-density surface EMG (HD-sEMG) offers a non-invasive alternative by placing a grid of electrodes on the skin. However, decoding these surface signals into individual motor unit spike trains is incredibly difficult. Current blind source separation (cBSS) algorithms rely on fixed mathematical contrast functions to isolate individual signals. These methods frequently fail in complex scenarios—such as during fast, explosive (ballistic) movements, when recording from deep muscles, or when filtering through thicker layers of subcutaneous fat (a common challenge in female subjects). They tend to repeatedly converge on the loudest, most dominant superficial signals, completely missing the smaller, deeper, or lower-amplitude motor units.

Methods

The authors address these limitations by adapting Swarm-Contrastive Decomposition (SCD)—originally designed for invasive recordings—for use with non-invasive HD-sEMG data.

SCD introduces two core innovations over standard cBSS methods:

  1. Adaptive Contrast Functions: Standard algorithms use a fixed contrast function (typically a polynomial approximation of skewness or kurtosis) to maximize the "sparseness" of the signal. SCD dynamically tunes the exponent ($e$) of its contrast function ($G(s) = E\{\text{sign}(s)|s|^e\}$) for each individual source using particle swarm optimization. This customizes the mathematical filter to the unique shape and distribution of each motor unit's action potential.
  2. Peel-Off Strategy: Instead of using standard deflation methods (which enforce mathematical orthogonality to separate signals), SCD implements a sequential "peel-off" approach. Once a motor unit is successfully identified, its contribution is completely removed from the EMG signal. This lowers the background interference and allows the algorithm to uncover smaller, subtle sources that were previously overshadowed.

Datasets

The algorithm was rigorously evaluated using both simulated and real-world experimental data:

  • Simulated Data: Generated using NeuroMotion, an advanced musculoskeletal and AI-driven volume conductor simulator. The authors simulated an isometric index finger contraction with a pool of 100 motor neurons across varying excitation levels (10% to 100% maximum voluntary contraction [MVC]), different noise intensities (10 to 30 dB), and rapid ballistic contractions.
  • Experimental Data:
  • Two-Source Validation: Conducted on two healthy male subjects performing ankle dorsiflexion. This included concurrent recordings of surface HD-sEMG and invasive intramuscular EMG (HD-iEMG) to cross-verify the accuracy of the surface decomposition against a ground truth.
  • Sex-Difference Yield: Conducted on the tibialis anterior muscle of two healthy female subjects at low force levels (5% and 10% MVC).
  • Complex Muscle Groups: Conducted on the highly complex forearm muscles of two healthy female subjects during index finger flexions at 15% MVC.

Key Insights and Ideas

  • Surface Signals Need Higher Exponents: The authors discovered that the optimal contrast function exponents were significantly higher for surface EMG than for intramuscular EMG. This indicates that non-invasive surface signals, which are naturally smoothed and mixed by the skin and fat layers, require more aggressive mathematical scaling to achieve clean separation.
  • Uncovering the "Hidden" Pool: The motor units uniquely identified by SCD (and missed by traditional cBSS) consistently exhibited lower peak-to-peak amplitudes, slower conduction velocities, and fewer innervated fibers. This proves that the combination of adaptive contrast and peel-off successfully exposes the smaller, deeper motor units that traditional methods are blind to.
  • Breaking the Deflation Bottleneck: Traditional deflation techniques can distort the remaining signal mixture as more sources are extracted. By physically subtracting (peeling off) the identified waveforms, SCD maintains signal integrity, preventing the accuracy degradation typically seen late in the decomposition process.

Key Results

SCD consistently outperformed standard cBSS across all tested conditions:

  • Double the Yield in Standard Simulations: Across all simulated force levels, cBSS identified an average of $13.9 \pm 2.7$ motor units per contraction, whereas SCD nearly doubled that figure, identifying $25.9 \pm 5.8$ units while maintaining a median Rate of Agreement (RoA) of 100%.
  • 3-Fold Increase in Ballistic Tasks: In highly synchronized, rapid contractions, SCD decoded three times as many motor units as cBSS ($31.2 \pm 4.3$ vs. $10.5 \pm 1.7$) with significantly higher precision.
  • Experimental Validation: In the concurrent surface-intramuscular trials, SCD successfully matched up to three times as many motor units to the invasive ground truth compared to cBSS.
  • 50% Improvement in Challenging Demographics: For female subjects and complex forearm muscle structures—scenarios notorious for low motor unit yields—SCD expanded the number of decoded units by roughly 50%, providing a substantial boost for inclusive physiological modeling.
Decomposed Motor Units (Simulation Average)
| Method | Standard Contraction | Ballistic Contraction |
|--------|----------------------|-----------------------|
| cBSS   | 13.9                 | 10.5                  |
| SCD    | 25.9 (~2x increase)  | 31.2 (~3x increase)   |

Limitations and Future Work

The primary limitation of this study is that SCD was validated exclusively under static muscle geometries (isometric contractions). The current framework assumes that the motor unit action potential shapes remain constant over time. During dynamic movements (like walking or reaching), muscles change length, shift beneath the skin, and alter their geometric relationship with the electrode grid. Under these non-stationary conditions, SCD's performance is expected to degrade.

Future Directions: The authors point out that adapting SCD to handle non-stationary signal changes and optimizing the algorithm for real-time, online processing are the logical next steps. This will be crucial for translating this high-yield decomposition technique into practical applications like advanced prosthetic control and real-time clinical neuro-rehabilitation interfaces.

Summary updated

Abstract

The decomposition of high-density surface electromyography (HD-sEMG) signals into motor unit discharge patterns has become a powerful tool for investigating the neural control of movement, providing insights into motor neuron recruitment and discharge behavior. However, current algorithms, while very effective under certain conditions, face significant challenges in complex scenarios, as their accuracy and motor unit yield are highly dependent on anatomical differences among individuals. This can limit the number of decomposed motor units, particularly in challenging conditions. To address this issue, we recently introduced Swarm-Contrastive Decomposition (SCD), which dynamically adjusts the separation function based on the distribution of the data and prevents convergence to the same source. Initially applied to intramuscular EMG signals, SCD is here adapted for HD-sEMG signals. We demonstrated its ability to address key challenges faced by existing methods, particularly in identifying low-amplitude motor unit action potentials and effectively handling complex decomposition scenarios, like high-interference signals. We extensively validated SCD using simulated and experimental HD-sEMG recordings and compared it with current state-of-the-art decomposition methods under varying conditions, including different excitation levels, noise intensities, force profiles, sexes, and muscle groups. The proposed method consistently outperformed existing techniques in both the quantity of decoded motor units and the precision of their firing time identification. For instance, under certain experimental conditions, SCD detected more than three times as many motor units compared to previous methods, while also significantly improving accuracy. These advancements represent a major step forward in non-invasive EMG technology for studying motor unit activity in complex scenarios.
BibTeX
@article{grison2025unlocking,
  author = {Agnese Grison and Irene Mendez Guerra and Alexander Kenneth Clarke and Silvia Muceli and Jaime Ibanez Pereda and Dario Farina},
  title = {{Unlocking the Full Potential of High-Density Surface EMG: Novel Non-Invasive High-Yield Motor Unit Decomposition}},
  journal = {The Journal of Physiology},
  year = {2025},
  doi = {10.1113/jp287913},
}
Figure